Podcast-Oberfläche: wartender Transkriptionsauftrag zeigt den Grund (sw CACHE v171)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>

.gitattributes

@@ -0,0 +1 @@
+*.pdf binary

accounts/keyshare.py

@@ -0,0 +1,134 @@
+"""Passing the book key from an unlocked device to a new one (models.KeyShare).
+
+The book key never reaches the server in the clear: the new device posts an ECDH P-256
+public key, an unlocked device of the same account answers with its own public key and
+the book key AES-GCM-encrypted under the shared secret. Both devices show a code made
+from the request's public key (static/js/app/33-unlock.js), so the user approves the
+right device. Requests run out after a few minutes; an answer is handed out once.
+
+  POST /accounts/keyshare/                  {pub, device}      -> {token}
+  GET  /accounts/keyshare/<token>/                             -> {state, pub, iv, ct}
+  GET  /accounts/keyshare/pending/                             -> {requests: [...]}
+  POST /accounts/keyshare/<token>/answer/   {pub, iv, ct} | {decline: true}
+"""
+import base64
+import binascii
+import json
+from datetime import timedelta
+
+from django.http import JsonResponse
+from django.utils import timezone
+from django.views.decorators.http import require_http_methods
+
+from .models import KeyShare
+
+LIFETIME = timedelta(minutes=10)
+# Only requests whose device asked lately are offered: a closed tab drops out quickly.
+ALIVE = timedelta(seconds=20)
+MAX_OPEN = 5
+
+
+def _auth(request):
+    if not request.user.is_authenticated:
+        return JsonResponse({'error': 'authentication required'}, status=401)
+    return None
+
+
+def _body(request):
+    try:
+        body = json.loads(request.body or b'{}')
+    except (json.JSONDecodeError, ValueError):
+        return None
+    return body if isinstance(body, dict) else None
+
+
+def _b64(value, low, high):
+    """A base64 string whose decoded length lies in [low, high], else None."""
+    if not isinstance(value, str) or len(value) > 4 * high:
+        return None
+    try:
+        raw = base64.b64decode(value, validate=True)
+    except (binascii.Error, ValueError):
+        return None
+    return value if low <= len(raw) <= high else None
+
+
+def _purge(user):
+    KeyShare.objects.filter(user=user, created_at__lt=timezone.now() - LIFETIME).delete()
+
+
+@require_http_methods(['POST'])
+def create(request):
+    if (r := _auth(request)):
+        return r
+    body = _body(request)
+    pub = _b64((body or {}).get('pub'), 65, 65)  # uncompressed P-256 point
+    if pub is None:
+        return JsonResponse({'error': 'pub required'}, status=400)
+    _purge(request.user)
+    # A device that reloads asks anew: the oldest requests make room instead of blocking it.
+    stale = KeyShare.objects.filter(user=request.user).order_by('-created_at').values_list('pk', flat=True)[MAX_OPEN - 1:]
+    KeyShare.objects.filter(pk__in=list(stale)).delete()
+    device = str(body.get('device') or '')[:100]
+    share = KeyShare.objects.create(user=request.user, request_pub=pub, device=device,
+                                    polled_at=timezone.now())
+    return JsonResponse({'token': share.token})
+
+
+@require_http_methods(['GET'])
+def status(request, token):
+    if (r := _auth(request)):
+        return r
+    _purge(request.user)
+    share = KeyShare.objects.filter(user=request.user, token=token).first()
+    if share is None:
+        return JsonResponse({'state': 'expired'})
+    if share.declined:
+        share.delete()
+        return JsonResponse({'state': 'declined'})
+    if not share.answer_ct:
+        share.polled_at = timezone.now()
+        share.save(update_fields=['polled_at'])
+        return JsonResponse({'state': 'pending'})
+    data = {'state': 'answered', 'pub': share.answer_pub, 'iv': share.answer_iv, 'ct': share.answer_ct}
+    share.delete()  # handed out once
+    return JsonResponse(data)
+
+
+@require_http_methods(['GET'])
+def pending(request):
+    if (r := _auth(request)):
+        return r
+    _purge(request.user)
+    open_ = KeyShare.objects.filter(user=request.user, declined=False, answer_ct='',
+                                    polled_at__gte=timezone.now() - ALIVE).order_by('-created_at')
+    return JsonResponse({'requests': [
+        {'token': s.token, 'pub': s.request_pub, 'device': s.device,
+         'created_at': s.created_at.isoformat()} for s in open_]})
+
+
+@require_http_methods(['POST'])
+def answer(request, token):
+    if (r := _auth(request)):
+        return r
+    _purge(request.user)
+    share = KeyShare.objects.filter(user=request.user, token=token).first()
+    if share is None:
+        return JsonResponse({'error': 'expired'}, status=404)
+    if share.answer_ct or share.declined:
+        return JsonResponse({'error': 'already answered'}, status=409)
+    body = _body(request) or {}
+    if body.get('decline'):
+        share.declined = True
+        share.answered_at = timezone.now()
+        share.save(update_fields=['declined', 'answered_at'])
+        return JsonResponse({'ok': True})
+    pub = _b64(body.get('pub'), 65, 65)
+    iv = _b64(body.get('iv'), 12, 12)
+    ct = _b64(body.get('ct'), 32 + 16, 32 + 16)  # 256-bit key + GCM tag
+    if not (pub and iv and ct):
+        return JsonResponse({'error': 'pub, iv and ct required'}, status=400)
+    share.answer_pub, share.answer_iv, share.answer_ct = pub, iv, ct
+    share.answered_at = timezone.now()
+    share.save(update_fields=['answer_pub', 'answer_iv', 'answer_ct', 'answered_at'])
+    return JsonResponse({'ok': True})

accounts/migrations/0008_oidc_keyshare.py

@@ -0,0 +1,50 @@
+# Generated by Django 6.1 on 2026-11-01 13:35
+
+import accounts.models
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('accounts', '0007_uservault'),
+        migrations.swappable_dependency(settings.AUTH_USER_MODEL),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name='KeyShare',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('token', models.CharField(default=accounts.models._generate_token, max_length=64, unique=True)),
+                ('device', models.CharField(blank=True, max_length=100)),
+                ('request_pub', models.TextField()),
+                ('answer_pub', models.TextField(blank=True)),
+                ('answer_iv', models.CharField(blank=True, max_length=64)),
+                ('answer_ct', models.TextField(blank=True)),
+                ('declined', models.BooleanField(default=False)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('polled_at', models.DateTimeField(blank=True, null=True)),
+                ('answered_at', models.DateTimeField(blank=True, null=True)),
+                ('user', models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='key_shares', to=settings.AUTH_USER_MODEL)),
+            ],
+        ),
+        migrations.CreateModel(
+            name='OidcIdentity',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('issuer', models.CharField(max_length=200)),
+                ('sub', models.CharField(max_length=255)),
+                ('username', models.CharField(blank=True, max_length=150)),
+                ('email', models.CharField(blank=True, max_length=254)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('last_login_at', models.DateTimeField(blank=True, null=True)),
+                ('user', models.ForeignKey(on_delete=django.db.models.deletion.CASCADE, related_name='oidc_identities', to=settings.AUTH_USER_MODEL)),
+            ],
+            options={
+                'unique_together': {('issuer', 'sub')},
+            },
+        ),
+    ]

accounts/models.py

@@ -142,3 +142,40 @@ class ApiToken(models.Model):
 
     def __str__(self):
         return f"ApiToken({self.user.username})"
+
+
+class OidcIdentity(models.Model):
+    """An account of the sign-in provider (Authelia) linked to a diora account
+    (accounts/oidc.py). `sub` is the provider's stable id; name and mail are only shown."""
+    user = models.ForeignKey(User, on_delete=models.CASCADE, related_name='oidc_identities')
+    issuer = models.CharField(max_length=200)
+    sub = models.CharField(max_length=255)
+    username = models.CharField(max_length=150, blank=True)
+    email = models.CharField(max_length=254, blank=True)
+    created_at = models.DateTimeField(auto_now_add=True)
+    last_login_at = models.DateTimeField(null=True, blank=True)
+
+    class Meta:
+        unique_together = ('issuer', 'sub')
+
+    def __str__(self):
+        return f"OidcIdentity({self.username or self.sub} -> {self.user_id})"
+
+
+class KeyShare(models.Model):
+    """A device without the book key asks an unlocked device of the same account for it
+    (accounts/keyshare.py). The server only passes on public keys and ciphertext: the new
+    device sends an ECDH public key, the unlocked one answers with its own public key and
+    the book key encrypted for the shared secret. Both show a code made from the request's
+    public key, so the user can see the answer goes to the right device."""
+    user = models.ForeignKey(User, on_delete=models.CASCADE, related_name='key_shares')
+    token = models.CharField(max_length=64, unique=True, default=_generate_token)
+    device = models.CharField(max_length=100, blank=True)
+    request_pub = models.TextField()   # base64 raw P-256 point of the asking device
+    answer_pub = models.TextField(blank=True)
+    answer_iv = models.CharField(max_length=64, blank=True)
+    answer_ct = models.TextField(blank=True)
+    declined = models.BooleanField(default=False)
+    created_at = models.DateTimeField(auto_now_add=True)
+    polled_at = models.DateTimeField(null=True, blank=True)  # the asking device still waits
+    answered_at = models.DateTimeField(null=True, blank=True)

accounts/oidc.py

@@ -0,0 +1,165 @@
+"""Sign-in with an OpenID Connect provider (Authelia, docs/accounts.md).
+
+An extra way in next to username + password: an account is linked once in the settings
+while signed in normally, after that the provider's login leads into it. Nobody gets a new
+account from the provider — an unlinked identity is turned away.
+
+Authorization code flow with PKCE (S256) and a confidential client (client_secret_basic).
+The id_token is checked here (RS256 against the provider's JWKS, iss, aud, exp, nonce)
+with `cryptography`, so no OIDC library is needed.
+
+The book key is not touched: it is derived from the diora password on the device. A device
+that signs in this way gets it from an unlocked device or by typing the password once
+(accounts/keyshare.py, static/js/app/33-unlock.js).
+"""
+import base64
+import hashlib
+import json
+import secrets
+import time
+
+from urllib.parse import urlencode
+
+import requests
+from cryptography.exceptions import InvalidSignature
+from cryptography.hazmat.primitives import hashes
+from cryptography.hazmat.primitives.asymmetric import padding, rsa
+from django.conf import settings
+
+TIMEOUT = 10
+_CACHE_SECONDS = 3600
+_cache = {}  # url -> (fetched_at, json)
+
+
+class OidcError(Exception):
+    pass
+
+
+def enabled() -> bool:
+    return bool(settings.OIDC_ISSUER and settings.OIDC_CLIENT_ID and settings.OIDC_CLIENT_SECRET)
+
+
+def key_enabled() -> bool:
+    return enabled() and bool(settings.OIDC_APP_CLIENT_ID)
+
+
+def _b64url_decode(s: str) -> bytes:
+    return base64.urlsafe_b64decode(s + '=' * (-len(s) % 4))
+
+
+def _b64url(b: bytes) -> str:
+    return base64.urlsafe_b64encode(b).rstrip(b'=').decode('ascii')
+
+
+def _get_json(url: str, fresh: bool = False) -> dict:
+    hit = _cache.get(url)
+    if hit and not fresh and time.time() - hit[0] < _CACHE_SECONDS:
+        return hit[1]
+    try:
+        resp = requests.get(url, timeout=TIMEOUT)
+        resp.raise_for_status()
+        data = resp.json()
+    except (requests.RequestException, ValueError) as exc:
+        raise OidcError(f'{url} nicht erreichbar: {exc}') from exc
+    _cache[url] = (time.time(), data)
+    return data
+
+
+def discovery() -> dict:
+    issuer = settings.OIDC_ISSUER.rstrip('/')
+    return _get_json(issuer + '/.well-known/openid-configuration')
+
+
+def new_flow() -> dict:
+    """Random state, nonce and PKCE verifier for one sign-in (kept in the session)."""
+    return {
+        'state': secrets.token_urlsafe(24),
+        'nonce': secrets.token_urlsafe(24),
+        'verifier': secrets.token_urlsafe(48),
+    }
+
+
+def authorize_url(flow: dict, redirect_uri: str) -> str:
+    challenge = _b64url(hashlib.sha256(flow['verifier'].encode('ascii')).digest())
+    params = {
+        'response_type': 'code',
+        'client_id': settings.OIDC_CLIENT_ID,
+        'redirect_uri': redirect_uri,
+        'scope': 'openid profile email groups',
+        'state': flow['state'],
+        'nonce': flow['nonce'],
+        'code_challenge': challenge,
+        'code_challenge_method': 'S256',
+    }
+    return discovery()['authorization_endpoint'] + '?' + urlencode(params)
+
+
+def exchange_code(code: str, flow: dict, redirect_uri: str) -> dict:
+    """Trade the code for tokens and return the checked id_token claims."""
+    try:
+        resp = requests.post(
+            discovery()['token_endpoint'],
+            data={
+                'grant_type': 'authorization_code',
+                'code': code,
+                'redirect_uri': redirect_uri,
+                'code_verifier': flow['verifier'],
+            },
+            auth=(settings.OIDC_CLIENT_ID, settings.OIDC_CLIENT_SECRET),
+            timeout=TIMEOUT,
+        )
+    except requests.RequestException as exc:
+        raise OidcError(f'Token-Abruf fehlgeschlagen: {exc}') from exc
+    if resp.status_code != 200:
+        raise OidcError(f'Token-Abruf abgelehnt ({resp.status_code})')
+    try:
+        id_token = resp.json()['id_token']
+    except (ValueError, KeyError) as exc:
+        raise OidcError('Antwort ohne id_token') from exc
+    return verify_id_token(id_token, flow['nonce'])
+
+
+def _public_key(kid: str):
+    jwks_uri = discovery()['jwks_uri']
+    for fresh in (False, True):  # a key rotation shows up as an unknown kid: fetch again once
+        for jwk in _get_json(jwks_uri, fresh=fresh).get('keys', []):
+            if jwk.get('kty') == 'RSA' and (not kid or jwk.get('kid') == kid):
+                n = int.from_bytes(_b64url_decode(jwk['n']), 'big')
+                e = int.from_bytes(_b64url_decode(jwk['e']), 'big')
+                return rsa.RSAPublicNumbers(e, n).public_key()
+    raise OidcError('Signaturschlüssel des Anbieters nicht gefunden')
+
+
+def verify_id_token(token: str, nonce, now: float = None, client_id: str = None) -> dict:
+    """Check signature and claims. `nonce` is the expected value, or a function that says
+    whether the token's nonce is good (the apps' signed ones, oidc_views.app_login)."""
+    try:
+        header_b64, payload_b64, sig_b64 = token.split('.')
+        header = json.loads(_b64url_decode(header_b64))
+        claims = json.loads(_b64url_decode(payload_b64))
+        signature = _b64url_decode(sig_b64)
+    except (ValueError, TypeError) as exc:
+        raise OidcError('id_token unlesbar') from exc
+    if header.get('alg') != 'RS256':
+        raise OidcError(f'nicht unterstützte Signatur {header.get("alg")!r} (erwartet RS256)')
+    try:
+        _public_key(header.get('kid', '')).verify(
+            signature, f'{header_b64}.{payload_b64}'.encode('ascii'),
+            padding.PKCS1v15(), hashes.SHA256())
+    except InvalidSignature as exc:
+        raise OidcError('Signatur des id_token ungültig') from exc
+
+    now = time.time() if now is None else now
+    if claims.get('iss', '').rstrip('/') != settings.OIDC_ISSUER.rstrip('/'):
+        raise OidcError('falscher Aussteller')
+    aud = claims.get('aud')
+    if (client_id or settings.OIDC_CLIENT_ID) not in (aud if isinstance(aud, list) else [aud]):
+        raise OidcError('id_token ist nicht für diora')
+    if not isinstance(claims.get('exp'), (int, float)) or claims['exp'] < now - 60:
+        raise OidcError('id_token abgelaufen')
+    good = nonce(claims.get('nonce') or '') if callable(nonce) else (nonce and claims.get('nonce') == nonce)
+    if not good:
+        raise OidcError('nonce passt nicht')
+    if not claims.get('sub'):
+        raise OidcError('id_token ohne sub')
+    return claims

accounts/oidc_views.py

@@ -0,0 +1,245 @@
+"""Sign-in with Authelia and linking it to an existing account (accounts/oidc.py)."""
+import json
+import logging
+import secrets
+
+from django.conf import settings
+from django.contrib import messages
+from django.contrib.auth import login
+from django.contrib.auth.decorators import login_required
+from django.core import signing
+from django.http import JsonResponse
+from django.shortcuts import redirect, render
+from django.utils import timezone
+from django.utils.http import url_has_allowed_host_and_scheme
+from django.views.decorators.csrf import csrf_exempt
+from django.views.decorators.http import require_http_methods
+
+from . import oidc
+from .models import ApiToken, OidcIdentity
+
+log = logging.getLogger(__name__)
+
+SESSION_KEY = 'oidc_flow'
+
+
+def _redirect_uri(request):
+    return settings.OIDC_REDIRECT_URI or request.build_absolute_uri('/accounts/oidc/callback/')
+
+
+def _safe_next(request, value):
+    if value and url_has_allowed_host_and_scheme(value, allowed_hosts={request.get_host()},
+                                                 require_https=request.is_secure()):
+        return value
+    return '/'
+
+
+def oidc_context(request):
+    """What the settings page shows about the provider."""
+    return {
+        'oidc_enabled': oidc.enabled(),
+        'oidc_provider': settings.OIDC_PROVIDER_NAME,
+        'oidc_identities': list(request.user.oidc_identities.all()) if request.user.is_authenticated else [],
+    }
+
+
+@require_http_methods(['GET'])
+def oidc_start(request):
+    """Off to the provider. `?link=1` (signed in) links the account instead of signing in."""
+    if not oidc.enabled():
+        messages.error(request, 'Die Anmeldung über Authelia ist auf diesem Server nicht eingerichtet.')
+        return redirect('login')
+    link = request.GET.get('link') == '1'
+    if link and not request.user.is_authenticated:
+        return redirect('login')
+    flow = oidc.new_flow()
+    flow['link'] = link
+    flow['next'] = _safe_next(request, request.GET.get('next', ''))
+    try:
+        url = oidc.authorize_url(flow, _redirect_uri(request))
+    except oidc.OidcError as exc:
+        log.warning('oidc start: %s', exc)
+        messages.error(request, 'Authelia ist gerade nicht erreichbar.')
+        return redirect('settings' if link else 'login')
+    request.session[SESSION_KEY] = flow
+    return redirect(url)
+
+
+@require_http_methods(['GET'])
+def oidc_callback(request):
+    flow = request.session.pop(SESSION_KEY, None)
+    fail_to = 'login'
+    if flow and flow.get('link'):
+        fail_to = 'settings'
+    if not flow or request.GET.get('state') != flow.get('state'):
+        messages.error(request, 'Die Anmeldung über Authelia ist abgelaufen. Bitte noch einmal.')
+        return redirect(fail_to)
+    if request.GET.get('error'):
+        messages.error(request, 'Authelia hat die Anmeldung abgebrochen.')
+        return redirect(fail_to)
+    code = request.GET.get('code', '')
+    try:
+        claims = oidc.exchange_code(code, flow, _redirect_uri(request))
+    except oidc.OidcError as exc:
+        log.warning('oidc callback: %s', exc)
+        messages.error(request, 'Die Anmeldung über Authelia hat nicht geklappt.')
+        return redirect(fail_to)
+
+    issuer = settings.OIDC_ISSUER.rstrip('/')
+    sub = claims['sub']
+    shown = {
+        'username': (claims.get('preferred_username') or '')[:150],
+        'email': (claims.get('email') or '')[:254],
+    }
+    identity = OidcIdentity.objects.filter(issuer=issuer, sub=sub).select_related('user').first()
+
+    if flow.get('link'):
+        if not request.user.is_authenticated:
+            return redirect('login')
+        if identity and identity.user_id != request.user.id:
+            messages.error(request, 'Dieses Authelia-Konto ist schon mit einem anderen diora-Konto verknüpft.')
+            return redirect('settings')
+        if identity is None:
+            OidcIdentity.objects.create(user=request.user, issuer=issuer, sub=sub, **shown)
+        else:
+            OidcIdentity.objects.filter(pk=identity.pk).update(**shown)
+        messages.success(request, f'Authelia-Konto „{shown["username"] or sub}“ verknüpft. '
+                                  'Du kannst dich jetzt auch darüber anmelden.')
+        if oidc.key_enabled():  # leave the book key with Authelia's help right away
+            return redirect('/accounts/oidc/key/?next=/accounts/settings/')
+        return redirect('settings')
+
+    if identity is None:
+        messages.error(request, 'Dieses Authelia-Konto ist mit keinem diora-Konto verknüpft. '
+                                'Melde dich einmal mit Passwort an und verknüpfe es in den Einstellungen.')
+        return redirect('login')
+    if not identity.user.is_active:
+        messages.error(request, 'Dieses diora-Konto ist gesperrt.')
+        return redirect('login')
+    OidcIdentity.objects.filter(pk=identity.pk).update(last_login_at=timezone.now(), **shown)
+    login(request, identity.user, backend='django.contrib.auth.backends.ModelBackend')
+    # A password login brings the book key along, this one cannot: the device fetches it
+    # (oidc_key below), else the page asks (static/js/app/33-unlock.js).
+    next_url = flow.get('next') or '/'
+    if oidc.key_enabled():
+        from urllib.parse import urlencode
+        return redirect('/accounts/oidc/key/?' + urlencode({'next': next_url}))
+    return redirect(next_url)
+
+
+@login_required
+@require_http_methods(['POST'])
+def oidc_unlink(request, pk):
+    OidcIdentity.objects.filter(user=request.user, pk=pk).delete()
+    messages.success(request, 'Verknüpfung mit Authelia gelöst.')
+    return redirect('settings')
+
+
+@login_required
+@require_http_methods(['GET'])
+def oidc_key(request):
+    """The book key with the provider's help, without asking (docs/accounts.md).
+
+    Runs entirely on the device: it signs in once more as the public client
+    (OIDC_APP_CLIENT_ID, PKCE, prompt=none — the provider session from just before makes it
+    invisible), reads its `diora_key` claim from userinfo and with it opens the book key the
+    server keeps wrapped (vault "authelia-key") — or, when the device has the key, wraps and
+    leaves it there. The server never sees the claim; this view only hands out the page.
+    The same page is the redirect target (`/accounts/oidc/key/callback/`)."""
+    next_url = _safe_next(request, request.GET.get('next', ''))
+    if not oidc.key_enabled():
+        return redirect(next_url)
+    try:
+        disco = oidc.discovery()
+    except oidc.OidcError as exc:
+        log.warning('oidc key: %s', exc)
+        return redirect(next_url)
+    config = {
+        'authorize': disco['authorization_endpoint'],
+        'token': disco['token_endpoint'],
+        'userinfo': disco.get('userinfo_endpoint', ''),
+        'client_id': settings.OIDC_APP_CLIENT_ID,
+        'redirect_uri': request.build_absolute_uri('/accounts/oidc/key/callback/')
+                        if not settings.OIDC_REDIRECT_URI
+                        else settings.OIDC_REDIRECT_URI.replace('/oidc/callback/', '/oidc/key/callback/'),
+        'subs': [i.sub for i in request.user.oidc_identities.all()],
+        'user_id': request.user.id,
+        'next': next_url,
+    }
+    return render(request, 'accounts/oidc_key.html', {'config': config})
+
+
+# ---------------------------------------------------------------------------
+# The apps (Android, desktop): sign in with Authelia as the public client themselves
+# ---------------------------------------------------------------------------
+#
+# 1. GET  /accounts/oidc/app/        -> where Authelia is, the client id and a nonce
+#    signed by this server (good for APP_NONCE_AGE).
+# 2. The app runs the code flow with PKCE in the system browser (redirect to its own
+#    scheme or loopback), trades the code itself and reads `diora_key` from userinfo.
+# 3. POST /accounts/oidc/app/login/  {id_token} -> the account's API token, as the
+#    password login (diora-android Auth.kt) hands it out.
+# The book key never passes here: the app opens the vault "authelia-key" itself.
+
+APP_NONCE_SALT = 'diora.oidc.app'
+APP_NONCE_AGE = 600
+
+
+@require_http_methods(['GET'])
+def app_config(request):
+    if not oidc.key_enabled():
+        return JsonResponse({'enabled': False})
+    try:
+        disco = oidc.discovery()
+    except oidc.OidcError as exc:
+        log.warning('oidc app config: %s', exc)
+        return JsonResponse({'enabled': False, 'error': 'provider unreachable'}, status=502)
+    return JsonResponse({
+        'enabled': True,
+        'provider': settings.OIDC_PROVIDER_NAME,
+        'issuer': settings.OIDC_ISSUER.rstrip('/'),
+        'authorize': disco['authorization_endpoint'],
+        'token': disco['token_endpoint'],
+        'userinfo': disco.get('userinfo_endpoint', ''),
+        'client_id': settings.OIDC_APP_CLIENT_ID,
+        'scope': 'openid profile diora_key',
+        'nonce': signing.dumps(secrets.token_urlsafe(16), salt=APP_NONCE_SALT),
+    })
+
+
+def _app_nonce_ok(value):
+    try:
+        signing.loads(value, salt=APP_NONCE_SALT, max_age=APP_NONCE_AGE)
+        return True
+    except signing.BadSignature:
+        return False
+
+
+@csrf_exempt  # a native client without cookies; the signed id_token is the proof
+@require_http_methods(['POST'])
+def app_login(request):
+    if not oidc.key_enabled():
+        return JsonResponse({'error': 'not enabled'}, status=404)
+    try:
+        body = json.loads(request.body or b'{}')
+        id_token = body['id_token']
+        if not isinstance(id_token, str):
+            raise TypeError
+    except (ValueError, KeyError, TypeError):
+        return JsonResponse({'error': 'id_token required'}, status=400)
+    try:
+        claims = oidc.verify_id_token(id_token, _app_nonce_ok, client_id=settings.OIDC_APP_CLIENT_ID)
+    except oidc.OidcError as exc:
+        log.warning('oidc app login: %s', exc)
+        return JsonResponse({'error': 'Die Anmeldung über Authelia war ungültig.'}, status=401)
+    identity = OidcIdentity.objects.filter(
+        issuer=settings.OIDC_ISSUER.rstrip('/'), sub=claims['sub']).select_related('user').first()
+    if identity is None:
+        return JsonResponse({'error': 'Dieses Authelia-Konto ist mit keinem diora-Konto verknüpft. '
+                                      'Verknüpfe es einmal im Browser unter Einstellungen.'}, status=403)
+    if not identity.user.is_active:
+        return JsonResponse({'error': 'Dieses diora-Konto ist gesperrt.'}, status=403)
+    OidcIdentity.objects.filter(pk=identity.pk).update(last_login_at=timezone.now())
+    token, _ = ApiToken.objects.get_or_create(user=identity.user)
+    return JsonResponse({'token': token.token, 'user_id': identity.user.id,
+                         'username': identity.user.username, 'sub': claims['sub']})

accounts/test_oidc.py

@@ -0,0 +1,321 @@
+import base64
+import json
+import time
+from unittest.mock import patch
+from urllib.parse import parse_qs, urlsplit
+
+from cryptography.hazmat.primitives import hashes
+from cryptography.hazmat.primitives.asymmetric import padding, rsa
+from django.contrib.auth.models import User
+from django.test import TestCase, override_settings
+
+from . import oidc
+from .models import ApiToken, KeyShare, OidcIdentity
+
+ISSUER = 'https://auth.example.org'
+PRIV = rsa.generate_private_key(public_exponent=65537, key_size=2048)
+OTHER_PRIV = rsa.generate_private_key(public_exponent=65537, key_size=2048)
+
+
+def _b64url(b):
+    return base64.urlsafe_b64encode(b).rstrip(b'=').decode()
+
+
+def _int_b64(i):
+    return _b64url(i.to_bytes((i.bit_length() + 7) // 8, 'big'))
+
+
+def make_token(claims, key=PRIV, alg='RS256', kid='k1'):
+    header = _b64url(json.dumps({'alg': alg, 'kid': kid, 'typ': 'JWT'}).encode())
+    payload = _b64url(json.dumps(claims).encode())
+    sig = key.sign(f'{header}.{payload}'.encode(), padding.PKCS1v15(), hashes.SHA256())
+    return f'{header}.{payload}.{_b64url(sig)}'
+
+
+DISCOVERY = {
+    'issuer': ISSUER,
+    'authorization_endpoint': ISSUER + '/api/oidc/authorization',
+    'token_endpoint': ISSUER + '/api/oidc/token',
+    'jwks_uri': ISSUER + '/jwks.json',
+}
+pub = PRIV.public_key().public_numbers()
+JWKS = {'keys': [{'kty': 'RSA', 'kid': 'k1', 'n': _int_b64(pub.n), 'e': _int_b64(pub.e)}]}
+
+
+class FakeResp:
+    def __init__(self, data, status=200):
+        self._data, self.status_code = data, status
+
+    def json(self):
+        return self._data
+
+    def raise_for_status(self):
+        if self.status_code >= 400:
+            raise oidc.requests.HTTPError(str(self.status_code))
+
+
+def fake_get(url, timeout=None):
+    return FakeResp(DISCOVERY if url.endswith('openid-configuration') else JWKS)
+
+
+@override_settings(OIDC_ISSUER=ISSUER, OIDC_CLIENT_ID='diora', OIDC_CLIENT_SECRET='s3cret',
+                   OIDC_REDIRECT_URI='https://diora.test/accounts/oidc/callback/')
+class OidcFlowTests(TestCase):
+    def setUp(self):
+        oidc._cache.clear()
+        self.user = User.objects.create_user(username='crm', password='pw12345678')
+        self.get = patch('accounts.oidc.requests.get', side_effect=fake_get)
+        self.get.start()
+        self.addCleanup(self.get.stop)
+
+    def _start(self, query=''):
+        resp = self.client.get('/accounts/oidc/start/' + query)
+        self.assertEqual(resp.status_code, 302)
+        q = parse_qs(urlsplit(resp['Location']).query)
+        self.assertEqual(q['code_challenge_method'], ['S256'])
+        self.assertEqual(q['client_id'], ['diora'])
+        return q['state'][0], q['nonce'][0]
+
+    def _callback(self, state, claims, token_status=200):
+        token = make_token(claims)
+        with patch('accounts.oidc.requests.post',
+                   return_value=FakeResp({'id_token': token}, token_status)) as post:
+            resp = self.client.get('/accounts/oidc/callback/', {'state': state, 'code': 'c0de'})
+        return resp, post
+
+    def _claims(self, nonce, **extra):
+        c = {'iss': ISSUER, 'aud': ['diora'], 'sub': 'uuid-1', 'exp': time.time() + 300,
+             'nonce': nonce, 'preferred_username': 'marwin', 'email': 'm@example.org'}
+        c.update(extra)
+        return c
+
+    def test_unlinked_identity_is_turned_away(self):
+        state, nonce = self._start()
+        resp, _ = self._callback(state, self._claims(nonce))
+        self.assertEqual(resp['Location'], '/accounts/login/')
+        self.assertNotIn('_auth_user_id', self.client.session)
+        self.assertFalse(User.objects.exclude(pk=self.user.pk).exists())
+
+    def test_link_then_login(self):
+        self.client.force_login(self.user)
+        state, nonce = self._start('?link=1')
+        resp, post = self._callback(state, self._claims(nonce))
+        self.assertEqual(resp['Location'], '/accounts/settings/')
+        ident = OidcIdentity.objects.get()
+        self.assertEqual((ident.user, ident.sub, ident.username), (self.user, 'uuid-1', 'marwin'))
+        # client_secret_basic + PKCE verifier
+        self.assertEqual(post.call_args.kwargs['auth'], ('diora', 's3cret'))
+        self.assertIn('code_verifier', post.call_args.kwargs['data'])
+
+        self.client.logout()
+        state, nonce = self._start('?next=/books/')
+        resp, _ = self._callback(state, self._claims(nonce))
+        self.assertEqual(resp['Location'], '/books/')
+        self.assertEqual(int(self.client.session['_auth_user_id']), self.user.pk)
+        ident.refresh_from_db()
+        self.assertIsNotNone(ident.last_login_at)
+
+    def test_next_to_other_host_is_dropped(self):
+        OidcIdentity.objects.create(user=self.user, issuer=ISSUER, sub='uuid-1')
+        state, nonce = self._start('?next=https://evil.example/')
+        resp, _ = self._callback(state, self._claims(nonce))
+        self.assertEqual(resp['Location'], '/')
+
+    def test_identity_of_other_account_is_not_moved(self):
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        OidcIdentity.objects.create(user=other, issuer=ISSUER, sub='uuid-1')
+        self.client.force_login(self.user)
+        state, nonce = self._start('?link=1')
+        self._callback(state, self._claims(nonce))
+        self.assertEqual(OidcIdentity.objects.get().user, other)
+
+    def test_wrong_state_nonce_audience_or_signature_fail(self):
+        OidcIdentity.objects.create(user=self.user, issuer=ISSUER, sub='uuid-1')
+        cases = [
+            ('state', {}),
+            ('nonce', {'nonce': 'other'}),
+            ('aud', {'aud': 'someone-else'}),
+            ('iss', {'iss': 'https://evil.example'}),
+            ('exp', {'exp': time.time() - 3600}),
+        ]
+        for name, extra in cases:
+            with self.subTest(name):
+                state, nonce = self._start()
+                claims = {**self._claims(nonce), **extra}
+                resp, _ = self._callback('wrong' if name == 'state' else state, claims)
+                self.assertEqual(resp['Location'], '/accounts/login/')
+                self.assertNotIn('_auth_user_id', self.client.session)
+        state, nonce = self._start()
+        with patch('accounts.oidc.requests.post',
+                   return_value=FakeResp({'id_token': make_token(self._claims(nonce), key=OTHER_PRIV)})):
+            self.client.get('/accounts/oidc/callback/', {'state': state, 'code': 'x'})
+        self.assertNotIn('_auth_user_id', self.client.session)
+
+    def test_state_is_single_use(self):
+        OidcIdentity.objects.create(user=self.user, issuer=ISSUER, sub='uuid-1')
+        state, nonce = self._start()
+        self._callback(state, self._claims(nonce))
+        self.client.logout()
+        resp, _ = self._callback(state, self._claims(nonce))
+        self.assertNotIn('_auth_user_id', self.client.session)
+
+    def test_unlink(self):
+        ident = OidcIdentity.objects.create(user=self.user, issuer=ISSUER, sub='uuid-1')
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        foreign = OidcIdentity.objects.create(user=other, issuer=ISSUER, sub='uuid-2')
+        self.client.force_login(self.user)
+        self.client.post(f'/accounts/oidc/{foreign.pk}/unlink/')
+        self.client.post(f'/accounts/oidc/{ident.pk}/unlink/')
+        self.assertEqual(list(OidcIdentity.objects.all()), [foreign])
+
+    def test_with_app_client_login_and_link_go_by_the_key_page(self):
+        OidcIdentity.objects.create(user=self.user, issuer=ISSUER, sub='uuid-1')
+        with override_settings(OIDC_APP_CLIENT_ID='diora-app'):
+            state, nonce = self._start('?next=/books/')
+            resp, _ = self._callback(state, self._claims(nonce))
+            self.assertEqual(resp['Location'], '/accounts/oidc/key/?next=%2Fbooks%2F')
+            page = self.client.get(resp['Location'])
+            self.assertContains(page, 'oidc-key-config')
+            cfg = page.context['config']
+            self.assertEqual(cfg['client_id'], 'diora-app')
+            self.assertEqual(cfg['redirect_uri'], 'https://diora.test/accounts/oidc/key/callback/')
+            self.assertEqual(cfg['subs'], ['uuid-1'])
+            self.assertEqual(cfg['next'], '/books/')
+            self.assertEqual(cfg['token'], DISCOVERY['token_endpoint'])
+            self.assertEqual(self.client.get('/accounts/oidc/key/callback/?code=x&state=y').status_code, 200)
+
+            state, nonce = self._start('?link=1')
+            resp, _ = self._callback(state, self._claims(nonce))
+            self.assertEqual(resp['Location'], '/accounts/oidc/key/?next=/accounts/settings/')
+
+    def test_key_page_without_app_client_goes_straight_on(self):
+        self.client.force_login(self.user)
+        resp = self.client.get('/accounts/oidc/key/?next=/books/')
+        self.assertEqual(resp['Location'], '/books/')
+
+    def test_login_page_shows_button_only_when_configured(self):
+        self.assertContains(self.client.get('/accounts/login/'), 'Mit Authelia anmelden')
+        with override_settings(OIDC_CLIENT_SECRET=''):
+            self.assertNotContains(self.client.get('/accounts/login/'), 'Mit Authelia anmelden')
+
+
+@override_settings(OIDC_ISSUER=ISSUER, OIDC_CLIENT_ID='diora', OIDC_CLIENT_SECRET='s3cret',
+                   OIDC_APP_CLIENT_ID='diora-app')
+class OidcAppLoginTests(TestCase):
+    def setUp(self):
+        oidc._cache.clear()
+        self.user = User.objects.create_user(username='crm', password='pw12345678')
+        OidcIdentity.objects.create(user=self.user, issuer=ISSUER, sub='uuid-1')
+        p = patch('accounts.oidc.requests.get', side_effect=fake_get)
+        p.start()
+        self.addCleanup(p.stop)
+
+    def _nonce(self):
+        cfg = self.client.get('/accounts/oidc/app/').json()
+        self.assertEqual((cfg['client_id'], cfg['token']), ('diora-app', DISCOVERY['token_endpoint']))
+        return cfg['nonce']
+
+    def _login(self, **claims):
+        c = {'iss': ISSUER, 'aud': ['diora-app'], 'sub': 'uuid-1', 'exp': time.time() + 300,
+             'nonce': self._nonce()}
+        c.update(claims)
+        return self.client.post('/accounts/oidc/app/login/', json.dumps({'id_token': make_token(c)}),
+                                content_type='application/json')
+
+    def test_linked_account_gets_its_existing_token(self):
+        existing = ApiToken.objects.create(user=self.user)
+        resp = self._login()
+        self.assertEqual(resp.status_code, 200)
+        self.assertEqual(resp.json(), {'token': existing.token, 'user_id': self.user.id,
+                                       'username': 'crm', 'sub': 'uuid-1'})
+
+    def test_token_is_made_when_there_is_none(self):
+        resp = self._login()
+        self.assertEqual(resp.json()['token'], ApiToken.objects.get(user=self.user).token)
+
+    def test_refused(self):
+        for name, claims, code in [
+            ('unlinked', {'sub': 'other'}, 403),
+            ('web client token', {'aud': 'diora'}, 401),
+            ('own nonce', {'nonce': 'made-up'}, 401),
+            ('expired', {'exp': time.time() - 3600}, 401),
+        ]:
+            with self.subTest(name):
+                self.assertEqual(self._login(**claims).status_code, code)
+        self.assertEqual(self.client.post('/accounts/oidc/app/login/', '{}',
+                                          content_type='application/json').status_code, 400)
+
+    def test_old_nonce_is_refused(self):
+        nonce = self._nonce()
+        with patch('django.core.signing.time.time', return_value=time.time() + 3600):
+            resp = self.client.post('/accounts/oidc/app/login/', json.dumps({'id_token': make_token(
+                {'iss': ISSUER, 'aud': 'diora-app', 'sub': 'uuid-1', 'exp': time.time() + 7200, 'nonce': nonce})}),
+                content_type='application/json')
+        self.assertEqual(resp.status_code, 401)
+
+    def test_off_without_app_client(self):
+        with override_settings(OIDC_APP_CLIENT_ID=''):
+            self.assertEqual(self.client.get('/accounts/oidc/app/').json(), {'enabled': False})
+
+
+PUB = base64.b64encode(b'\x04' + b'\x01' * 64).decode()
+
+
+class KeyShareTests(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username='crm', password='pw12345678')
+        self.client.force_login(self.user)
+
+    def _post(self, url, body):
+        return self.client.post(url, json.dumps(body), content_type='application/json')
+
+    def test_round_trip_hands_answer_out_once(self):
+        token = self._post('/accounts/keyshare/', {'pub': PUB, 'device': 'Firefox auf Linux'}).json()['token']
+        self.assertEqual(self.client.get(f'/accounts/keyshare/{token}/').json(), {'state': 'pending'})
+        pending = self.client.get('/accounts/keyshare/pending/').json()['requests']
+        self.assertEqual([(r['token'], r['pub'], r['device']) for r in pending], [(token, PUB, 'Firefox auf Linux')])
+
+        answer = {'pub': PUB, 'iv': base64.b64encode(b'i' * 12).decode(),
+                  'ct': base64.b64encode(b'c' * 48).decode()}
+        self.assertEqual(self._post(f'/accounts/keyshare/{token}/answer/', answer).status_code, 200)
+        self.assertEqual(self._post(f'/accounts/keyshare/{token}/answer/', answer).status_code, 409)
+        self.assertEqual(self.client.get('/accounts/keyshare/pending/').json()['requests'], [])
+        got = self.client.get(f'/accounts/keyshare/{token}/').json()
+        self.assertEqual(got, {'state': 'answered', **answer})
+        self.assertEqual(self.client.get(f'/accounts/keyshare/{token}/').json(), {'state': 'expired'})
+
+    def test_requests_of_devices_that_stopped_asking_are_not_offered(self):
+        token = self._post('/accounts/keyshare/', {'pub': PUB}).json()['token']
+        KeyShare.objects.update(polled_at=KeyShare.objects.get().created_at.replace(year=2000))
+        self.assertEqual(self.client.get('/accounts/keyshare/pending/').json()['requests'], [])
+        self.client.get(f'/accounts/keyshare/{token}/')  # asks again
+        self.assertEqual(len(self.client.get('/accounts/keyshare/pending/').json()['requests']), 1)
+
+    def test_decline(self):
+        token = self._post('/accounts/keyshare/', {'pub': PUB}).json()['token']
+        self._post(f'/accounts/keyshare/{token}/answer/', {'decline': True})
+        self.assertEqual(self.client.get(f'/accounts/keyshare/{token}/').json(), {'state': 'declined'})
+
+    def test_bad_input_is_refused(self):
+        self.assertEqual(self._post('/accounts/keyshare/', {'pub': 'nope'}).status_code, 400)
+        token = self._post('/accounts/keyshare/', {'pub': PUB}).json()['token']
+        bad = {'pub': PUB, 'iv': base64.b64encode(b'i' * 12).decode(), 'ct': base64.b64encode(b'c' * 10).decode()}
+        self.assertEqual(self._post(f'/accounts/keyshare/{token}/answer/', bad).status_code, 400)
+
+    def test_other_accounts_see_nothing(self):
+        token = self._post('/accounts/keyshare/', {'pub': PUB}).json()['token']
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        self.client.force_login(other)
+        self.assertEqual(self.client.get('/accounts/keyshare/pending/').json()['requests'], [])
+        self.assertEqual(self._post(f'/accounts/keyshare/{token}/answer/', {'decline': True}).status_code, 404)
+        self.client.logout()
+        self.assertEqual(self.client.get('/accounts/keyshare/pending/').status_code, 401)
+
+    def test_open_requests_are_limited_and_run_out(self):
+        tokens = [self._post('/accounts/keyshare/', {'pub': PUB}).json()['token'] for _ in range(7)]
+        self.assertEqual(KeyShare.objects.count(), 5)
+        self.assertEqual(self.client.get(f'/accounts/keyshare/{tokens[0]}/').json(), {'state': 'expired'})
+        self.assertEqual(self.client.get(f'/accounts/keyshare/{tokens[-1]}/').json(), {'state': 'pending'})
+        KeyShare.objects.update(created_at=KeyShare.objects.first().created_at.replace(year=2000))
+        self.assertEqual(self.client.get('/accounts/keyshare/pending/').json()['requests'], [])
+        self.assertEqual(self._post('/accounts/keyshare/', {'pub': PUB}).status_code, 200)

accounts/urls.py

@@ -1,6 +1,6 @@
 from django.contrib.auth.views import LogoutView
 from django.urls import path
-from . import views
+from . import keyshare, oidc_views, views
 
 urlpatterns = [
     path('register/', views.register, name='register'),
@@ -21,4 +21,15 @@ urlpatterns = [
     path('webdav/<int:pk>/test/', views.webdav_test, name='webdav_test'),
     path('webdav/<int:pk>/delete/', views.webdav_delete, name='webdav_delete'),
     path('vault/<slug:name>/', views.vault, name='vault'),
+    path('oidc/start/', oidc_views.oidc_start, name='oidc_start'),
+    path('oidc/callback/', oidc_views.oidc_callback, name='oidc_callback'),
+    path('oidc/app/', oidc_views.app_config, name='oidc_app_config'),
+    path('oidc/app/login/', oidc_views.app_login, name='oidc_app_login'),
+    path('oidc/key/', oidc_views.oidc_key, name='oidc_key'),
+    path('oidc/key/callback/', oidc_views.oidc_key, name='oidc_key_callback'),
+    path('oidc/<int:pk>/unlink/', oidc_views.oidc_unlink, name='oidc_unlink'),
+    path('keyshare/', keyshare.create, name='keyshare_create'),
+    path('keyshare/pending/', keyshare.pending, name='keyshare_pending'),
+    path('keyshare/<str:token>/', keyshare.status, name='keyshare_status'),
+    path('keyshare/<str:token>/answer/', keyshare.answer, name='keyshare_answer'),
 ]

accounts/views.py

@@ -13,6 +13,7 @@ from django.views.decorators.http import require_http_methods
 from radio import lastfm as lastfm_module
 
 from .models import ApiToken, UserVault, WebDAVSource
+from .oidc_views import oidc_context
 from .webdav import WebDAVError, list_directory
 
 User = get_user_model()
@@ -56,7 +57,7 @@ def login_view(request):
     else:
         form = AuthenticationForm()
 
-    return render(request, 'accounts/login.html', {'form': form})
+    return render(request, 'accounts/login.html', {'form': form, **oidc_context(request)})
 
 
 # ---------------------------------------------------------------------------
@@ -78,6 +79,7 @@ def settings_view(request):
         'password_form': PasswordChangeForm(request.user),
         'api_token': getattr(request.user, 'api_token', None),
         'webdav_sources': request.user.webdav_sources.all(),
+        **oidc_context(request),
     }
     return render(request, 'accounts/settings.html', context)
 
@@ -217,6 +219,7 @@ def change_password(request):
         'password_form_open': True,
         'api_token': getattr(request.user, 'api_token', None),
         'webdav_sources': request.user.webdav_sources.all(),
+        **oidc_context(request),
     })
 
 

deploy/whisper/Dockerfile

@@ -0,0 +1,10 @@
+# diora's built-in Whisper (docs/transcription.md). Built on the server next to diora-prod:
+#   docker compose build whisper
+FROM python:3.12-slim
+RUN apt-get update && apt-get install -y --no-install-recommends ffmpeg && rm -rf /var/lib/apt/lists/*
+RUN pip install --no-cache-dir faster-whisper==1.2.1 fastapi==0.143.1 uvicorn==0.54.0 python-multipart==0.0.32
+WORKDIR /app
+COPY server.py .
+ENV WHISPER_CACHE=/models HF_HOME=/models/hf
+EXPOSE 8000
+CMD ["uvicorn", "server:app", "--host", "0.0.0.0", "--port", "8000"]

deploy/whisper/server.py

@@ -0,0 +1,85 @@
+"""diora's built-in Whisper: a small OpenAI-compatible transcription server.
+
+POST /v1/audio/transcriptions (multipart: file, model, language, prompt, response_format)
+returns {text} or, with response_format=verbose_json, {language, duration, text, segments}.
+The `model` field is ignored; WHISPER_MODEL decides. The model is loaded on the first request
+and freed again after WHISPER_IDLE seconds without work, so it holds no memory while unused.
+One transcription at a time; parallel requests wait (the server has few cores to share).
+See docs/transcription.md.
+"""
+import os
+import subprocess
+import threading
+import time
+
+# Windows with a GPU: where the CUDA libraries (cuBLAS, cuDNN) are, separated by ';'.
+for _d in filter(None, os.environ.get('WHISPER_DLL_DIRS', '').split(';')):
+    os.add_dll_directory(_d)
+    os.environ['PATH'] = _d + os.pathsep + os.environ['PATH']
+
+import numpy as np  # noqa: E402
+from fastapi import FastAPI, File, Form, HTTPException, UploadFile  # noqa: E402
+from faster_whisper import WhisperModel  # noqa: E402
+
+MODEL = os.environ.get('WHISPER_MODEL', 'large-v3-turbo')
+THREADS = int(os.environ.get('WHISPER_THREADS', '4'))
+DEVICE = os.environ.get('WHISPER_DEVICE', 'cpu')            # cpu or cuda
+COMPUTE = os.environ.get('WHISPER_COMPUTE', 'int8')          # int8_float16 on a GPU
+IDLE = int(os.environ.get('WHISPER_IDLE', '600'))
+CACHE = os.environ.get('WHISPER_CACHE', '/models')
+
+app = FastAPI()
+work = threading.Lock()      # one transcription at a time
+state = {'model': None, 'last': 0.0}
+
+
+def model():
+    if state['model'] is None:
+        state['model'] = WhisperModel(MODEL, device=DEVICE, compute_type=COMPUTE, cpu_threads=THREADS,
+                                      download_root=CACHE)
+    return state['model']
+
+
+def unloader():
+    while True:
+        time.sleep(30)
+        if state['model'] is not None and time.time() - state['last'] > IDLE and work.acquire(blocking=False):
+            try:
+                state['model'] = None
+            finally:
+                work.release()
+
+
+threading.Thread(target=unloader, daemon=True).start()
+
+
+def decode(data: bytes) -> np.ndarray:
+    p = subprocess.run(['ffmpeg', '-nostdin', '-loglevel', 'error', '-i', 'pipe:0', '-ac', '1', '-ar', '16000',
+                        '-f', 'f32le', 'pipe:1'], input=data, capture_output=True)
+    if p.returncode != 0:
+        raise HTTPException(400, 'not audio: ' + p.stderr.decode('utf-8', 'replace')[-200:])
+    return np.frombuffer(p.stdout, np.float32)
+
+
+@app.get('/health')
+def health():
+    return {'ok': True, 'model': MODEL, 'device': DEVICE, 'loaded': state['model'] is not None}
+
+
+@app.post('/v1/audio/transcriptions')
+def transcriptions(file: UploadFile = File(...), language: str = Form(''), prompt: str = Form(''),
+                   response_format: str = Form('json')):
+    audio = decode(file.file.read())
+    t0 = time.time()
+    with work:
+        segs, info = model().transcribe(audio, language=language or None, initial_prompt=prompt or None,
+                                        beam_size=5, vad_filter=True)
+        segs = [{'id': i, 'start': round(s.start, 2), 'end': round(s.end, 2), 'text': s.text}
+                for i, s in enumerate(segs)]
+        state['last'] = time.time()
+    dur = len(audio) / 16000
+    print(f'{dur:.0f} s audio in {time.time() - t0:.0f} s', flush=True)
+    text = ''.join(s['text'] for s in segs).strip()
+    if response_format == 'verbose_json':
+        return {'task': 'transcribe', 'language': info.language, 'duration': dur, 'text': text, 'segments': segs}
+    return {'text': text}

diora/settings.py

@@ -84,6 +84,9 @@ DATABASES = {
     }
 }
 
+# The built-in Whisper instance (OpenAI-compatible, base such as http://whisper:8000/v1);
+# empty = none (docs/transcription.md).
+WHISPER_URL = os.environ.get('DIORA_WHISPER_URL', '').strip().rstrip('/')
 PODCAST_MAX_EPISODES_PER_FEED = int(os.environ.get('PODCAST_MAX_EPISODES_PER_FEED', '200'))
 
 AUTH_PASSWORD_VALIDATORS = [
@@ -138,6 +141,17 @@ DEFAULT_AUTO_FIELD = 'django.db.models.BigAutoField'
 
 LOGIN_URL = '/accounts/login/'
 LOGIN_REDIRECT_URL = '/'
+
+# Sign-in with Authelia (accounts/oidc.py); off while one of the three is empty.
+OIDC_ISSUER = os.environ.get('DIORA_OIDC_ISSUER', '')
+OIDC_CLIENT_ID = os.environ.get('DIORA_OIDC_CLIENT_ID', '')
+OIDC_CLIENT_SECRET = os.environ.get('DIORA_OIDC_CLIENT_SECRET', '')
+# Fixed callback address; empty = built from the request (needs the real https host).
+OIDC_REDIRECT_URI = os.environ.get('DIORA_OIDC_REDIRECT_URI', '')
+OIDC_PROVIDER_NAME = os.environ.get('DIORA_OIDC_PROVIDER_NAME', 'Authelia')
+# Public client the devices use to fetch their `diora_key` claim from the provider themselves
+# (docs/accounts.md, "Schlüssel über Authelia"); empty = off.
+OIDC_APP_CLIENT_ID = os.environ.get('DIORA_OIDC_APP_CLIENT_ID', '')
 LOGOUT_REDIRECT_URL = '/'
 
 # Last.fm

docs/accounts.md

@@ -0,0 +1,91 @@
+# Anmeldung und Bücher-Schlüssel
+
+## Zwei Wege hinein
+
+**Benutzername und Passwort** (`accounts/views.py` `login_view`): Das Login-Formular leitet im
+Browser aus dem Passwort den Bücher-Schlüssel ab (PBKDF2-SHA256, Salt `diora:<username>`,
+200 000 Runden) und legt ihn bis zur nächsten Seite in `localStorage['diora_pending_enc_key']`
+ab. Die App-Seite übernimmt ihn als `diora_enc_key_<userId>` (`32-crypto.js`, `99-init.js`).
+
+**Authelia (OpenID Connect)** (`accounts/oidc.py`, `accounts/oidc_views.py`): als zusätzlicher
+Weg, die Anmeldung mit Passwort bleibt. Der Ablauf ist Authorization Code mit PKCE (S256) und
+einem vertraulichen Client (`client_secret_basic`). Das id_token wird hier geprüft (RS256 gegen
+die JWKS des Anbieters, `iss`, `aud`, `exp`, `nonce`), dafür reicht `cryptography`.
+
+- Ein Konto wird **einmal verknüpft**: in den Einstellungen „Mit Authelia verknüpfen“, während
+  man normal angemeldet ist (`/accounts/oidc/start/?link=1`). Gespeichert wird `OidcIdentity`
+  (`issuer`, `sub`; Name und Mail nur zur Anzeige). Danach führt „Mit Authelia anmelden“ auf der
+  Login-Seite in dieses Konto.
+- Eine nicht verknüpfte Authelia-Identität wird abgewiesen. Es entstehen **keine neuen
+  Konten** über Authelia, und es wird auch nicht über Name oder Mail automatisch verknüpft.
+- Wer hinein darf, regelt Authelia (Policy `mitglieder` = Gruppen `admins` oder `freunde`).
+- Einstellungen per Umgebung: `DIORA_OIDC_ISSUER`, `DIORA_OIDC_CLIENT_ID`,
+  `DIORA_OIDC_CLIENT_SECRET`, `DIORA_OIDC_REDIRECT_URI` (fest, weil hinter dem Proxy kein
+  https erkannt wird), optional `DIORA_OIDC_PROVIDER_NAME`. Fehlt eins der ersten drei, gibt es
+  keinen Knopf. Auf prod stehen sie in `~/docker-compose/diora-prod/.env` (Strato), das
+  Gegenstück liegt auf Hetzner in `/opt/authelia/oidc-clients.env`.
+- Redirect-URI im Authelia-Client: `https://diora.creamfresh.xyz/accounts/oidc/callback/`.
+
+## Der Schlüssel nach einer Authelia-Anmeldung
+
+### Normalfall: über Authelia, ohne Nachfrage
+
+Authelia hält je Nutzer ein geheimes Attribut `diora_key` (32 Byte, base64). Es geht nur an den
+öffentlichen Client `diora-app` (`DIORA_OIDC_APP_CLIENT_ID`, PKCE, Scope `diora_key`) und nur
+über userinfo, nicht im id_token. Nach jedem Authelia-Login (und nach dem Verknüpfen) führt
+diora auf `/accounts/oidc/key/` (`templates/accounts/oidc_key.html`). Die Seite meldet sich
+**im Browser** noch einmal mit `prompt=none` als `diora-app` an, unsichtbar, weil die
+Authelia-Sitzung noch besteht, und holt per CORS Token und userinfo. Daraus leitet sie mit
+HKDF-SHA256 (Salt = `sub`, info `diora authelia key v1`) einen AES-GCM-Schlüssel ab:
+
+- Hat das Gerät den Bücher-Schlüssel (und passt er zu einem Buch), wird er damit verschlüsselt
+  und im Vault `authelia-key` abgelegt, bei jedem Authelia-Login neu. So heilt ein geänderter
+  `diora_key` von selbst.
+- Hat es keinen, öffnet sie den Vault-Eintrag damit und legt den Schlüssel ab.
+- Es zählt nur das Authelia-Konto, das mit diesem diora-Konto verknüpft ist (`sub`).
+
+Der diora-Server sieht `diora_key` nie. **Grenze, ehrlich:** Wer Authelia (Hetzner-Root) oder
+dein Authelia-Konto (Passwort, Mail-Reset) übernimmt, kommt damit an die Bücher. Wer Strato
+übernimmt, kann im Browser ohnehin manipuliertes JS ausliefern. Geschützt ist gegen geleakte
+Daten **einer** Seite (diora-DB/Backup ohne Authelia-Nutzerdatei oder umgekehrt). Geht
+`diora_key` verloren, sind die Bücher nicht weg: Der Schlüssel kommt weiter aus dem Passwort.
+Entschieden 2026-11-01 vom Nutzer (Komfort vor stärkerer Trennung).
+
+### Notweg: Code oder Passwort
+
+Ohne Passwort gibt es keinen Schlüssel. Deshalb erzeugt `getOrCreateEncKey()` **nie mehr
+selbst einen neuen Schlüssel**: Früher machte es einen zufälligen, wenn keiner da war, und damit
+waren die vorhandenen Bücher auf diesem Gerät verloren. Jetzt fragt es `dioraUnlock()`
+(`static/js/app/33-unlock.js`). Es gibt zwei Wege:
+
+1. **Von einem entsperrten Gerät übergeben** (`accounts/keyshare.py`, Modell `KeyShare`). Das
+   neue Gerät schickt einen ECDH-P-256-Public-Key und zeigt einen 6-stelligen Code, der aus
+   diesem Key berechnet wird. Entsperrte Geräte fragen alle 15 s
+   `/accounts/keyshare/pending/` ab und zeigen denselben Code mit „Freigeben/Ablehnen“. Die
+   Freigabe verschlüsselt den Bücher-Schlüssel mit AES-GCM unter dem gemeinsamen Geheimnis
+   (HKDF-SHA256, info `diora keyshare v1`). Der Server sieht nur Public Keys und Chiffrat und
+   gibt die Antwort genau einmal heraus. Anträge laufen nach 10 min ab. Angeboten werden nur
+   solche, deren Gerät in den letzten 20 s nachgefragt hat, und je Konto höchstens 5, die
+   ältesten weichen.
+2. **Das diora-Passwort einmal eingeben**: dieselbe Ableitung wie beim Login.
+
+In beiden Fällen wird der Schlüssel an einem Buch geprüft, bevor er gespeichert wird. Mit
+„Später“ bleiben Bücher gesperrt, und die Bibliothek zeigt einen Knopf „Entsperren“. Die
+Einstellungsseite erzeugt ebenfalls keinen Schlüssel mehr.
+
+## Apps
+
+Android und Desktop melden sich bisher nur mit Passwort an (eigene Login-Seiten in
+`diora-android/web-native/login.html` und `diora-desktop/src/web/login.js`, die einen
+API-Token holen). Mit Authelia melden sie sich selbst als öffentlicher Client `diora-app` an:
+
+1. `GET /accounts/oidc/app/` sagt, wo Authelia ist, gibt die Client-ID und eine vom Server
+   signierte nonce (10 min gültig).
+2. Die App führt den Code-Flow mit PKCE im Systembrowser aus. Redirect:
+   `xyz.creamfresh.diora:/oauth` bzw. `xyz.creamfresh.diora.reader:/oauth` (Android),
+   Loopback `http://127.0.0.1:<port>/oauth` (Desktop, jeder Port). Den Code tauscht sie selbst
+   und liest `diora_key` aus userinfo.
+3. `POST /accounts/oidc/app/login/ {id_token}` prüft das Token (`aud` = `diora-app`, nonce) und
+   gibt den API-Token des verknüpften Kontos heraus (den vorhandenen, kein neuer).
+4. Die App holt mit dem Token `/accounts/vault/authelia-key/` und öffnet den Bücher-Schlüssel
+   wie die Webseite (HKDF, Salt `sub`). Ist dort noch nichts, fragt sie nach dem diora-Passwort.

docs/ai.md

@@ -29,6 +29,8 @@ kept by the server as a blob it can't read:
   the key can't be read: the settings ask for it again. Not on diora Reader (e-readers).
 - Server-side jobs (in the background, without an app open) were left out on purpose: they
   would need the key on the server. The project Radar has its own server key (docs/projects.md).
+  Podcast articles wait on the server as a wish until a device with the key is open
+  (docs/transcription.md).
 
 ## Building blocks (`static/js/ai.js`, an ES module)
 
@@ -41,6 +43,8 @@ kept by the server as a blob it can't read:
   (`output_config.format` json_schema), effort `low` by default, streamed, `fallbacks: "default"`
   (beta `server-side-fallback-2026-07-01`) for declined requests; refusals and over-long
   answers become plain German messages.
+- `ask` also takes `onProgress(words)` (words written so far while streaming; OpenRouter is not streamed and reports once at the end)
+  and `signal` (an `AbortSignal`), both optional; `DioraAI.modelName()` is the model of the chosen provider. Used by the podcast articles.
 - `DioraAI.review({title, intro, items, onApply})`: the suggestions with checkboxes — nothing
   changes until the user ticks and presses "Übernehmen".
 - `DioraAI.settings()`, `withKey(fn)`, `progress()`, `showError()`, `toast()`; the sheets close
@@ -53,5 +57,6 @@ kept by the server as a blob it can't read:
 | Bücher, Zeitschriften | ✦ in the library's head (`static/js/app/55-library-tools.js`) | per book: id, title, author, folders, series, magazine mark; the existing folders | folders to add (existing first; new ones only for ≥ 3 books), series with volume, cleaned title/author where broken, magazine mark |
 | Podcasts | ✦ in Abos (diora-android `web-native/static/native-podcast.js`, copied to `static/podcasts/`) | per subscription: title, author, folders, level, heard in 12 months, publishing interval, three newest episode titles | folders to add (≤ 2), level Stamm / Strom / Auf Abruf (only with a clear reason) |
 | Nachrichten, Blogs | ✦ in the feeds' bar (`static/js/feeds.js`; the apps' section bar presses it) | per source: title, address, kind, Ressort or folders, recent titles shown | Ressort for news, folders for blogs, news ↔ blog when clearly wrong |
+| Podcast-Artikel | ✦ Artikel in an episode (`static/js/podcast-article.js`, docs/transcription.md) | the episode's transcript (feed or Whisper), title, author, shownotes, chapters | an article from the speakers' standpoint with summary and sections, saved to the episode and to Blogs (no review sheet: it is new text, removable) |
 
 Applied through `_updateBookMeta` (the encrypted meta, as when editing by hand).

docs/podcasts.md

@@ -182,3 +182,12 @@ Listening without committing: one button plays the Strom like a station.
 - Web host (`static/js/podcasts-web.js`, `DioraPodcastHost`): `podcastLaterAdd(json episodes)`,
   `podcastLaterRemove(json ids)`, `podcastListsSwitch(on)`, `podcastStreamSeen()` (keeps the
   previous time as `state.stream_seen_before` for the marks).
+
+## Transcripts
+
+An episode's Transkript tab shows one of two kinds. **From the feed**: `<podcast:transcript>`
+(SRT, VTT, JSON, Podlove), fetched and parsed on first use and cached in `EpisodeTranscript`
+(`podcasts/transcripts.py`, `GET /podcasts/episodes/<id>/transcript/`). **Made by Whisper**
+(since 2026-11): only when the user asks, kept on `TranscriptionJob`; the feed's wins when both
+exist. From a transcript the device can write an article that is pinned to the episode and shows
+in Blogs. All of that is in docs/transcription.md.

docs/transcription.md

@@ -0,0 +1,262 @@
+# Smarte Transkription: Transkript und Artikel zu einer Folge
+
+Decided with the user on 2026-11-09. Plan and order of work:
+diora-android `plans/smarte-transkription.md`.
+
+## What it is for
+
+Sometimes the shownotes don't tell whether an episode still has something to teach. Then the
+user presses a button and, a few hours later at most, gets a readable article written from the
+speakers' own standpoint. It is meant to let them judge the episode for themselves, so it keeps
+arguments, reasons and evidence and does not just summarise them. The article sits on the
+episode and in Blogs.
+
+Nothing happens on its own. The server never transcribes ahead, by cron or on subscribing; the
+only way to start a transcription is the `POST` below, sent when the user presses a button.
+
+## Who does what
+
+The **server** transcribes. It already downloads and decodes episodes for skipping silence
+(`podcasts/silence.py`), and browsers and the WebView can't fetch podcast audio (CORS). It cuts
+the audio at pauses and sends the pieces in parallel to a Whisper instance with an
+OpenAI-compatible API (`POST /v1/audio/transcriptions`): the built-in one by default (`DIORA_WHISPER_URL`),
+or any instance the user enters. YouTube episodes are
+not transcribed: the server never fetches their sound (docs/youtube.md).
+
+The **device** writes the article with the user's own key (docs/ai.md). The key never reaches the
+server. It is one LLM call of a few minutes. If diora is open when the transcript is done, it
+happens right away; otherwise the next time diora is opened on any of the user's devices. The
+server only remembers that an article was wanted.
+
+Transcript and article are stored in plain text on the server, like all podcast and feed data.
+
+## Flow
+
+1. In the episode the user presses "Transkript erstellen" (the Transkript tab) or
+   "✦ Artikel erstellen" / "✦ Transkribieren und Artikel erstellen". The second sends
+   `article: true`. With a transcript already there (from the feed or an earlier run) only the
+   article is written.
+2. The server works through the episode (minutes to hours, no GPU on Strato). The UI shows
+   "Transkribiere 7/38 …" while it is open.
+3. Done: a notice for `GET /feeds/notifications/`. Android's BackgroundSync and the desktop show
+   "Transkript fertig: <Folge>", with "Artikel wird beim Öffnen geschrieben" when one is wanted.
+   Tapping it opens diora at the episode.
+4. A device with a key (`static/js/podcast-article.js`) picks up the open wishes, reserves one,
+   writes the article and saves it. It then says "Artikel fertig" (on Android as a system
+   notification through the bridge, `DioraNative.notify(title, text, target)` with target
+   `episode:<id>` or `item:<id>`; elsewhere a toast) and opens it on a tap.
+5. The article is the first tab of the episode ("Artikel") and an item of the favourite blog
+   "Podcast-Artikel", linking back to the episode.
+
+## API
+
+**Settings.** `GET /podcasts/transcription/settings/` returns
+`{engine: "builtin"|"custom", url, model, language, parallel, has_token, builtin_available}`.
+`POST` takes the same fields and `token` (write-only; `""` removes it). The token is kept
+encrypted with Fernet like `feeds.SiteLogin.secret`. `language` is `""` for automatic or an ISO
+code (`de`). `parallel` is 1 to 8, default 3. `builtin_available` is whether `DIORA_WHISPER_URL`
+is set. `url` is the instance's base, e.g. `https://whisper.example.org/v1`; the server appends
+`/audio/transcriptions`.
+
+An empty `model` is sent as `whisper-1`. The built-in URL is read into `settings.WHISPER_URL` from
+`DIORA_WHISPER_URL`. A custom engine needs a `url`; the token is only sent to custom instances.
+
+`POST /podcasts/transcription/test/` sends one second of silence to the configured instance and
+returns `{ok, model, error}`. A custom URL goes through the same SSRF check as chapters
+(`podcasts/chapters.py` `_open`): no private, loopback or link-local addresses. Only the built-in
+URL from `DIORA_WHISPER_URL` is exempt.
+
+**Job.** `POST /podcasts/episodes/<id>/transcribe/` with `{article: true|false}` starts a job, or
+updates `want_article` on a running one. With a transcript already there it returns
+`{state: "done"}` straight away (and keeps `want_article`). `GET` on the same address returns
+`{state, chunks_done, chunks_total, error, want_article, started_at, finished_at}`, with `state`
+one of `none`, `queued`, `downloading`, `transcribing`, `done`, `error`. `DELETE` cancels a
+queued or running job (its row is deleted, so `GET` then says `none`; a finished one stays). The server runs one job at a time; the others wait as `queued`. A YouTube
+episode returns 400 `{error: "youtube"}`, an episode over 800 MB 400 `{error: "too large"}`. The size is asked with a `HEAD`
+request (`Content-Length`; if that fails the job starts anyway) and checked again while
+downloading, where it ends the job with the error `too large`. `article: true` is only ever
+raised, never lowered, by a repeated `POST`; a `POST` after an `error` starts a new job (at the
+back of the queue). Jobs run in id order, which is why the notice below can use ids as its cursor.
+A feed transcript already there gives a job that is `done` with no `data`: it only holds the wish
+for an article and causes no `transcripts` notice.
+
+**Transcript.** The result is kept on the job (`TranscriptionJob.data`) in the format of
+`EpisodeTranscript.data`: `{segments: [{start, end, speaker, text}], source: {type: "whisper",
+model, language}}`. `speaker` is `""`. It is kept apart from `EpisodeTranscript`, which is a cache
+of the feed's transcripts and is rewritten when the feed changes them.
+`GET /podcasts/episodes/<id>/transcript/` returns the feed's transcript when the feed has a
+usable one, else the Whisper one, else 404 as before. `episode_detail` adds
+`transcript_origin` (`"feed"`, `"whisper"` or `null`), `has_transcript` is true for both, and it
+adds `article_item_id` (or `null`) and `transcription` (the job's state as above, or `null`).
+
+**Article.** `POST /podcasts/episodes/<id>/article/` takes `{title, summary, body_html, model,
+kind}` and returns `{item_id}`. The server cleans `body_html` with `feeds/fetch.py clean_html`.
+It creates the user's source "Podcast-Artikel" when missing (`kind` blog, `fav` true,
+`synthetic` true, `url` `diora://podcast-artikel`) and adds a `feeds.Item` there, or replaces the
+episode's earlier one. Item fields: the episode's link (else its audio URL) as `url`, the feed's
+author as `author`, the episode's artwork as `image_url`, now as `published` (so it is on top of
+the blog stream under "heute"), `words` counted. Saving clears the wish. `DELETE` on the same
+address removes the article and its item. `GET /feeds/items/<id>/` and `_item_data` carry
+`episode_id` for such items.
+`body_html` may contain `href="#t=<seconds>"` links; `clean_html` keeps exactly this fragment form
+(`podcasts/shownotes.py` `TIME_LINK`) and no other `#` links. Validation: `title` 1 to 300
+characters, `body_html` non-empty after cleaning and at most 1,000,000 characters, `summary` at most
+2000, `model` at most 100, `kind` at most 20 (400 otherwise). The item has `content_full` true (no
+page is ever fetched) and, as the reader's furniture filter would strip repeated "▶ ab" blocks,
+`GET /feeds/items/<id>/` skips that filter for synthetic sources. The endpoint is
+`podcasts/article.py`.
+
+Synthetic sources are skipped by the feed refresh (`refresh_newsfeeds`, `refresh_now`) and by the
+pruning of old blog items (`feeds/refresh.py`). They can be hidden or have their bell turned off
+like any blog; they can't be refreshed, renamed to another URL or turned into news. `refresh.refresh()` itself drops them, the OPML export leaves them out, and `_source_data` carries `synthetic`.
+
+**Open wishes.** `GET /podcasts/transcription/pending/` returns
+`[{episode_id, title, podcast, duration_seconds}]`: the user's episodes whose transcript is done,
+that want an article, have none and aren't reserved. `POST
+/podcasts/episodes/<id>/article/claim/` reserves one for 15 minutes for the calling device
+(200), or answers 409 when another device holds it, so phone and desktop don't both write it.
+`DELETE` on the claim releases it (the device gave up, e.g. no key or an error the user saw).
+The device names itself with `device` in the JSON body (or, if that is missing, the
+`X-Diora-Device` header): a short stable name such as `android`, `desktop` or `browser`, at most 60
+characters; a claim without one is 400. The same device claiming again renews its 15 minutes. A
+claim on an episode with no open wish (no job done, no article wanted, or an article already
+there) is 404. `DELETE` by a device that does not hold the claim is 409.
+
+**Notice.** `GET /feeds/notifications/` additionally takes `t_after` and returns
+`transcripts: [{job_id, episode_id, title, podcast, want_article}]` (jobs done after that id) and
+`t_last`. Without `t_after` (a device's first time) the list is empty and only `t_last` comes,
+like `after`/`last`. Devices that don't know the fields ignore them.
+
+## Transcribing (`podcasts/transcribe.py`)
+
+Download with the same limits as `silence.py`, decode with imageio-ffmpeg to 16 kHz mono. Cut
+into pieces of about eight minutes; each cut moves to the longest pause within ±30 seconds
+(`silence.py`'s detection), so no word is cut. Pieces go out as FLAC to the built-in instance
+and as Opus 32 kbit/s to custom ones, to keep uploads small.
+
+Each piece is sent with `model`, `language` (when set), `response_format=verbose_json` and a
+`prompt` of podcast title, episode title and author names, so Whisper spells names right. Pieces
+run in a thread pool of `parallel` workers; each is retried up to three times with a growing
+pause; progress is written after each. Segment times are shifted by the piece's start; the
+segments are joined and smoothed with `transcripts.merge`. A job runs in a daemon thread, like
+`silence.ensure()`; one per server. A job found `downloading` or `transcribing` after a restart
+is started again. This happens when the first
+request (`GET`/`POST` on `/transcribe/` or `/transcription/pending/`) after a start reaches the server.
+Cancelling deletes the job's row; the worker notices at the latest after five seconds, stops
+sending pieces and removes its temporary files (they live in a temporary directory that is always
+removed). Only the worker's main thread writes to the database. The settings used are those of the
+episode's owner, read when the job starts. A refusal by the instance with a 4xx status other than
+408 and 429 is not retried and ends the job with `Whisper: HTTP <code>`.
+
+## The article (`static/js/podcast-article.js`)
+
+Input: the transcript as plain text, one line per segment group of about 30 seconds,
+`[mm:ss] Sprecher: text` (or `[mm:ss] text` without speakers), plus the episode's metadata:
+podcast title, author, episode title, date, duration, the shownotes as plain text and the
+chapters. Over about 120,000 tokens (more than about nine hours) the device stops with a clear
+message for now. A five-hour episode is about 70,000 tokens.
+
+`DioraAI.ask({system, prompt, schema, effort: "medium", maxTokens: 32000, onProgress, signal})`.
+`onProgress(words)` reports the words written so far while streaming ("Schreibe Artikel … 2.300
+Wörter"); `signal` aborts. Both are optional and new in ai.js; the existing callers don't change.
+
+Schema:
+
+```json
+{
+  "type": "object",
+  "additionalProperties": false,
+  "required": ["title", "summary", "kind", "speakers", "sections"],
+  "properties": {
+    "title": {"type": "string"},
+    "summary": {"type": "string"},
+    "kind": {"type": "string", "enum": ["Interview", "Gespräch", "Monolog", "Diskussion"]},
+    "speakers": {"type": "array", "items": {"type": "string"}},
+    "sections": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "required": ["heading", "start", "body_markdown"],
+        "properties": {
+          "heading": {"type": "string"},
+          "start": {"type": "integer"},
+          "body_markdown": {"type": "string"}
+        }
+      }
+    }
+  }
+}
+```
+
+The HTML: the summary as a lead paragraph, then per section an `<h2>`, below it a link
+"▶ ab 12:34" (`<a href="#t=754">`; `clean_html` keeps fragment links, and feeds.js plays the
+item's `episode_id` from there when such a link is tapped) and the body. Markdown is limited to paragraphs, emphasis, lists and block
+quotes, converted on the device; the server cleans the result anyway.
+
+System prompt (the user's words decide the perspective; keep its substance when changing it):
+
+```text
+You turn the transcript of a podcast episode into an article that reads as if the people
+speaking had written it themselves instead of recording the episode.
+
+The reader wants to decide whether this episode still has something to teach them. They do not
+want a summary; they want to understand what is argued here well enough to judge it for
+themselves. Write for that reader.
+
+Perspective. Write from the speakers' standpoint and state their positions directly, as their
+own claims. Never distance yourself from them: no "im Podcast wird die These aufgestellt", "der
+Sprecher meint", "in dieser Folge geht es um", "die Gäste diskutieren", "laut X". If the episode
+is one person speaking, write as that person argues, in the third person or impersonal voice,
+but without reporting verbs. Do not add your own opinion, criticism or fact checks, and do not
+soften what is said.
+
+Conversations. Interviews, conversations and discussions are condensed into their core theses;
+the form of question and answer disappears. Where people hold different positions, keep them
+apart and name who holds which ("Zuckermann hält dagegen, dass …"), but do not turn it into
+minutes. Questions from a host matter only where they shape the argument. When the transcript
+has no speaker names, infer them from the context (introductions, the shownotes, how people
+address each other); if you can't, do not invent names.
+
+Substance over summary. Give the arguments with their reasons, examples and evidence, not just
+the conclusions. Keep figures, dates, quotations that carry an argument (short, in quotation
+marks, attributed), and every work mentioned (books, articles, studies, films) with author and
+title. Make clear what is new, distinctive or surprising in the episode and where it restates
+what is commonly said, by how you weight and order it, without grading it.
+
+Leave out: advertising and sponsors, appeals for donations, subscriptions or ratings, notes on
+the podcast itself (schedules, technology, community), greetings, small talk and digressions that
+lead nowhere.
+
+Form. Write in the language of the episode. `title`: a title for the article, not the episode
+title repeated. `summary`: three to six sentences up front that state the central thesis and how
+the episode gets there. `sections`: as many as the content needs (usually four to ten), each
+with a heading that says something (not "Einleitung"), `start` = the second in the episode where
+its content begins (from the [mm:ss] marks), and `body_markdown` in plain paragraphs; lists only
+where the speakers list things themselves. Follow the order of the episode unless a different
+order makes the argument clearer. Length: as long as needed for the reader to judge the episode,
+roughly a tenth of the transcript, at most about 5,000 words.
+
+The transcript was made by speech recognition: fix obvious mishearings, especially of names,
+using the metadata, and don't quote garbled passages.
+```
+
+Prompt (user turn): the metadata block, then `Transkript:` and the text.
+
+## The built-in instance (since 2026-11-09)
+
+Whisper does not fit on the Strato server: with about 2 GB of memory free, large-v3-turbo needed
+more than 2.2 GB and even `small` broke a 1.2 GB limit, and earlyoom then killed other services.
+So the built-in instance runs on the user's PC (multi11, GTX 1660 Ti) and diora-prod reaches it over
+the tailnet: `DIORA_WHISPER_URL=http://100.64.0.3:8766/v1` in `~/docker-compose/diora-prod/.env`.
+It is `deploy/whisper/server.py` (a small OpenAI-compatible server around faster-whisper,
+large-v3-turbo, int8_float16 on CUDA; one request at a time; the model is loaded on demand and freed
+after 10 idle minutes), started at logon from `F:\diora-whisper` (start.cmd via start-hidden.vbs,
+shortcut in the user's Startup folder; log `F:\diora-whisper\server.log`), listening only on the
+tailnet address. One hour of speech takes about five minutes.
+
+When the PC is off, a job does not fail: the server checks the instance before downloading, and a
+connection error (there or while pieces are out) puts the job back into the queue with
+`error = "Whisper-Instanz nicht erreichbar, neuer Versuch um HH:MM"`, tried again every ten
+minutes (`WAIT_RETRY`) for up to two days after the button was pressed (`WAIT_MAX`). Only the
+built-in instance waits like this; a custom instance's connection errors fail the job as before.

e2e/fixtures/make_test_pdf.py

@@ -0,0 +1,129 @@
+"""A small PDF with the cases the PDF reader has to get right (e2e and by hand).
+
+  python e2e/fixtures/make_test_pdf.py e2e/fixtures/test.pdf
+
+12 pages: A4 portrait text pages, one A5 page and one A4 landscape page (mixed
+sizes), page 7 nearly empty (only a page number: the smart crop must not zoom
+it to infinity), an outline (table of contents) with three entries, internal
+links from page 1 to pages 5 and 9 and one external link, a coloured picture on
+page 2 (it must stay as it is in the reader's colours) and page labels like a
+book's: i, ii for the first two pages, then 1, 2, 3 … Plain PDF 1.4, Helvetica,
+no dependencies.
+"""
+import sys
+
+A4 = (595, 842)
+A5 = (420, 595)
+A4L = (842, 595)
+
+WORDS = ('Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. '
+         'Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. ').split()
+
+
+def esc(t):
+    return t.replace('\\', '\\\\').replace('(', '\\(').replace(')', '\\)')
+
+
+def text_lines(n, words_per_line=11):
+    out, i = [], 0
+    for _ in range(n):
+        out.append(' '.join(WORDS[(i + k) % len(WORDS)] for k in range(words_per_line)))
+        i += words_per_line
+    return out
+
+
+def page_content(no, size):
+    w, h = size
+    ops = ['BT', '/F1 22 Tf', f'50 {h - 70} Td', f'(Kapitel {no} - Seite {no}) Tj', 'ET']
+    if no == 7:
+        # Nearly empty: just the number at the bottom.
+        return '\n'.join(['BT', '/F1 10 Tf', f'{w / 2 - 10:.0f} 30 Td', f'({no}) Tj', 'ET'])
+    y = h - 110
+    lines = text_lines(28 if h > 700 else 18, 11 if w > 500 else 8)
+    ops += ['BT', '/F1 11 Tf', f'50 {y} Td', '14 TL']
+    for ln in lines:
+        ops.append(f'({esc(ln)}) Tj T*')
+    ops.append('ET')
+    if no == 2:
+        ops += ['q', f'200 0 0 120 {w - 260} 60 cm', '/Im1 Do', 'Q']
+    if no == 1:
+        ops += ['BT', '/F1 12 Tf', f'50 {y - 30 - 14 * len(lines)} Td',
+                '(Link zu Seite 5) Tj', '0 -20 Td', '(Link zu Seite 9) Tj', '0 -20 Td', '(Externer Link: example.org) Tj', 'ET']
+    ops += ['BT', '/F1 10 Tf', f'{w / 2 - 10:.0f} 30 Td', f'({no}) Tj', 'ET']
+    return '\n'.join(ops)
+
+
+def build(path):
+    sizes = [A4] * 12
+    sizes[3] = A5          # page 4
+    sizes[8] = A4L         # page 9
+    objs = []              # index = object number - 1
+
+    def add(body):
+        objs.append(body)
+        return len(objs)
+
+    catalog = add(None)
+    pages_obj = add(None)
+    font = add('<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /Encoding /WinAnsiEncoding >>')
+    # A picture of four colours (2 x 2 pixels, drawn large).
+    pixels = b'FF0000 00A000 0000FF FFD700>'
+    image = add(b'<< /Type /XObject /Subtype /Image /Width 2 /Height 2 /ColorSpace /DeviceRGB '
+                b'/BitsPerComponent 8 /Filter /ASCIIHexDecode /Length %d >>\nstream\n' % len(pixels)
+                + pixels + b'\nendstream')
+    page_ids = []
+    for i, size in enumerate(sizes):
+        content = page_content(i + 1, size).encode('cp1252')
+        cid = add(b'<< /Length %d >>\nstream\n' % len(content) + content + b'\nendstream')
+        pid = add(None)
+        page_ids.append((pid, cid, size))
+    # links on page 1 (placed where the text is: bottom part of the text block)
+    p1_h = sizes[0][1]
+    base_y = p1_h - 110 - 30 - 14 * 28
+
+    def link_to(page_no, y):
+        return add(f'<< /Type /Annot /Subtype /Link /Rect [48 {y - 4} 220 {y + 14}] /Border [0 0 0] '
+                   f'/Dest [{page_ids[page_no - 1][0]} 0 R /XYZ 0 {sizes[page_no - 1][1]} null] >>')
+    annots = [link_to(5, base_y), link_to(9, base_y - 20),
+              add(f'<< /Type /Annot /Subtype /Link /Rect [48 {base_y - 44} 300 {base_y - 26}] /Border [0 0 0] '
+                  f'/A << /S /URI /URI (https://example.org/) >> >>')]
+    for n, (pid, cid, (w, h)) in enumerate(page_ids):
+        extra = f' /Annots [{" ".join(f"{a} 0 R" for a in annots)}]' if n == 0 else ''
+        xobj = f' /XObject << /Im1 {image} 0 R >>' if n == 1 else ''
+        objs[pid - 1] = (f'<< /Type /Page /Parent {pages_obj} 0 R /MediaBox [0 0 {w} {h}] '
+                         f'/Resources << /Font << /F1 {font} 0 R >>{xobj} >> /Contents {cid} 0 R{extra} >>')
+    objs[pages_obj - 1] = f'<< /Type /Pages /Kids [{" ".join(f"{p} 0 R" for p, _, _ in page_ids)}] /Count {len(page_ids)} >>'
+    # outline: three entries
+    outline = add(None)
+    entries = [('Anfang', 1), ('Querformat', 9), ('Fast leer', 7)]
+    ids = [add(None) for _ in entries]
+    for k, ((title, page), oid) in enumerate(zip(entries, ids)):
+        nav = ''
+        if k > 0:
+            nav += f' /Prev {ids[k - 1]} 0 R'
+        if k < len(ids) - 1:
+            nav += f' /Next {ids[k + 1]} 0 R'
+        objs[oid - 1] = (f'<< /Title ({esc(title)}) /Parent {outline} 0 R{nav} '
+                         f'/Dest [{page_ids[page - 1][0]} 0 R /XYZ 0 {sizes[page - 1][1]} null] >>')
+    objs[outline - 1] = f'<< /Type /Outlines /First {ids[0]} 0 R /Last {ids[-1]} 0 R /Count {len(ids)} >>'
+    objs[catalog - 1] = (f'<< /Type /Catalog /Pages {pages_obj} 0 R /Outlines {outline} 0 R /PageMode /UseOutlines '
+                         f'/PageLabels << /Nums [0 << /S /r >> 2 << /S /D >>] >> >>')
+    info = add('<< /Title (Testdokument PDF) /Author (diora Test) >>')
+
+    out = bytearray(b'%PDF-1.4\n%\xe2\xe3\xcf\xd3\n')
+    offsets = []
+    for i, body in enumerate(objs):
+        offsets.append(len(out))
+        data = body if isinstance(body, bytes) else body.encode('cp1252')
+        out += b'%d 0 obj\n' % (i + 1) + data + b'\nendobj\n'
+    xref = len(out)
+    out += b'xref\n0 %d\n0000000000 65535 f \n' % (len(objs) + 1)
+    for off in offsets:
+        out += b'%010d 00000 n \n' % off
+    out += (b'trailer\n<< /Size %d /Root %d 0 R /Info %d 0 R >>\nstartxref\n%d\n%%%%EOF\n'
+            % (len(objs) + 1, catalog, info, xref))
+    open(path, 'wb').write(out)
+
+
+if __name__ == '__main__':
+    build(sys.argv[1] if len(sys.argv) > 1 else 'test.pdf')

e2e/fixtures/test.pdf

@@ -0,0 +1,652 @@
+%PDF-1.4
+%âãÏÓ
+1 0 obj
+<< /Type /Catalog /Pages 2 0 R /Outlines 32 0 R /PageMode /UseOutlines /PageLabels << /Nums [0 << /S /r >> 2 << /S /D >>] >> >>
+endobj
+2 0 obj
+<< /Type /Pages /Kids [6 0 R 8 0 R 10 0 R 12 0 R 14 0 R 16 0 R 18 0 R 20 0 R 22 0 R 24 0 R 26 0 R 28 0 R] /Count 12 >>
+endobj
+3 0 obj
+<< /Type /Font /Subtype /Type1 /BaseFont /Helvetica /Encoding /WinAnsiEncoding >>
+endobj
+4 0 obj
+<< /Type /XObject /Subtype /Image /Width 2 /Height 2 /ColorSpace /DeviceRGB /BitsPerComponent 8 /Filter /ASCIIHexDecode /Length 28 >>
+stream
+FF0000 00A000 0000FF FFD700>
+endstream
+endobj
+5 0 obj
+<< /Length 2463 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 1 - Seite 1) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 12 Tf
+50 310 Td
+(Link zu Seite 5) Tj
+0 -20 Td
+(Link zu Seite 9) Tj
+0 -20 Td
+(Externer Link: example.org) Tj
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(1) Tj
+ET
+endstream
+endobj
+6 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 5 0 R /Annots [29 0 R 30 0 R 31 0 R] >>
+endobj
+7 0 obj
+<< /Length 2379 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 2 - Seite 2) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+q
+200 0 0 120 335 60 cm
+/Im1 Do
+Q
+BT
+/F1 10 Tf
+288 30 Td
+(2) Tj
+ET
+endstream
+endobj
+8 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> /XObject << /Im1 4 0 R >> >> /Contents 7 0 R >>
+endobj
+9 0 obj
+<< /Length 2345 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 3 - Seite 3) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(3) Tj
+ET
+endstream
+endobj
+10 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 9 0 R >>
+endobj
+11 0 obj
+<< /Length 1196 >>
+stream
+BT
+/F1 22 Tf
+50 525 Td
+(Kapitel 4 - Seite 4) Tj
+ET
+BT
+/F1 11 Tf
+50 485 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+ET
+BT
+/F1 10 Tf
+200 30 Td
+(4) Tj
+ET
+endstream
+endobj
+12 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 420 595] /Resources << /Font << /F1 3 0 R >> >> /Contents 11 0 R >>
+endobj
+13 0 obj
+<< /Length 2345 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 5 - Seite 5) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(5) Tj
+ET
+endstream
+endobj
+14 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 13 0 R >>
+endobj
+15 0 obj
+<< /Length 2345 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 6 - Seite 6) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(6) Tj
+ET
+endstream
+endobj
+16 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 15 0 R >>
+endobj
+17 0 obj
+<< /Length 32 >>
+stream
+BT
+/F1 10 Tf
+288 30 Td
+(7) Tj
+ET
+endstream
+endobj
+18 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 17 0 R >>
+endobj
+19 0 obj
+<< /Length 2345 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 8 - Seite 8) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(8) Tj
+ET
+endstream
+endobj
+20 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 19 0 R >>
+endobj
+21 0 obj
+<< /Length 1547 >>
+stream
+BT
+/F1 22 Tf
+50 525 Td
+(Kapitel 9 - Seite 9) Tj
+ET
+BT
+/F1 11 Tf
+50 485 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+ET
+BT
+/F1 10 Tf
+411 30 Td
+(9) Tj
+ET
+endstream
+endobj
+22 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 842 595] /Resources << /Font << /F1 3 0 R >> >> /Contents 21 0 R >>
+endobj
+23 0 obj
+<< /Length 2348 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 10 - Seite 10) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(10) Tj
+ET
+endstream
+endobj
+24 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 23 0 R >>
+endobj
+25 0 obj
+<< /Length 2348 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 11 - Seite 11) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(11) Tj
+ET
+endstream
+endobj
+26 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 25 0 R >>
+endobj
+27 0 obj
+<< /Length 2348 >>
+stream
+BT
+/F1 22 Tf
+50 772 Td
+(Kapitel 12 - Seite 12) Tj
+ET
+BT
+/F1 11 Tf
+50 732 Td
+14 TL
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+(etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed) Tj T*
+(do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und) Tj T*
+(Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit,) Tj T*
+(sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche) Tj T*
+(und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur adipiscing) Tj T*
+(elit, sed do eiusmod tempor. Die Seite trägt Text, damit Auswahl,) Tj T*
+(Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet, consectetur) Tj T*
+(adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text, damit) Tj T*
+(Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit amet,) Tj T*
+(consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt Text,) Tj T*
+(damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor sit) Tj T*
+(amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite trägt) Tj T*
+(Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum dolor) Tj T*
+(sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die Seite) Tj T*
+(trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem ipsum) Tj T*
+(dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor. Die) Tj T*
+(Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden. Lorem) Tj T*
+(ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod tempor.) Tj T*
+(Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas finden.) Tj T*
+(Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do eiusmod) Tj T*
+(tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen etwas) Tj T*
+(finden. Lorem ipsum dolor sit amet, consectetur adipiscing elit, sed do) Tj T*
+(eiusmod tempor. Die Seite trägt Text, damit Auswahl, Suche und Vorlesen) Tj T*
+ET
+BT
+/F1 10 Tf
+288 30 Td
+(12) Tj
+ET
+endstream
+endobj
+28 0 obj
+<< /Type /Page /Parent 2 0 R /MediaBox [0 0 595 842] /Resources << /Font << /F1 3 0 R >> >> /Contents 27 0 R >>
+endobj
+29 0 obj
+<< /Type /Annot /Subtype /Link /Rect [48 306 220 324] /Border [0 0 0] /Dest [14 0 R /XYZ 0 842 null] >>
+endobj
+30 0 obj
+<< /Type /Annot /Subtype /Link /Rect [48 286 220 304] /Border [0 0 0] /Dest [22 0 R /XYZ 0 595 null] >>
+endobj
+31 0 obj
+<< /Type /Annot /Subtype /Link /Rect [48 266 300 284] /Border [0 0 0] /A << /S /URI /URI (https://example.org/) >> >>
+endobj
+32 0 obj
+<< /Type /Outlines /First 33 0 R /Last 35 0 R /Count 3 >>
+endobj
+33 0 obj
+<< /Title (Anfang) /Parent 32 0 R /Next 34 0 R /Dest [6 0 R /XYZ 0 842 null] >>
+endobj
+34 0 obj
+<< /Title (Querformat) /Parent 32 0 R /Prev 33 0 R /Next 35 0 R /Dest [22 0 R /XYZ 0 595 null] >>
+endobj
+35 0 obj
+<< /Title (Fast leer) /Parent 32 0 R /Prev 34 0 R /Dest [18 0 R /XYZ 0 842 null] >>
+endobj
+36 0 obj
+<< /Title (Testdokument PDF) /Author (diora Test) >>
+endobj
+xref
+0 37
+0000000000 65535 f 
+0000000015 00000 n 
+0000000158 00000 n 
+0000000292 00000 n 
+0000000389 00000 n 
+0000000584 00000 n 
+0000003099 00000 n 
+0000003256 00000 n 
+0000005687 00000 n 
+0000005839 00000 n 
+0000008236 00000 n 
+0000008363 00000 n 
+0000009612 00000 n 
+0000009740 00000 n 
+0000012138 00000 n 
+0000012266 00000 n 
+0000014664 00000 n 
+0000014792 00000 n 
+0000014875 00000 n 
+0000015003 00000 n 
+0000017401 00000 n 
+0000017529 00000 n 
+0000019129 00000 n 
+0000019257 00000 n 
+0000021658 00000 n 
+0000021786 00000 n 
+0000024187 00000 n 
+0000024315 00000 n 
+0000026716 00000 n 
+0000026844 00000 n 
+0000026964 00000 n 
+0000027084 00000 n 
+0000027218 00000 n 
+0000027292 00000 n 
+0000027388 00000 n 
+0000027502 00000 n 
+0000027602 00000 n 
+trailer
+<< /Size 37 /Root 1 0 R /Info 36 0 R >>
+startxref
+27671
+%%EOF

feeds/management/commands/refresh_newsfeeds.py

@@ -13,7 +13,7 @@ class Command(BaseCommand):
         parser.add_argument('--source-id', type=int, help='Refresh only this source (even when backing off)')
 
     def handle(self, *args, **options):
-        qs = Source.objects.select_related('user').order_by('last_refreshed_at')
+        qs = Source.objects.filter(synthetic=False).select_related('user').order_by('last_refreshed_at')
         if options['source_id']:
             qs = qs.filter(pk=options['source_id'])
         else:

feeds/migrations/0007_transcription.py

@@ -0,0 +1,18 @@
+# Generated by Django 6.1 on 2026-11-09 08:20
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('feeds', '0006_folder_order'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='source',
+            name='synthetic',
+            field=models.BooleanField(default=False),
+        ),
+    ]

feeds/models.py

@@ -35,6 +35,9 @@ class Source(models.Model):
     fav = models.BooleanField(default=False)
     hidden = models.BooleanField(default=False)
     notify = models.BooleanField(default=False)
+    # Filled by diora itself (the blog "Podcast-Artikel"), not by a feed: never
+    # fetched, pruned, given another URL or turned into news.
+    synthetic = models.BooleanField(default=False)
 
     # Refresh bookkeeping, as for podcast feeds (see podcasts.views._fetch_feed).
     last_refreshed_at = models.DateTimeField(null=True, blank=True)

feeds/refresh.py

@@ -216,7 +216,7 @@ def expire(user, now=None):
     Item.objects.filter(user=user, gone=True, published__lt=now - timedelta(days=GONE_KEEP_DAYS)).delete()
     Story.objects.filter(user=user).exclude(items__gone=False).delete()
     # Blogs: the oldest read articles beyond BLOG_KEEP per blog.
-    for source in Source.objects.filter(user=user, kind=Source.BLOG):
+    for source in Source.objects.filter(user=user, kind=Source.BLOG, synthetic=False):
         ids = list(source.items.order_by('-published').values_list('id', flat=True)[BLOG_KEEP:])
         if ids:
             Item.objects.filter(id__in=ids, starred=False).exclude(read_at=None).delete()
@@ -225,7 +225,7 @@ def expire(user, now=None):
 def refresh(sources, deadline=None):
     """Refresh `sources` (downloads in parallel, writes one by one).
     Returns (refreshed, failed, new items)."""
-    sources = list(sources)
+    sources = [s for s in sources if not s.synthetic]   # synthetic sources have no feed to fetch
     now = timezone.now()
     keep = {}
     refreshed = failed = 0

feeds/tests.py

@@ -539,3 +539,56 @@ class NewsTextApiTest(TestCase):
         self.assertEqual(got['related'], [])
         single = self.client.get(f'/feeds/stories/{self.items[2].story_id}/overview/').json()
         self.assertNotIn('summary', single)
+
+
+class SyntheticSourceTest(TestCase):
+    """A source diora fills itself ("Podcast-Artikel") is never fetched or pruned."""
+
+    def setUp(self):
+        self.u = User.objects.create_user('tester', password='pw12345678')
+        self.client.force_login(self.u)
+        self.synth = Source.objects.create(user=self.u, url='diora://podcast-artikel', title='Podcast-Artikel',
+                                           kind='blog', fav=True, synthetic=True)
+        self.real = Source.objects.create(user=self.u, url='https://real.example/feed', title='Real', kind='blog')
+
+    def post(self, url, data):
+        return self.client.post(url, json.dumps(data), content_type='application/json')
+
+    def test_refresh_skips_it(self):
+        with mock.patch('feeds.refresh._fetch_feed', return_value=fetched(rss([]), 'https://real.example/feed')) as get:
+            refresh.refresh([self.synth, self.real])
+        self.assertEqual([c.args[0] for c in get.call_args_list], ['https://real.example/feed'])
+
+    def test_refresh_now_and_command_skip_it(self):
+        with mock.patch('feeds.refresh._fetch_feed', return_value=fetched(rss([]), 'https://real.example/feed')) as get:
+            r = self.post('/feeds/refresh/', {})
+            from django.core.management import call_command
+            call_command('refresh_newsfeeds', stdout=mock.MagicMock())
+        self.assertEqual(r.status_code, 200)
+        self.assertTrue(get.call_args_list)
+        self.assertTrue(all(c.args[0] == 'https://real.example/feed' for c in get.call_args_list))
+        call_command('refresh_newsfeeds', '--source-id', str(self.synth.id), stdout=mock.MagicMock())
+
+    def test_old_items_are_not_pruned(self):
+        old = timezone.now() - timedelta(days=400)
+        for source in (self.synth, self.real):
+            for n in range(refresh.BLOG_KEEP + 5):
+                Item.objects.create(source=source, user=self.u, kind='blog', guid_hash=f'{source.pk}-{n}',
+                                    title='x', published=old - timedelta(minutes=n), read_at=old)
+        refresh.expire(self.u)
+        self.assertEqual(self.synth.items.count(), refresh.BLOG_KEEP + 5)
+        self.assertEqual(self.real.items.count(), refresh.BLOG_KEEP)
+
+    def test_cannot_become_news_but_can_be_hidden(self):
+        r = self.post(f'/feeds/sources/{self.synth.id}/', {'kind': 'news'})
+        self.assertEqual(r.status_code, 400)
+        r = self.post(f'/feeds/sources/{self.synth.id}/', {'hidden': True, 'notify': True})
+        self.assertEqual(r.status_code, 200)
+        self.synth.refresh_from_db()
+        self.assertEqual((self.synth.kind, self.synth.hidden, self.synth.notify), ('blog', True, True))
+        self.assertTrue(r.json()['source']['synthetic'])
+
+    def test_not_in_opml(self):
+        body = self.client.get('/feeds/export.opml').content.decode()
+        self.assertIn('real.example', body)
+        self.assertNotIn('diora://', body)

feeds/views.py

@@ -93,7 +93,7 @@ def _source_data(s, unread=0, new=0, interval=None):
         'id': s.id, 'url': s.url, 'title': s.title, 'site_url': s.site_url, 'icon_url': s.icon_url,
         'kind': s.kind, 'category': s.category, 'folders': s.folders, 'fulltext': s.fulltext,
         'unread': unread, 'new': new,
-        'fav': s.fav, 'hidden': s.hidden, 'notify': s.notify,
+        'fav': s.fav, 'hidden': s.hidden, 'notify': s.notify, 'synthetic': s.synthetic,
         # Days between posts (blogs, for "Seltene zuerst"); a blog with one post counts as rare.
         'interval_days': interval,
         'last_refreshed_at': _iso(s.last_refreshed_at), 'last_error': s.last_error, 'error_count': s.error_count,
@@ -108,6 +108,11 @@ def _item_data(i, content=False):
         'summary': i.summary, 'image_url': i.image_url, 'read': i.read_at is not None, 'progress': i.progress,
         'starred': i.starred, 'story': i.story_id,
     }
+    if i.source.synthetic:
+        # The article of a podcast episode (docs/transcription.md); the lists fetch it with the item.
+        art = getattr(i, 'episode_article', None)
+        if art is not None:
+            d['episode_id'] = art.episode_id
     if content:
         d['content'] = i.content
         d['content_full'] = i.content_full
@@ -356,7 +361,7 @@ def add_sources(request):
             defaults=_place(request.user, kind, x.get('category'), x.get('folders'),
                             title=(x.get('title') or url)[:300]))
         added += created
-    sources = Source.objects.filter(user=request.user, last_refreshed_at=None)
+    sources = Source.objects.filter(user=request.user, last_refreshed_at=None, synthetic=False)
     refresh.refresh(sources, deadline=time.monotonic() + REFRESH_WAIT)
     return JsonResponse({'added': added})
 
@@ -386,6 +391,8 @@ def edit_source(request, pk):
     if body.get('fulltext') in dict(Source.FULLTEXTS):
         source.fulltext = body['fulltext']
     new_kind = _kind(body.get('kind'))
+    if source.synthetic and new_kind == Source.NEWS:
+        return JsonResponse({'error': 'Diese Quelle bleibt ein Blog'}, status=400)
     if new_kind and new_kind != source.kind:
         # A section becomes a folder and back.
         if new_kind == Source.BLOG and source.category and not source.folders:
@@ -416,7 +423,7 @@ def export_opml(request):
     if (err := _auth(request)):
         return err
     groups = {}
-    for s in Source.objects.filter(user=request.user).order_by('kind', 'category', 'title'):
+    for s in Source.objects.filter(user=request.user, synthetic=False).order_by('kind', 'category', 'title'):
         kind = 'Nachrichten' if s.kind == Source.NEWS else 'Blogs'
         place = s.category if s.kind == Source.NEWS else ((s.folders or [''])[0])
         label = kind + (f' – {place}' if place else '')
@@ -440,7 +447,7 @@ def export_opml(request):
 def refresh_now(request):
     if (err := _auth(request)):
         return err
-    refreshed, failed, new = refresh.refresh(Source.objects.filter(user=request.user),
+    refreshed, failed, new = refresh.refresh(Source.objects.filter(user=request.user, synthetic=False),
                                              deadline=time.monotonic() + REFRESH_WAIT)
     return JsonResponse({'refreshed': refreshed, 'failed': failed, 'new': new})
 
@@ -611,7 +618,7 @@ def story_overview(request, pk):
 def blogs(request):
     if (err := _auth(request)):
         return err
-    qs = Item.objects.filter(user=request.user, kind=Source.BLOG, gone=False).select_related('source')
+    qs = Item.objects.filter(user=request.user, kind=Source.BLOG, gone=False).select_related('source', 'episode_article')
     if request.GET.get('source', '').isdigit():
         qs = qs.filter(source_id=int(request.GET['source']))
     elif 'folder' in request.GET:
@@ -640,7 +647,7 @@ def blogs(request):
 def starred(request):
     if (err := _auth(request)):
         return err
-    items = Item.objects.filter(user=request.user, starred=True).select_related('source').order_by('-published')
+    items = Item.objects.filter(user=request.user, starred=True).select_related('source', 'episode_article').order_by('-published')
     return JsonResponse({'items': [_item_data(i) for i in items[:200]]})
 
 
@@ -649,12 +656,13 @@ def starred(request):
 def item(request, pk):
     if (err := _auth(request)):
         return err
-    i = Item.objects.filter(user=request.user, pk=pk, gone=False).select_related('source').first()
+    i = Item.objects.filter(user=request.user, pk=pk, gone=False).select_related('source', 'episode_article').first()
     if i is None:
         return JsonResponse({'error': 'not found'}, status=404)
     if refresh.needs_article(i, i.source):
         refresh.load_article(i)
     d = _item_data(i, content=True)
+    if not i.source.synthetic:
         d['content'] = textproc.html_without_furniture(i.content, _furniture(i.source))
     return JsonResponse({'item': d})
 
@@ -744,17 +752,32 @@ def star(request, pk):
 def notifications(request):
     """What a device should announce: posts marked when they came in, still
     unread, of blogs that still have the bell on, after the id the device saw
-    last. Without `after` (a device's first time) nothing, only where to start."""
+    last. Without `after` (a device's first time) nothing, only where to start.
+    The same for finished transcripts (`t_after`, `transcripts`, `t_last`; jobs the
+    user's episodes have, docs/transcription.md)."""
     if (err := _auth(request)):
         return err
     qs = Item.objects.filter(user=request.user, notify=True)
     last = qs.order_by('-id').values_list('id', flat=True).first() or 0
+    # Jobs run in id order, one at a time, so a job never finishes below an id already seen.
+    # (A job that is done with no data only reflects a feed transcript: no notice for it.)
+    from podcasts.models import TranscriptionJob
+    jobs = TranscriptionJob.objects.filter(
+        episode__feed__user=request.user, state=TranscriptionJob.DONE, data__isnull=False)
+    t_last = jobs.order_by('-id').values_list('id', flat=True).first() or 0
+    out = {'items': [], 'last': last, 'transcripts': [], 't_last': t_last}
+    t_after = request.GET.get('t_after', '')
+    if t_after.isdigit():
+        out['transcripts'] = [{
+            'job_id': j.id, 'episode_id': j.episode_id, 'title': j.episode.title,
+            'podcast': j.episode.feed.title, 'want_article': j.want_article,
+        } for j in jobs.filter(id__gt=int(t_after)).select_related('episode__feed').order_by('id')[:50]]
     after = request.GET.get('after', '')
-    if not after.isdigit():
+    if after.isdigit():
-        return JsonResponse({'items': [], 'last': last})
         items = (qs.filter(id__gt=int(after), read_at=None, gone=False, source__notify=True)
                  .select_related('source').order_by('-published')[:50])
-    return JsonResponse({'items': [_item_data(i) for i in items], 'last': last})
+        out['items'] = [_item_data(i) for i in items]
+    return JsonResponse(out)
 
 
 @csrf_exempt
@@ -770,6 +793,8 @@ def save_login(request):
     password = body.get('password') or ''
     if source is None or not username or not password:
         return JsonResponse({'error': 'Benutzername und Passwort nötig'}, status=400)
+    if source.synthetic:
+        return JsonResponse({'error': 'Diese Quelle hat keine Adresse'}, status=400)
     try:
         adapter = logins.check_and_store(request.user, source.url, username, password)
     except logins.LoginFailed as e:

podcasts/article.py

@@ -0,0 +1,119 @@
+"""The article written from an episode's transcript (docs/transcription.md): its text
+is a feeds.Item in the user's favourite blog "Podcast-Artikel"."""
+
+import hashlib
+import json
+
+from django.db import transaction
+from django.http import JsonResponse
+from django.utils import timezone
+from django.views.decorators.csrf import csrf_exempt
+from django.views.decorators.http import require_http_methods
+
+from feeds import fetch
+from feeds.models import Item, Source
+
+from .models import EpisodeArticle, PodcastEpisode, TranscriptionJob
+
+SOURCE_URL = 'diora://podcast-artikel'
+SOURCE_TITLE = 'Podcast-Artikel'
+TITLE_MAX = 300
+SUMMARY_MAX = 2000
+BODY_MAX = 1_000_000       # characters of body_html
+MODEL_MAX = 100
+KIND_MAX = 20
+
+
+def _auth(request):
+    if not request.user.is_authenticated:
+        return JsonResponse({'error': 'authentication required'}, status=401)
+    return None
+
+
+def _body(request):
+    try:
+        body = json.loads(request.body or b'{}')
+    except (ValueError, UnicodeDecodeError):
+        return None
+    return body if isinstance(body, dict) else None
+
+
+def article_source(user):
+    """The user's blog "Podcast-Artikel", made on first use."""
+    source, _ = Source.objects.get_or_create(user=user, url=SOURCE_URL, defaults={
+        'title': SOURCE_TITLE, 'kind': Source.BLOG, 'fav': True, 'synthetic': True})
+    return source
+
+
+def _text(value, limit):
+    return value.strip() if isinstance(value, str) and len(value.strip()) <= limit else None
+
+
+@csrf_exempt
+@require_http_methods(['POST', 'DELETE'])
+def episode_article(request, pk):
+    if (err := _auth(request)):
+        return err
+    ep = PodcastEpisode.objects.filter(pk=pk, feed__user=request.user).select_related('feed').first()
+    if ep is None:
+        return JsonResponse({'error': 'not found'}, status=404)
+    if request.method == 'DELETE':
+        art = EpisodeArticle.objects.filter(episode=ep).select_related('item').first()
+        if art is not None:
+            art.item.delete()   # the article goes with it (cascade)
+        return JsonResponse({'ok': True})
+
+    body = _body(request)
+    if body is None:
+        return JsonResponse({'error': 'invalid JSON'}, status=400)
+    title = _text(body.get('title'), TITLE_MAX)
+    if not title:
+        return JsonResponse({'error': 'title needed (at most %d characters)' % TITLE_MAX}, status=400)
+    html_in = body.get('body_html')
+    if not isinstance(html_in, str) or not html_in.strip():
+        return JsonResponse({'error': 'body_html needed'}, status=400)
+    if len(html_in) > BODY_MAX:
+        return JsonResponse({'error': 'body_html too large'}, status=400)
+    summary = body.get('summary') or ''
+    if not isinstance(summary, str) or len(summary) > SUMMARY_MAX:
+        return JsonResponse({'error': 'summary too long'}, status=400)
+    model = body.get('model') or ''
+    kind = body.get('kind') or ''
+    if not isinstance(model, str) or len(model) > MODEL_MAX or not isinstance(kind, str) or len(kind) > KIND_MAX:
+        return JsonResponse({'error': 'model or kind too long'}, status=400)
+
+    url = getattr(ep, 'link', '') or ep.audio_url
+    content = fetch.clean_html(html_in, url)
+    if not fetch.plain(content):
+        return JsonResponse({'error': 'body_html empty'}, status=400)
+    fields = {
+        'url': url[:2000],
+        'title': title,
+        'author': (ep.feed.author or '')[:300],
+        'published': timezone.now(),
+        'summary': fetch.plain(summary),
+        'content': content,
+        'content_full': True,     # nothing to fetch from a page
+        'words': fetch.word_count(content),
+        'image_url': (ep.artwork_url or ep.feed.artwork_url or '')[:2000],
+        'read_at': None, 'progress': 0, 'progress_at': None,
+    }
+    with transaction.atomic():
+        source = article_source(request.user)
+        art = EpisodeArticle.objects.filter(episode=ep).select_related('item').first()
+        if art is not None:
+            item = art.item
+            for k, v in fields.items():
+                setattr(item, k, v)
+            item.source = source
+            item.save()
+            art.model, art.kind = model, kind
+            art.save(update_fields=['model', 'kind'])
+        else:
+            guid = hashlib.sha256(f'diora-article:{ep.pk}'.encode()).hexdigest()
+            item = Item.objects.create(source=source, user=request.user, kind=Source.BLOG,
+                                       guid_hash=guid, **fields)
+            EpisodeArticle.objects.create(episode=ep, item=item, model=model, kind=kind)
+        # Saving fulfils the wish and frees the reservation.
+        TranscriptionJob.objects.filter(episode=ep).update(want_article=False, claimed_by='', claimed_until=None)
+    return JsonResponse({'item_id': item.id})

podcasts/chapters.py

@@ -128,25 +128,28 @@ FetchUnavailable = ChaptersUnavailable
 FetchInvalid = ChaptersInvalid
 
 
-def _open(url, headers):
+def _open(url, headers, method='GET', timeout=CHAPTERS_TIMEOUT, **kwargs):
-    """GET `url` with every redirect hop checked; the streamed response of
+    """`method` (GET or POST) `url` with every redirect hop checked; the streamed response of
-    the final hop, which the caller has to close.
+    the final hop, which the caller has to close. `kwargs` (data, files) go to
+    requests; a POST only follows 307/308, which keep the method and body.
 
     Raises ChaptersInvalid for permanent problems and ChaptersUnavailable for
     ones that may go away (network trouble, server errors).
     """
+    redirects = (301, 302, 303, 307, 308) if method == 'GET' else (307, 308)
     for _ in range(CHAPTERS_MAX_REDIRECTS + 1):
         try:
             assert_safe_url(url)
         except WebDAVInputError as e:
             raise ChaptersInvalid(str(e))
         try:
-            r = requests.get(url, timeout=CHAPTERS_TIMEOUT, allow_redirects=False, stream=True, headers=headers)
+            send = requests.get if method == 'GET' else requests.post
+            r = send(url, timeout=timeout, allow_redirects=False, stream=True, headers=headers, **kwargs)
         except requests.RequestException as e:
             raise ChaptersUnavailable(e.__class__.__name__)
         try:
             _assert_peer_is_safe(r)
-            if r.status_code in (301, 302, 303, 307, 308):
+            if r.status_code in redirects:
                 location = r.headers.get('Location')
                 if not location:
                     raise ChaptersInvalid('redirect without a location')
@@ -155,7 +158,7 @@ def _open(url, headers):
                 continue
             if r.status_code >= 500 or r.status_code == 429:
                 raise ChaptersUnavailable(f'HTTP {r.status_code}')
-            if r.status_code >= 400:
+            if r.status_code >= 400 or r.status_code in (301, 302, 303):
                 raise ChaptersInvalid(f'HTTP {r.status_code}')
         except WebDAVInputError as e:
             r.close()

podcasts/migrations/0027_podcastepisode_link.py

@@ -0,0 +1,18 @@
+# Generated by Django 6.1 on 2026-11-09 08:19
+
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('podcasts', '0026_podcastepisode_sponsor'),
+    ]
+
+    operations = [
+        migrations.AddField(
+            model_name='podcastepisode',
+            name='link',
+            field=models.URLField(blank=True, max_length=1000),
+        ),
+    ]

podcasts/migrations/0028_transcription.py

@@ -0,0 +1,59 @@
+# Generated by Django 6.1 on 2026-11-09 08:20
+
+import django.db.models.deletion
+from django.conf import settings
+from django.db import migrations, models
+
+
+class Migration(migrations.Migration):
+
+    dependencies = [
+        ('feeds', '0007_transcription'),
+        ('podcasts', '0027_podcastepisode_link'),
+        migrations.swappable_dependency(settings.AUTH_USER_MODEL),
+    ]
+
+    operations = [
+        migrations.CreateModel(
+            name='EpisodeArticle',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('model', models.CharField(blank=True, max_length=100)),
+                ('kind', models.CharField(blank=True, max_length=20)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('episode', models.OneToOneField(on_delete=django.db.models.deletion.CASCADE, related_name='article', to='podcasts.podcastepisode')),
+                ('item', models.OneToOneField(on_delete=django.db.models.deletion.CASCADE, related_name='episode_article', to='feeds.item')),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TranscriptionJob',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('state', models.CharField(choices=[('none', 'none'), ('queued', 'queued'), ('downloading', 'downloading'), ('transcribing', 'transcribing'), ('done', 'done'), ('error', 'error')], default='none', max_length=15)),
+                ('chunks_done', models.PositiveIntegerField(default=0)),
+                ('chunks_total', models.PositiveIntegerField(default=0)),
+                ('error', models.CharField(blank=True, max_length=500)),
+                ('want_article', models.BooleanField(default=False)),
+                ('claimed_by', models.CharField(blank=True, max_length=60)),
+                ('claimed_until', models.DateTimeField(blank=True, null=True)),
+                ('data', models.JSONField(blank=True, null=True)),
+                ('created_at', models.DateTimeField(auto_now_add=True)),
+                ('started_at', models.DateTimeField(blank=True, null=True)),
+                ('finished_at', models.DateTimeField(blank=True, null=True)),
+                ('episode', models.OneToOneField(on_delete=django.db.models.deletion.CASCADE, related_name='transcription_job', to='podcasts.podcastepisode')),
+            ],
+        ),
+        migrations.CreateModel(
+            name='TranscriptionSettings',
+            fields=[
+                ('id', models.BigAutoField(auto_created=True, primary_key=True, serialize=False, verbose_name='ID')),
+                ('engine', models.CharField(choices=[('builtin', 'Eingebaut'), ('custom', 'Eigene Instanz')], default='builtin', max_length=10)),
+                ('url', models.CharField(blank=True, max_length=500)),
+                ('token_encrypted', models.TextField(blank=True)),
+                ('model', models.CharField(blank=True, max_length=100)),
+                ('language', models.CharField(blank=True, max_length=10)),
+                ('parallel', models.PositiveSmallIntegerField(default=3)),
+                ('user', models.OneToOneField(on_delete=django.db.models.deletion.CASCADE, related_name='transcription_settings', to=settings.AUTH_USER_MODEL)),
+            ],
+        ),
+    ]

podcasts/models.py

@@ -164,6 +164,8 @@ class PodcastEpisode(models.Model):
     title = models.CharField(max_length=500)
     description = models.TextField(blank=True)
     audio_url = models.URLField(max_length=1000)
+    # The episode's page, from the feed item's <link> (for sharing); '' when it has none.
+    link = models.URLField(max_length=1000, blank=True)
     duration_seconds = models.IntegerField(default=0)
     pub_date = models.DateTimeField(null=True, blank=True)
     episode_number = models.IntegerField(null=True, blank=True)
@@ -217,6 +219,68 @@ class EpisodeTranscript(models.Model):
         return f'Transcript of {self.episode}'
 
 
+class TranscriptionSettings(models.Model):
+    """Where a user's episodes are transcribed (docs/transcription.md): the
+    built-in Whisper instance (DIORA_WHISPER_URL) or their own."""
+    BUILTIN = 'builtin'
+    CUSTOM = 'custom'
+    ENGINES = [(BUILTIN, 'Eingebaut'), (CUSTOM, 'Eigene Instanz')]
+
+    user = models.OneToOneField(User, on_delete=models.CASCADE, related_name='transcription_settings')
+    engine = models.CharField(max_length=10, choices=ENGINES, default=BUILTIN)
+    # The instance's base, e.g. https://whisper.example.org/v1 (custom only).
+    url = models.CharField(max_length=500, blank=True)
+    # The API token, Fernet-encrypted like feeds.SiteLogin.secret; never handed out.
+    token_encrypted = models.TextField(blank=True)
+    model = models.CharField(max_length=100, blank=True)
+    # '' = automatic, else an ISO code ("de").
+    language = models.CharField(max_length=10, blank=True)
+    parallel = models.PositiveSmallIntegerField(default=3)   # 1 to 8 pieces at once
+
+    def __str__(self):
+        return f'Transcription settings of {self.user}'
+
+
+class TranscriptionJob(models.Model):
+    """One episode's transcription run, its Whisper result and the wish for an
+    article. Kept apart from EpisodeTranscript, which caches the feed's
+    transcripts and is rewritten when the feed changes them."""
+    NONE, QUEUED, DOWNLOADING, TRANSCRIBING, DONE, ERROR = (
+        'none', 'queued', 'downloading', 'transcribing', 'done', 'error')
+    STATES = [(s, s) for s in (NONE, QUEUED, DOWNLOADING, TRANSCRIBING, DONE, ERROR)]
+
+    episode = models.OneToOneField(PodcastEpisode, on_delete=models.CASCADE, related_name='transcription_job')
+    state = models.CharField(max_length=15, choices=STATES, default=NONE)
+    chunks_done = models.PositiveIntegerField(default=0)
+    chunks_total = models.PositiveIntegerField(default=0)
+    error = models.CharField(max_length=500, blank=True)
+    want_article = models.BooleanField(default=False)
+    # The device writing the article (a short name) and until when it holds the wish.
+    claimed_by = models.CharField(max_length=60, blank=True)
+    claimed_until = models.DateTimeField(null=True, blank=True)
+    # {"segments": [{start, end, speaker, text}], "source": {type: "whisper", model, language}}
+    data = models.JSONField(null=True, blank=True)
+    created_at = models.DateTimeField(auto_now_add=True)
+    started_at = models.DateTimeField(null=True, blank=True)
+    finished_at = models.DateTimeField(null=True, blank=True)
+
+    def __str__(self):
+        return f'Transcription of {self.episode} ({self.state})'
+
+
+class EpisodeArticle(models.Model):
+    """The article written from an episode's transcript; its text is the
+    feeds.Item (in the blog "Podcast-Artikel")."""
+    episode = models.OneToOneField(PodcastEpisode, on_delete=models.CASCADE, related_name='article')
+    item = models.OneToOneField('feeds.Item', on_delete=models.CASCADE, related_name='episode_article')
+    model = models.CharField(max_length=100, blank=True)
+    kind = models.CharField(max_length=20, blank=True)   # Interview, Gespräch, Monolog, Diskussion
+    created_at = models.DateTimeField(auto_now_add=True)
+
+    def __str__(self):
+        return f'Article on {self.episode}'
+
+
 # An episode counts as played this close to its end (AntennaPod's default),
 # the same rule as the apps' "als gehört zählen".
 PLAYED_BEFORE_END_SECONDS = 30

podcasts/shownotes.py

@@ -37,6 +37,9 @@ ALLOWED_ATTRS = {
 URL_ATTRS = {'href', 'src'}
 LINK_SCHEMES = {'http', 'https', 'mailto'}
 IMAGE_SCHEMES = {'http', 'https'}
+# A link to a second in the episode ("#t=754"), as in the article of a transcript
+# (docs/transcription.md): the only fragment link that is kept.
+TIME_LINK = re.compile(r'^#t=\d{1,6}$')
 
 _CONTROL_AND_SPACE = re.compile(r'[\x00-\x20\x7f]+')
 _SCHEME = re.compile(r'^([a-z][a-z0-9+.\-]*):')
@@ -81,7 +84,9 @@ class _Sanitizer(HTMLParser):
             name = name.lower()
             if name not in allowed or value is None:
                 continue
-            if name in URL_ATTRS:
+            if tag == 'a' and name == 'href' and TIME_LINK.match(value.strip()):
+                value = value.strip()
+            elif name in URL_ATTRS:
                 value = _safe_url(value, IMAGE_SCHEMES if tag == 'img' else LINK_SCHEMES)
                 if value is None:
                     continue

podcasts/test_article.py

@@ -0,0 +1,146 @@
+import json
+
+from django.contrib.auth.models import User
+from django.test import TestCase
+from django.utils import timezone
+
+from feeds.models import Item, Source
+
+from .models import EpisodeArticle, PodcastEpisode, PodcastFeed, TranscriptionJob
+
+BODY = ('<p>Zusammenfassung <script>alert(1)</script></p><h2>Teil eins</h2>'
+        '<p><a href="#t=754" onclick="evil()">▶ ab 12:34</a> und <a href="https://example.org/x">Quelle</a></p>')
+
+
+class ArticleTests(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username='alice', password='pw12345678')
+        self.client.force_login(self.user)
+        self.feed = PodcastFeed.objects.create(user=self.user, rss_url='https://p.example/rss', title='Salon',
+                                               author='Anna Autorin', artwork_url='https://p.example/feed.jpg')
+        self.ep = PodcastEpisode.objects.create(feed=self.feed, guid='e1', title='Folge 1',
+                                                audio_url='https://p.example/e1.mp3', artwork_url='https://p.example/e1.jpg')
+
+    def url(self, ep=None):
+        return f'/podcasts/episodes/{(ep or self.ep).id}/article/'
+
+    def post(self, data=None, ep=None):
+        data = {'title': 'Ein Artikel', 'summary': 'Kurz.', 'body_html': BODY, 'model': 'claude', 'kind': 'Interview'} \
+            if data is None else data
+        return self.client.post(self.url(ep), json.dumps(data), content_type='application/json')
+
+    def test_requires_login_and_ownership(self):
+        self.client.logout()
+        self.assertEqual(self.post().status_code, 401)
+        self.assertEqual(self.client.delete(self.url()).status_code, 401)
+        bob = User.objects.create_user(username='bob', password='pw12345678')
+        self.client.force_login(bob)
+        self.assertEqual(self.post().status_code, 404)
+        self.assertEqual(self.client.delete(self.url()).status_code, 404)
+        self.assertFalse(Item.objects.exists())
+
+    def test_create(self):
+        r = self.post()
+        self.assertEqual(r.status_code, 200)
+        item = Item.objects.get(pk=r.json()['item_id'])
+        self.assertEqual(item.user, self.user)
+        self.assertEqual(item.kind, Source.BLOG)
+        self.assertEqual(item.title, 'Ein Artikel')
+        self.assertEqual(item.summary, 'Kurz.')
+        self.assertEqual(item.author, 'Anna Autorin')
+        self.assertEqual(item.image_url, 'https://p.example/e1.jpg')
+        self.assertEqual(item.url, 'https://p.example/e1.mp3')
+        self.assertTrue(item.content_full)
+        self.assertGreater(item.words, 3)
+        self.assertLess((timezone.now() - item.published).total_seconds(), 60)
+        art = EpisodeArticle.objects.get(episode=self.ep)
+        self.assertEqual((art.item, art.model, art.kind), (item, 'claude', 'Interview'))
+        self.assertEqual(item.episode_article, art)
+
+    def test_image_falls_back_to_feed_artwork(self):
+        self.ep.artwork_url = ''
+        self.ep.save()
+        item = Item.objects.get(pk=self.post().json()['item_id'])
+        self.assertEqual(item.image_url, 'https://p.example/feed.jpg')
+
+    def test_sanitizer_keeps_time_links_only(self):
+        item = Item.objects.get(pk=self.post().json()['item_id'])
+        self.assertNotIn('<script', item.content)
+        self.assertNotIn('alert(1)', item.content)
+        self.assertNotIn('onclick', item.content)
+        self.assertIn('href="#t=754"', item.content)
+        self.assertIn('href="https://example.org/x"', item.content)
+        evil = self.post({'title': 'x', 'body_html': '<a href="#t=1&quot; onfocus=&quot;x">a</a><a href="#top">b</a>'})
+        content = Item.objects.get(pk=evil.json()['item_id']).content
+        self.assertNotIn('onfocus', content)
+        self.assertNotIn('#top', content)
+
+    def test_source_made_once(self):
+        other = PodcastEpisode.objects.create(feed=self.feed, guid='e2', title='Folge 2', audio_url='https://p.example/e2.mp3')
+        self.post()
+        self.post(ep=other)
+        src = Source.objects.get(user=self.user, url='diora://podcast-artikel')
+        self.assertEqual((src.title, src.kind, src.fav, src.synthetic), ('Podcast-Artikel', Source.BLOG, True, True))
+        self.assertEqual(Source.objects.count(), 1)
+        self.assertEqual(src.items.count(), 2)
+
+    def test_replace_keeps_one_item(self):
+        first = self.post().json()['item_id']
+        Item.objects.filter(pk=first).update(read_at=timezone.now(), progress=1)
+        second = self.post({'title': 'Neu', 'body_html': '<p>Ganz anderer Text</p>', 'model': 'm2'}).json()['item_id']
+        self.assertEqual(first, second)
+        self.assertEqual(Item.objects.count(), 1)
+        item = Item.objects.get()
+        self.assertEqual((item.title, item.read_at, item.progress), ('Neu', None, 0))
+        self.assertIn('Ganz anderer Text', item.content)
+        self.assertEqual(EpisodeArticle.objects.get().model, 'm2')
+
+    def test_delete(self):
+        self.post()
+        self.assertEqual(self.client.delete(self.url()).status_code, 200)
+        self.assertFalse(Item.objects.exists())
+        self.assertFalse(EpisodeArticle.objects.exists())
+        self.assertEqual(Source.objects.count(), 1)
+        self.assertEqual(self.client.delete(self.url()).status_code, 200)   # again: nothing to do
+
+    def test_validation(self):
+        for data in ({}, {'title': '', 'body_html': BODY}, {'title': 'x' * 301, 'body_html': BODY},
+                     {'title': 'x'}, {'title': 'x', 'body_html': '<script>1</script>'},
+                     {'title': 'x', 'body_html': 'a' * 1_000_001},
+                     {'title': 'x', 'body_html': BODY, 'model': 'm' * 101}):
+            self.assertEqual(self.post(data).status_code, 400, data and list(data))
+        self.assertEqual(self.client.post(self.url(), 'nope', content_type='application/json').status_code, 400)
+        self.assertFalse(Item.objects.exists())
+        self.assertEqual(self.client.get(self.url()).status_code, 405)
+
+    def test_saving_clears_the_wish(self):
+        TranscriptionJob.objects.create(episode=self.ep, state='done', want_article=True, claimed_by='phone',
+                                        claimed_until=timezone.now() + timezone.timedelta(minutes=10))
+        self.post()
+        job = TranscriptionJob.objects.get()
+        self.assertEqual((job.want_article, job.claimed_by, job.claimed_until), (False, '', None))
+
+    def test_in_favourite_blogs_with_episode_id(self):
+        item_id = self.post().json()['item_id']
+        items = self.client.get('/feeds/blogs/?fav=1').json()['items']
+        self.assertEqual([i['id'] for i in items], [item_id])
+        self.assertEqual(items[0]['episode_id'], self.ep.id)
+        detail = self.client.get(f'/feeds/items/{item_id}/').json()['item']
+        self.assertEqual(detail['episode_id'], self.ep.id)
+        self.assertIn('href="#t=754"', detail['content'])
+
+    def test_ordinary_items_have_no_episode_id(self):
+        src = Source.objects.create(user=self.user, url='https://b.example/feed', title='B', kind=Source.BLOG, fav=True)
+        Item.objects.create(source=src, user=self.user, kind=Source.BLOG, guid_hash='g', title='T',
+                            published=timezone.now(), content='<p>x</p>', content_full=True)
+        self.post()
+        items = self.client.get('/feeds/blogs/?fav=1').json()['items']
+        self.assertEqual(len(items), 2)
+        self.assertEqual(sorted('episode_id' in i for i in items), [False, True])
+
+    def test_blog_list_needs_no_query_per_article(self):
+        for n in range(5):
+            self.post(ep=PodcastEpisode.objects.create(feed=self.feed, guid=f'x{n}', title=str(n),
+                                                       audio_url='https://p.example/x.mp3'))
+        with self.assertNumQueries(3):
+            self.client.get('/feeds/blogs/?fav=1')

podcasts/test_player.py

@@ -89,7 +89,7 @@ class PlayerSyncTests(TestCase):
         self.assertEqual(d['seq'], 2)   # the report, and device a coming online
         self.assertEqual((d['kind'], d['playing'], d['position_ms'], d['speed']), ('podcast', True, 12345, 1.5))
         self.assertEqual((d['device_id'], d['device_name']), ('a', 'A'))
-        self.assertEqual(set(d['episode']), {'id', 'title', 'audio_url', 'duration_seconds', 'pub_date', 'artwork_url',
+        self.assertEqual(set(d['episode']), {'id', 'title', 'audio_url', 'link', 'duration_seconds', 'pub_date', 'artwork_url',
                                              'feed__id', 'feed__title', 'feed__artwork_url', 'position_seconds',
                                              'in_queue'})
         self.assertEqual((d['episode']['id'], d['episode']['in_queue']), (self.ep.id, True))

podcasts/test_share.py

@@ -0,0 +1,45 @@
+import feedparser
+from django.contrib.auth.models import User
+from django.test import TestCase
+
+from .models import PodcastEpisode, PodcastFeed
+from .views import FeedFetch, _apply_feed, _entry_link
+
+
+def _feed(link):
+    item_link = f'<link>{link}</link>' if link is not None else ''
+    return feedparser.parse(
+        '<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>99 ZU EINS</title>'
+        '<item><title>Folge 724</title><guid>g724</guid>' + item_link +
+        '<enclosure url="https://cdn.example/724.mp3" type="audio/mpeg" length="1"/></item>'
+        '</channel></rss>'
+    )
+
+
+class EpisodeLinkTests(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username='alice', password='pw12345678')
+        self.client.force_login(self.user)
+        self.feed = PodcastFeed.objects.create(user=self.user, rss_url='https://example.org/rss', title='99 ZU EINS')
+
+    def fetch(self, link):
+        _apply_feed(self.feed, FeedFetch(200, _feed(link)))
+        return PodcastEpisode.objects.get(feed=self.feed)
+
+    def test_entry_link_takes_only_web_addresses(self):
+        self.assertEqual(_entry_link({'link': ' https://x.fireside.fm/724 '}), 'https://x.fireside.fm/724')
+        self.assertEqual(_entry_link({'link': 'javascript:alert(1)'}), '')
+        self.assertEqual(_entry_link({}), '')
+
+    def test_a_new_episode_gets_the_items_link(self):
+        self.assertEqual(self.fetch('https://99zueins.fireside.fm/724').link, 'https://99zueins.fireside.fm/724')
+
+    def test_a_refresh_fills_the_link_of_an_episode_that_has_none(self):
+        self.assertEqual(self.fetch(None).link, '')
+        self.assertEqual(self.fetch('https://99zueins.fireside.fm/724').link, 'https://99zueins.fireside.fm/724')
+
+    def test_episode_detail_returns_the_link(self):
+        ep = self.fetch('https://99zueins.fireside.fm/724')
+        r = self.client.get(f'/podcasts/episodes/{ep.id}/')
+        self.assertEqual(r.status_code, 200)
+        self.assertEqual(r.json()['link'], 'https://99zueins.fireside.fm/724')

podcasts/test_transcribe.py

@@ -0,0 +1,821 @@
+"""The transcription pipeline (podcasts/transcribe.py) against a fake Whisper instance, the
+endpoints around it and the notice in /feeds/notifications/."""
+import email
+import io
+import json
+import math
+import os
+import struct
+import subprocess
+import tempfile
+import threading
+import time
+import wave
+from datetime import timedelta
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from unittest.mock import patch
+
+from django.contrib.auth.models import User
+from django.db import OperationalError
+from django.test import TestCase, TransactionTestCase, override_settings
+from django.utils import timezone
+
+from feeds.models import Item, Source
+
+from . import transcribe, transcription, youtube
+from .models import (
+    EpisodeArticle, EpisodeTranscript, PodcastEpisode, PodcastFeed, TranscriptionJob, TranscriptionSettings,
+)
+
+RATE = 16000
+TOTAL = 120
+# Where the long pauses are (start, length): the pieces should be cut in these. The short
+# ones (0.3 s every 4 s) are only there to tempt.
+LONG_PAUSES = [(36.5, 1.6), (83.0, 1.4)]
+SEGMENTS = [{'start': 0.0, 'end': 2.0, 'text': ' Hallo '}, {'start': 2.0, 'end': 5.5, 'text': 'Welt.'}]
+
+
+def make_audio():
+    """~2 minutes of tones with pauses, as mp3 bytes (and the wav as a check of the length)."""
+    samples = []
+    t = 0.0
+    pauses = [(s, s + n) for s, n in LONG_PAUSES]
+    for i in range(RATE * TOTAL):
+        t = i / RATE
+        silent = any(a <= t < b for a, b in pauses) or (t % 4.0) > 3.7
+        value = 0 if silent else int(9000 * math.sin(2 * math.pi * (180 + 40 * ((i // 8000) % 5)) * t))
+        samples.append(value)
+    buf = io.BytesIO()
+    with wave.open(buf, 'wb') as w:
+        w.setnchannels(1)
+        w.setsampwidth(2)
+        w.setframerate(RATE)
+        w.writeframes(struct.pack('<%dh' % len(samples), *samples))
+    import imageio_ffmpeg
+    with tempfile.TemporaryDirectory() as tmp:
+        src, dst = os.path.join(tmp, 'a.wav'), os.path.join(tmp, 'a.mp3')
+        open(src, 'wb').write(buf.getvalue())
+        subprocess.run([imageio_ffmpeg.get_ffmpeg_exe(), '-loglevel', 'error', '-i', src, '-b:a', '64k', dst], check=True)
+        return open(dst, 'rb').read()
+
+
+_AUDIO = []
+
+
+def audio():
+    if not _AUDIO:
+        _AUDIO.append(make_audio())
+    return _AUDIO[0]
+
+
+def decoded_seconds(data, suffix):
+    """How long the uploaded piece is, by decoding it."""
+    import imageio_ffmpeg
+    with tempfile.TemporaryDirectory() as tmp:
+        path = os.path.join(tmp, 'p' + suffix)
+        open(path, 'wb').write(data)
+        out = subprocess.run([imageio_ffmpeg.get_ffmpeg_exe(), '-loglevel', 'error', '-i', path, '-f', 's16le',
+                              '-ac', '1', '-ar', str(RATE), '-'], stdout=subprocess.PIPE, check=True).stdout
+    return len(out) / 2 / RATE
+
+
+class Fake:
+    """A local HTTP server: GET /ep.mp3 is the test audio, POST /v1/audio/transcriptions a Whisper."""
+
+    def __init__(self, fail_first=0, fail_status=500, delay=0, on_call=None):
+        self.fail_first, self.fail_status, self.delay, self.on_call = fail_first, fail_status, delay, on_call
+        self.calls = []   # {'fields', 'filename', 'body', 'auth'}
+        self.lock = threading.Lock()
+        fake = self
+
+        class Handler(BaseHTTPRequestHandler):
+            def log_message(self, *a):
+                pass
+
+            def do_GET(self):
+                if self.path == '/missing.mp3':
+                    self.send_response(404)
+                    self.end_headers()
+                    return
+                body = audio_holder.data or audio()
+                self.send_response(200)
+                self.send_header('Content-Type', 'audio/mpeg')
+                self.send_header('Content-Length', str(len(body)))
+                self.end_headers()
+                self.wfile.write(body)
+
+            def do_POST(self):
+                raw = self.rfile.read(int(self.headers['Content-Length']))
+                msg = email.message_from_bytes(
+                    b'Content-Type: ' + self.headers['Content-Type'].encode() + b'\r\n\r\n' + raw)
+                fields, filename, body = {}, '', b''
+                for part in msg.get_payload():
+                    name = part.get_param('name', header='content-disposition')
+                    if part.get_filename():
+                        filename, body = part.get_filename(), part.get_payload(decode=True)
+                    else:
+                        fields[name] = part.get_payload(decode=True).decode()
+                with fake.lock:
+                    fake.calls.append({'fields': fields, 'filename': filename, 'body': body,
+                                       'auth': self.headers.get('Authorization')})
+                    n = len(fake.calls)
+                if fake.on_call:
+                    fake.on_call(n)
+                if fake.delay:
+                    time.sleep(fake.delay)
+                if n <= fake.fail_first or fake.fail_first < 0:
+                    self.send_response(fake.fail_status)
+                    self.send_header('Content-Length', '0')
+                    self.end_headers()
+                    return
+                out = json.dumps({'language': 'german', 'segments': SEGMENTS}).encode()
+                self.send_response(200)
+                self.send_header('Content-Type', 'application/json')
+                self.send_header('Content-Length', str(len(out)))
+                self.end_headers()
+                self.wfile.write(out)
+
+        self.server = ThreadingHTTPServer(('127.0.0.1', 0), Handler)
+        self.server.daemon_threads = True
+        self.port = self.server.server_address[1]
+        threading.Thread(target=self.server.serve_forever, daemon=True).start()
+
+    @property
+    def whisper(self):
+        return f'http://127.0.0.1:{self.port}/v1'
+
+    def url(self, path='/ep.mp3'):
+        return f'http://127.0.0.1:{self.port}{path}'
+
+    def stop(self):
+        self.server.shutdown()
+        self.server.server_close()
+
+
+def leftovers():
+    base = tempfile.gettempdir()
+    return {d for d in os.listdir(base) if d.startswith('diora-transcribe-')}
+
+
+FAST = dict(CHUNK_SECONDS=40, CUT_WINDOW=8, MIN_TAIL=10, RETRY_PAUSES=(0, 0, 0))
+
+
+def wait_for(cond, timeout=30):
+    end = time.time() + timeout
+    while time.time() < end:
+        if cond():
+            return True
+        time.sleep(0.05)
+    return False
+
+
+class PipelineBase(TransactionTestCase):
+    def setUp(self):
+        for k, v in FAST.items():
+            patcher = patch.object(transcribe, k, v)
+            patcher.start()
+            self.addCleanup(patcher.stop)
+        self.user = User.objects.create_user(username='alice', password='pw12345678')
+        self.feed = PodcastFeed.objects.create(user=self.user, rss_url='https://example.org/f', title='Der Podcast',
+                                               author='Anna Autorin')
+        self.fake = Fake()
+        self.addCleanup(self.fake.stop)
+        self.before = leftovers()
+
+    def job(self, audio_path='/ep.mp3', **kw):
+        n = PodcastEpisode.objects.count()
+        ep = PodcastEpisode.objects.create(feed=self.feed, guid=f'g{n}', title=f'Folge {n}',
+                                           audio_url=self.fake.url(audio_path), duration_seconds=TOTAL)
+        return TranscriptionJob.objects.create(episode=ep, state=TranscriptionJob.QUEUED, **kw)
+
+    def reload(self, job):
+        for _ in range(50):   # the in-memory test database may be locked for a moment by the worker
+            try:
+                return TranscriptionJob.objects.get(pk=job.pk)
+            except OperationalError:
+                time.sleep(0.05)
+        return TranscriptionJob.objects.get(pk=job.pk)
+
+    def assertClean(self):
+        self.assertEqual(leftovers(), self.before)
+
+
+class PlanTests(PipelineBase):
+    def test_cuts_land_in_the_long_pauses(self):
+        with tempfile.TemporaryDirectory() as tmp:
+            src, pcm = os.path.join(tmp, 'a.mp3'), os.path.join(tmp, 'a.pcm')
+            open(src, 'wb').write(audio())
+            transcribe._decode(src, pcm)
+            total = os.path.getsize(pcm) / 2 / RATE
+            pieces = transcribe.plan(total, transcribe.pauses_of(pcm))
+        self.assertAlmostEqual(total, TOTAL, delta=0.5)
+        self.assertEqual(len(pieces), 3)
+        cuts = [p[1] for p in pieces[:-1]]
+        for cut, (start, length) in zip(cuts, LONG_PAUSES):
+            self.assertTrue(start <= cut <= start + length, f'{cut} not in the pause {start}+{length}')
+        self.assertEqual(pieces[0][0], 0)
+        self.assertEqual(pieces[-1][1], total)
+        self.assertEqual([a[1] for a in pieces[:-1]], [b[0] for b in pieces[1:]])
+
+    def test_plan_without_pauses_cuts_at_the_target(self):
+        self.assertEqual(transcribe.plan(100, []), [(0.0, 40), (40, 80), (80, 100)])
+        self.assertEqual(transcribe.plan(30, []), [(0.0, 30)])
+        # a short tail is joined to the last piece
+        self.assertEqual(transcribe.plan(85, []), [(0.0, 40), (40, 85)])
+
+
+@override_settings(WHISPER_URL='')
+class RunTests(PipelineBase):
+    def configure(self, **kw):
+        conf = TranscriptionSettings(user=self.user, **kw)
+        conf.save()
+        return conf
+
+    def test_builtin_run_end_to_end(self):
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            self.configure(language='de', parallel=2)
+            job = self.job()
+            started = time.time()
+            transcribe.run(job.pk)
+            took = time.time() - started
+        print(f'\n[pipeline run of the {TOTAL} s test audio took {took:.1f} s]')
+        job = self.reload(job)
+        self.assertEqual((job.state, job.error), ('done', ''))
+        self.assertEqual((job.chunks_done, job.chunks_total), (3, 3))
+        self.assertIsNotNone(job.finished_at)
+        self.assertEqual(len(self.fake.calls), 3)
+        for c in self.fake.calls:
+            self.assertEqual(c['fields']['model'], 'whisper-1')
+            self.assertEqual(c['fields']['language'], 'de')
+            self.assertEqual(c['fields']['response_format'], 'verbose_json')
+            self.assertEqual(c['fields']['prompt'], 'Der Podcast. Folge 0. Anna Autorin')
+            self.assertTrue(c['filename'].endswith('.flac'))
+            self.assertIsNone(c['auth'])   # the token only goes to custom instances
+        # the uploaded pieces are the planned ones: the same total, cut at the pauses
+        lengths = sorted(decoded_seconds(c['body'], '.flac') for c in self.fake.calls)
+        self.assertAlmostEqual(sum(lengths), TOTAL, delta=0.5)
+        data = job.data
+        self.assertEqual(data['source'], {'type': 'whisper', 'model': 'whisper-1', 'language': 'de'})
+        segs = data['segments']
+        starts = [s['start'] for s in segs]
+        self.assertEqual(starts, sorted(starts))
+        self.assertTrue(all(s['speaker'] == '' for s in segs))
+        # every piece answered "Hallo Welt." at 0 to 5.5 s: its start is where the piece starts,
+        # which lies on a cut in a long pause, and the first at zero
+        self.assertEqual(starts[0], 0)
+        cuts = starts[1:]
+        self.assertEqual(len(cuts), 2)
+        for cut, (p, n) in zip(cuts, LONG_PAUSES):
+            self.assertTrue(p <= cut <= p + n, f'{cut} not in {p}+{n}')
+        self.assertEqual(segs[0]['text'], 'Hallo Welt.')
+        self.assertAlmostEqual(segs[0]['end'], 5.5)
+        self.assertClean()
+
+    def test_language_empty_takes_what_whisper_detects_and_custom_gets_opus_and_token(self):
+        conf = self.configure(engine='custom', url=self.fake.whisper, model='large-v3', language='')
+        conf.token_encrypted = transcription.logins.encrypt('tok')
+        conf.save()
+        job = self.job()
+        with patch.object(transcribe.chapters, 'assert_safe_url'), \
+                patch.object(transcribe.chapters, '_assert_peer_is_safe'):
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual((job.state, job.error), ('done', ''))
+        self.assertEqual(job.data['source'], {'type': 'whisper', 'model': 'large-v3', 'language': 'german'})
+        for c in self.fake.calls:
+            self.assertTrue(c['filename'].endswith('.ogg'))
+            self.assertNotIn('language', c['fields'])
+            self.assertEqual(c['auth'], 'Bearer tok')
+            self.assertLess(len(c['body']), decoded_seconds(c['body'], '.ogg') * 6500)   # 32 kbit/s: about 4 kB a second, ogg and VBR on top
+        self.assertAlmostEqual(sum(decoded_seconds(c['body'], '.ogg') for c in self.fake.calls), TOTAL, delta=0.5)
+        self.assertClean()
+
+    def test_custom_instance_on_a_private_address_is_refused(self):
+        self.configure(engine='custom', url=self.fake.whisper)
+        job = self.job()
+        transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual(job.state, 'error')
+        self.assertIn('Whisper', job.error)
+        self.assertEqual(self.fake.calls, [])
+
+    def test_unreachable_builtin_waits_and_tries_again(self):
+        import socket
+        sock = socket.socket()
+        sock.bind(('127.0.0.1', 0))
+        port = sock.getsockname()[1]
+        sock.close()   # nothing listens there now
+        with override_settings(WHISPER_URL=f'http://127.0.0.1:{port}/v1'):
+            self.configure()
+            job = self.job()
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual(job.state, 'queued')
+        self.assertIn('nicht erreichbar', job.error)
+        self.assertGreater(transcribe._not_before[job.pk], time.time())
+        self.assertEqual(self.fake.calls, [])   # nothing downloaded while the instance is away
+        self.assertEqual(transcribe._next_job(), transcribe.WAIT)
+        transcribe._not_before[job.pk] = 0   # due again: the instance is back
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual((job.state, job.error), ('done', ''))
+        transcribe._not_before.clear()
+
+    def test_waiting_gives_up_after_two_days(self):
+        from datetime import timedelta
+        from django.utils import timezone
+        with override_settings(WHISPER_URL='http://127.0.0.1:9/v1'):
+            self.configure()
+            job = self.job()
+            TranscriptionJob.objects.filter(pk=job.pk).update(created_at=timezone.now() - timedelta(days=3))
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual(job.state, 'error')
+        self.assertIn('zwei Tage', job.error)
+
+    @override_settings(WHISPER_URL='')
+    def test_retries_then_succeeds(self):
+        self.fake.fail_first = 2
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            self.configure(parallel=1)
+            job = self.job()
+            transcribe.run(job.pk)
+        self.assertEqual(self.reload(job).state, 'done')
+        self.assertEqual(len(self.fake.calls), 3 + 2)
+
+    def test_gives_up_after_three_retries(self):
+        self.fake.fail_first = -1
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            self.configure(parallel=1)
+            job = self.job()
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual(job.state, 'error')
+        self.assertEqual(job.error, 'Whisper: HTTP 500')
+        self.assertEqual(len(self.fake.calls), 4)   # the first try and three more
+        self.assertIsNone(job.data)
+        self.assertClean()
+
+    def test_a_refusal_is_not_retried(self):
+        self.fake.fail_first, self.fake.fail_status = -1, 401
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            self.configure(parallel=1)
+            job = self.job()
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual((job.state, job.error), ('error', 'Whisper: HTTP 401'))
+        self.assertEqual(len(self.fake.calls), 1)
+
+    def test_download_failure(self):
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            job = self.job('/missing.mp3')
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual((job.state, job.error), ('error', 'Download: HTTP 404'))
+        self.assertClean()
+
+    def test_not_audio(self):
+        with override_settings(WHISPER_URL=self.fake.whisper), patch.object(audio_holder, 'data', b'<html>no</html>'):
+            job = self.job()
+            transcribe.run(job.pk)
+        job = self.reload(job)
+        self.assertEqual(job.state, 'error')
+        self.assertTrue(job.error.startswith('ffmpeg:'), job.error)
+        self.assertClean()
+
+    def test_no_builtin_instance(self):
+        job = self.job()
+        transcribe.run(job.pk)
+        self.assertEqual(self.reload(job).error, 'Keine eingebaute Instanz eingerichtet')
+
+    def test_cancel_while_running(self):
+        job = self.job()
+
+        def cancel(n):
+            if n == 1:
+                TranscriptionJob.objects.filter(pk=job.pk).delete()
+                from django.db import connection
+                connection.close()
+        self.fake.on_call, self.fake.delay = cancel, 0.2
+        with override_settings(WHISPER_URL=self.fake.whisper):
+            self.configure(parallel=1)
+            transcribe.run(job.pk)
+        self.assertFalse(TranscriptionJob.objects.filter(pk=job.pk).exists())
+        self.assertLessEqual(len(self.fake.calls), 1)   # no further piece was sent
+        self.assertClean()
+
+    def test_cancelled_while_waiting_is_not_started(self):
+        job = self.job()
+        job.delete()
+        transcribe.run(job.pk)
+        self.assertEqual(self.fake.calls, [])
+
+
+class _Holder:
+    """Lets a test swap the audio the fake server sends."""
+    data = None
+
+
+audio_holder = _Holder()
+
+
+class QueueTests(PipelineBase):
+    """The runner and its queue. The work itself (_transcribe) is replaced by a stand-in the test
+    holds back, so only the runner thread writes to the database while the test looks at it."""
+
+    def setUp(self):
+        super().setUp()
+        self.gate = threading.Event()
+        self.started = []
+        patcher = patch.object(transcribe, '_resumed', True)
+        patcher.start()
+        self.addCleanup(patcher.stop)
+        work = patch.object(transcribe, '_transcribe', side_effect=self.work)
+        work.start()
+        self.addCleanup(work.stop)
+        self.addCleanup(self.gate.set)
+        self.addCleanup(lambda: wait_for(lambda: transcribe._runner is None))
+
+    def work(self, job_id):
+        self.started.append(job_id)
+        self.gate.wait(30)
+        return {'segments': [{'start': 0, 'end': 1, 'speaker': '', 'text': 'x'}], 'source': {'type': 'whisper'}}
+
+    def test_one_job_at_a_time_oldest_first(self):
+        first, second, third = self.job(), self.job(), self.job()
+        transcribe.kick()
+        self.assertTrue(wait_for(lambda: self.started == [first.pk]))
+        transcribe.kick()   # a second runner doesn't start
+        time.sleep(0.3)
+        self.assertEqual(self.started, [first.pk])
+        self.assertEqual([self.reload(j).state for j in (first, second, third)], ['downloading', 'queued', 'queued'])
+        self.gate.set()
+        self.assertTrue(wait_for(lambda: transcribe._runner is None))
+        self.assertEqual(self.started, [first.pk, second.pk, third.pk])
+        jobs = [self.reload(j) for j in (first, second, third)]
+        self.assertEqual([j.state for j in jobs], ['done'] * 3)
+        self.assertEqual(sorted(j.finished_at for j in jobs), [j.finished_at for j in jobs])
+
+    def test_a_job_queued_while_the_runner_works_is_taken_too(self):
+        first = self.job()
+        transcribe.kick()
+        self.assertTrue(wait_for(lambda: self.started == [first.pk]))
+        second = self.job()
+        transcribe.kick()
+        self.gate.set()
+        self.assertTrue(wait_for(lambda: transcribe._runner is None))
+        self.assertEqual(self.started, [first.pk, second.pk])
+
+    def test_a_cancelled_waiting_job_is_skipped(self):
+        first, second, third = self.job(), self.job(), self.job()
+        transcribe.kick()
+        self.assertTrue(wait_for(lambda: self.started == [first.pk]))
+        second.delete()
+        self.gate.set()
+        self.assertTrue(wait_for(lambda: transcribe._runner is None))
+        self.assertEqual(self.started, [first.pk, third.pk])
+
+    def test_a_crashing_job_does_not_stop_the_queue(self):
+        first, second = self.job(), self.job()
+        self.gate.set()
+        calls = []
+
+        def work(job_id):
+            calls.append(job_id)
+            if job_id == first.pk:
+                raise RuntimeError('boom')
+            return self.work(job_id)
+        with patch.object(transcribe, '_transcribe', side_effect=work):
+            transcribe.kick()
+            self.assertTrue(wait_for(lambda: transcribe._runner is None))
+        self.assertEqual(calls, [first.pk, second.pk])
+        self.assertEqual((self.reload(first).state, self.reload(first).error), ('error', 'RuntimeError: boom'))
+        self.assertEqual(self.reload(second).state, 'done')
+
+    def test_stuck_jobs_start_again_after_a_restart(self):
+        self.gate.set()
+        job = self.job()
+        TranscriptionJob.objects.filter(pk=job.pk).update(state='transcribing', chunks_done=2, chunks_total=3)
+        with patch.object(transcribe, '_resumed', False):
+            transcribe.kick()
+        self.assertTrue(wait_for(lambda: transcribe._runner is None))
+        self.assertEqual(self.started, [job.pk])
+        self.assertEqual(self.reload(job).state, 'done')
+
+    def test_nothing_queued_starts_no_runner(self):
+        transcribe.kick()
+        self.assertIsNone(transcribe._runner)
+
+
+# --------------------------------------------------------------------------- endpoints
+
+class EndpointBase(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username='alice', password='pw12345678')
+        self.client.force_login(self.user)
+        self.feed = PodcastFeed.objects.create(user=self.user, rss_url='https://example.org/f', title='Der Podcast')
+        self.ep = self.episode('e1')
+        kick = patch.object(transcribe, 'kick')
+        self.kick = kick.start()
+        self.addCleanup(kick.stop)
+
+    def episode(self, guid, **kw):
+        kw.setdefault('audio_url', f'https://example.org/{guid}.mp3')
+        return PodcastEpisode.objects.create(feed=self.feed, guid=guid, title=f'Folge {guid}', duration_seconds=600, **kw)
+
+    def post(self, url, body=None, **extra):
+        return self.client.post(url, json.dumps(body or {}), content_type='application/json', **extra)
+
+    def done_job(self, ep, **kw):
+        kw.setdefault('data', {'segments': [{'start': 0, 'end': 4, 'speaker': '', 'text': 'Hallo.'}],
+                               'source': {'type': 'whisper', 'model': 'whisper-1', 'language': 'de'}})
+        return TranscriptionJob.objects.create(episode=ep, state='done', finished_at=timezone.now(), **kw)
+
+
+class TranscribeEndpointTests(EndpointBase):
+    def url(self, ep=None):
+        return f'/podcasts/episodes/{(ep or self.ep).id}/transcribe/'
+
+    def test_requires_login_and_ownership(self):
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        self.client.force_login(other)
+        self.assertEqual(self.client.get(self.url()).status_code, 404)
+        self.assertEqual(self.post(self.url()).status_code, 404)
+        self.client.logout()
+        self.assertEqual(self.client.get(self.url()).status_code, 401)
+
+    def test_none_before_a_job(self):
+        self.assertEqual(self.client.get(self.url()).json(), {
+            'state': 'none', 'chunks_done': 0, 'chunks_total': 0, 'error': '', 'want_article': False,
+            'started_at': None, 'finished_at': None})
+
+    @patch('podcasts.transcription._too_large', return_value=False)
+    def test_post_queues_a_job_and_starts_the_runner(self, _):
+        r = self.post(self.url())
+        self.assertEqual(r.status_code, 200)
+        self.assertEqual((r.json()['state'], r.json()['want_article']), ('queued', False))
+        self.assertEqual(TranscriptionJob.objects.get(episode=self.ep).state, 'queued')
+        self.kick.assert_called()
+
+    @patch('podcasts.transcription._too_large', return_value=False)
+    def test_post_again_only_raises_the_article_wish(self, _):
+        self.post(self.url())
+        first = TranscriptionJob.objects.get(episode=self.ep).pk
+        r = self.post(self.url(), {'article': True})
+        self.assertTrue(r.json()['want_article'])
+        r = self.post(self.url(), {'article': False})   # never lowers it
+        self.assertTrue(r.json()['want_article'])
+        self.assertEqual(TranscriptionJob.objects.filter(episode=self.ep).count(), 1)
+        self.assertEqual(TranscriptionJob.objects.get(episode=self.ep).pk, first)
+
+    @patch('podcasts.transcription._too_large', return_value=False)
+    def test_post_with_article_true_sets_the_wish(self, _):
+        self.assertTrue(self.post(self.url(), {'article': True}).json()['want_article'])
+
+    def test_youtube_is_refused(self):
+        ep = self.episode('yt', audio_url=youtube.WATCH + 'abc123')
+        r = self.post(self.url(ep))
+        self.assertEqual((r.status_code, r.json()), (400, {'error': 'youtube'}))
+        self.assertFalse(TranscriptionJob.objects.exists())
+
+    @patch('podcasts.transcription._too_large', return_value=True)
+    def test_too_large_is_refused(self, _):
+        r = self.post(self.url())
+        self.assertEqual((r.status_code, r.json()), (400, {'error': 'too large'}))
+        self.assertFalse(TranscriptionJob.objects.exists())
+
+    def test_the_size_check_reads_the_content_length(self):
+        from . import silence
+        resp = type('R', (), {'headers': {'Content-Length': str(silence.MAX_BYTES + 1)}})()
+        with patch('podcasts.transcription.requests.head', return_value=resp):
+            self.assertTrue(transcription._too_large('https://example.org/x.mp3'))
+        resp.headers = {'Content-Length': '1000'}
+        with patch('podcasts.transcription.requests.head', return_value=resp):
+            self.assertFalse(transcription._too_large('https://example.org/x.mp3'))
+        with patch('podcasts.transcription.requests.head', side_effect=transcription.requests.ConnectionError):
+            self.assertFalse(transcription._too_large('https://example.org/x.mp3'))
+
+    def test_a_whisper_transcript_is_done_already(self):
+        self.done_job(self.ep)
+        r = self.post(self.url(), {'article': True})
+        self.assertEqual((r.json()['state'], r.json()['want_article']), ('done', True))
+        self.kick.assert_not_called()
+        self.assertTrue(TranscriptionJob.objects.get(episode=self.ep).want_article)
+
+    def test_a_feed_transcript_is_done_already(self):
+        self.ep.transcripts = [{'url': 'https://example.org/t.vtt', 'type': 'text/vtt'}]
+        self.ep.save()
+        r = self.post(self.url(), {'article': True})
+        self.assertEqual((r.json()['state'], r.json()['want_article']), ('done', True))
+        job = TranscriptionJob.objects.get(episode=self.ep)
+        self.assertIsNone(job.data)
+        self.kick.assert_not_called()
+
+    @patch('podcasts.transcription._too_large', return_value=False)
+    def test_after_an_error_post_starts_again_at_the_back(self, _):
+        old = TranscriptionJob.objects.create(episode=self.ep, state='error', error='Whisper: HTTP 500')
+        r = self.post(self.url(), {'article': True})
+        self.assertEqual((r.json()['state'], r.json()['error']), ('queued', ''))
+        new = TranscriptionJob.objects.get(episode=self.ep)
+        self.assertGreater(new.pk, old.pk)
+
+    @patch('podcasts.transcription._too_large', return_value=False)
+    def test_delete_cancels_queued_and_running_but_not_a_finished_one(self, _):
+        self.post(self.url())
+        self.assertEqual(self.client.delete(self.url()).json()['state'], 'none')
+        self.assertFalse(TranscriptionJob.objects.exists())
+        TranscriptionJob.objects.create(episode=self.ep, state='transcribing', chunks_total=5, chunks_done=2)
+        self.assertEqual(self.client.delete(self.url()).json()['state'], 'none')
+        self.assertFalse(TranscriptionJob.objects.exists())
+        self.done_job(self.ep)
+        self.assertEqual(self.client.delete(self.url()).json()['state'], 'done')
+        self.assertTrue(TranscriptionJob.objects.exists())
+
+    def test_get_shows_progress(self):
+        TranscriptionJob.objects.create(episode=self.ep, state='transcribing', chunks_total=38, chunks_done=7,
+                                        started_at=timezone.now())
+        d = self.client.get(self.url()).json()
+        self.assertEqual((d['state'], d['chunks_done'], d['chunks_total']), ('transcribing', 7, 38))
+        self.assertIsNotNone(d['started_at'])
+        self.kick.assert_called()   # a restart's first request starts the queue again
+
+
+class TranscriptTests(EndpointBase):
+    def transcript(self, ep=None):
+        return self.client.get(f'/podcasts/episodes/{(ep or self.ep).id}/transcript/')
+
+    def detail(self, ep=None):
+        return self.client.get(f'/podcasts/episodes/{(ep or self.ep).id}/').json()
+
+    def test_404_as_before_without_any(self):
+        self.assertEqual((self.transcript().status_code, self.transcript().json()), (404, {'error': 'no transcript'}))
+        d = self.detail()
+        self.assertEqual((d['has_transcript'], d['transcript_origin'], d['transcription'], d['article_item_id']),
+                         (False, None, None, None))
+
+    def test_falls_back_to_the_whisper_transcript(self):
+        job = self.done_job(self.ep)
+        r = self.transcript()
+        self.assertEqual(r.status_code, 200)
+        self.assertEqual(r.json(), job.data)
+        d = self.detail()
+        self.assertEqual((d['has_transcript'], d['transcript_origin']), (True, 'whisper'))
+        self.assertEqual(d['transcription']['state'], 'done')
+
+    def test_a_running_job_has_no_transcript_yet(self):
+        TranscriptionJob.objects.create(episode=self.ep, state='transcribing', data=None)
+        self.assertEqual(self.transcript().status_code, 404)
+        d = self.detail()
+        self.assertEqual((d['has_transcript'], d['transcript_origin'], d['transcription']['state']),
+                         (False, None, 'transcribing'))
+
+    def feed_transcript(self, error=''):
+        ep = self.ep
+        ep.transcripts = [{'url': 'https://example.org/t.vtt', 'type': 'text/vtt'}]
+        ep.save()
+        EpisodeTranscript.objects.create(
+            episode=ep, sources=['https://example.org/t.vtt'], error=error,
+            data=None if error else {'segments': [{'start': 0, 'end': 1, 'speaker': 'A', 'text': 'Aus dem Feed'}],
+                                     'source': {'url': 'https://example.org/t.vtt'}})
+
+    def test_the_feeds_transcript_wins(self):
+        self.done_job(self.ep)
+        self.feed_transcript()
+        self.assertEqual(self.transcript().json()['segments'][0]['text'], 'Aus dem Feed')
+        self.assertEqual(self.detail()['transcript_origin'], 'feed')
+
+    def test_an_unusable_feed_transcript_gives_way(self):
+        self.done_job(self.ep)
+        self.feed_transcript(error='invalid')
+        self.assertEqual(self.transcript().json()['segments'][0]['text'], 'Hallo.')
+        self.assertEqual(self.detail()['transcript_origin'], 'whisper')
+
+    def test_a_feed_transcript_that_cannot_be_loaded_now_gives_way_too(self):
+        self.done_job(self.ep)
+        self.ep.transcripts = [{'url': 'https://example.org/t.vtt', 'type': 'text/vtt'}]
+        self.ep.save()
+        from . import chapters
+        with patch('podcasts.views.transcripts_mod.load', side_effect=chapters.FetchUnavailable('x')):
+            self.assertEqual(self.transcript().json()['segments'][0]['text'], 'Hallo.')
+
+    def test_article_item_id(self):
+        source = Source.objects.create(user=self.user, url='diora://podcast-artikel', title='Podcast-Artikel',
+                                       kind='blog')
+        item = Item.objects.create(source=source, user=self.user, kind='blog', guid_hash='x', title='A',
+                                   published=timezone.now())
+        EpisodeArticle.objects.create(episode=self.ep, item=item)
+        self.assertEqual(self.detail()['article_item_id'], item.id)
+
+
+class WishTests(EndpointBase):
+    PENDING = '/podcasts/transcription/pending/'
+
+    def claim(self, ep=None, method='post', **kw):
+        return getattr(self.client, method)(
+            f'/podcasts/episodes/{(ep or self.ep).id}/article/claim/', json.dumps(kw.pop('body', {})),
+            content_type='application/json', **kw)
+
+    def test_pending_lists_open_wishes_only(self):
+        wished = self.done_job(self.ep, want_article=True)
+        e2, e3, e4, e5 = (self.episode(g) for g in ('e2', 'e3', 'e4', 'e5'))
+        self.done_job(e2)   # no wish
+        TranscriptionJob.objects.create(episode=e3, state='transcribing', want_article=True)   # not done
+        self.done_job(e4, want_article=True, claimed_by='tablet', claimed_until=timezone.now() + timedelta(minutes=5))
+        job5 = self.done_job(e5, want_article=True)
+        source = Source.objects.create(user=self.user, url='diora://podcast-artikel', title='P', kind='blog')
+        item = Item.objects.create(source=source, user=self.user, kind='blog', guid_hash='x', title='A',
+                                   published=timezone.now())
+        EpisodeArticle.objects.create(episode=e5, item=item)   # written already
+        r = self.client.get(self.PENDING).json()
+        self.assertEqual(r, [{'episode_id': self.ep.id, 'title': 'Folge e1', 'podcast': 'Der Podcast',
+                              'duration_seconds': 600}])
+        self.assertTrue(wished.want_article and job5.want_article)
+
+    def test_pending_is_per_user_and_needs_login(self):
+        self.done_job(self.ep, want_article=True)
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        self.client.force_login(other)
+        self.assertEqual(self.client.get(self.PENDING).json(), [])
+        self.client.logout()
+        self.assertEqual(self.client.get(self.PENDING).status_code, 401)
+
+    def test_an_expired_claim_is_open_again(self):
+        self.done_job(self.ep, want_article=True, claimed_by='tablet',
+                      claimed_until=timezone.now() - timedelta(seconds=1))
+        self.assertEqual(len(self.client.get(self.PENDING).json()), 1)
+
+    def test_claim_blocks_the_other_device_until_it_expires(self):
+        job = self.done_job(self.ep, want_article=True)
+        r = self.claim(body={'device': 'phone'})
+        self.assertEqual(r.status_code, 200)
+        self.assertEqual(r.json()['claimed_by'], 'phone')
+        job.refresh_from_db()
+        left = job.claimed_until - timezone.now()
+        self.assertTrue(timedelta(minutes=14) < left <= timedelta(minutes=15))
+        self.assertEqual(self.client.get(self.PENDING).json(), [])
+        r = self.claim(body={'device': 'desktop'})
+        self.assertEqual(r.status_code, 409)
+        self.assertEqual(r.json()['error'], 'claimed')
+        self.assertEqual(self.claim(body={'device': 'phone'}).status_code, 200)   # the holder renews
+        TranscriptionJob.objects.filter(pk=job.pk).update(claimed_until=timezone.now() - timedelta(seconds=1))
+        self.assertEqual(self.claim(body={'device': 'desktop'}).status_code, 200)
+        job.refresh_from_db()
+        self.assertEqual(job.claimed_by, 'desktop')
+
+    def test_the_device_may_come_in_a_header(self):
+        self.done_job(self.ep, want_article=True)
+        self.assertEqual(self.claim(HTTP_X_DIORA_DEVICE='phone').status_code, 200)
+        self.assertEqual(self.claim(HTTP_X_DIORA_DEVICE='desktop').status_code, 409)
+
+    def test_a_device_name_is_required(self):
+        self.done_job(self.ep, want_article=True)
+        self.assertEqual(self.claim().status_code, 400)
+
+    def test_release(self):
+        self.done_job(self.ep, want_article=True)
+        self.claim(body={'device': 'phone'})
+        self.assertEqual(self.claim(method='delete', body={'device': 'desktop'}).status_code, 409)
+        self.assertEqual(self.claim(method='delete', body={'device': 'phone'}).status_code, 200)
+        self.assertEqual(len(self.client.get(self.PENDING).json()), 1)
+
+    def test_no_wish_no_claim(self):
+        self.done_job(self.ep)   # done, but no article wanted
+        self.assertEqual(self.claim(body={'device': 'phone'}).status_code, 404)
+        self.assertEqual(self.claim(self.episode('e9'), body={'device': 'phone'}).status_code, 404)
+
+
+class NotificationTests(EndpointBase):
+    def get(self, **q):
+        return self.client.get('/feeds/notifications/', q).json()
+
+    def test_first_time_only_the_cursor(self):
+        j = self.done_job(self.ep, want_article=True)
+        d = self.get()
+        self.assertEqual((d['transcripts'], d['t_last'], d['items']), ([], j.id, []))
+        self.assertEqual(d['last'], 0)
+
+    def test_jobs_done_after_the_cursor(self):
+        e2, e3 = self.episode('e2'), self.episode('e3')
+        a = self.done_job(self.ep)
+        b = self.done_job(e2, want_article=True)
+        TranscriptionJob.objects.create(episode=e3, state='transcribing')   # not done: no notice
+        d = self.get(t_after=a.id)
+        self.assertEqual(d['transcripts'], [{'job_id': b.id, 'episode_id': e2.id, 'title': 'Folge e2',
+                                              'podcast': 'Der Podcast', 'want_article': True}])
+        self.assertEqual(d['t_last'], b.id)
+        self.assertEqual(self.get(t_after=b.id)['transcripts'], [])
+        self.assertEqual([t['job_id'] for t in self.get(t_after=0)['transcripts']], [a.id, b.id])
+
+    def test_only_the_users_own_and_not_feed_only_jobs(self):
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        feed = PodcastFeed.objects.create(user=other, rss_url='https://example.org/o', title='O')
+        ep = PodcastEpisode.objects.create(feed=feed, guid='o', title='O', audio_url='https://example.org/o.mp3')
+        self.done_job(ep)
+        self.done_job(self.episode('e2'), data=None)   # only the wish for an article from a feed transcript
+        d = self.get(t_after=0)
+        self.assertEqual((d['transcripts'], d['t_last']), ([], 0))
+
+    def test_items_still_work_beside_it(self):
+        d = self.get(after=0)
+        self.assertEqual((d['items'], d['last'], d['transcripts'], d['t_last']), ([], 0, [], 0))

podcasts/test_transcription.py

@@ -0,0 +1,153 @@
+import io
+import json
+import wave
+from unittest.mock import MagicMock, patch
+
+from django.contrib.auth.models import User
+from django.test import TestCase, override_settings
+
+from feeds import logins
+
+from . import chapters, transcription
+from .models import TranscriptionSettings
+
+SETTINGS = '/podcasts/transcription/settings/'
+TEST = '/podcasts/transcription/test/'
+
+
+def fake_response(status=200):
+    r = MagicMock()
+    r.status_code = status
+    r.headers = {}
+    r.raw = None   # nothing to introspect for the peer check
+    return r
+
+
+class SettingsTests(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username='alice', password='pw12345678')
+        self.client.force_login(self.user)
+
+    def post(self, data, url=SETTINGS):
+        return self.client.post(url, json.dumps(data), content_type='application/json')
+
+    def test_requires_login(self):
+        self.client.logout()
+        self.assertEqual(self.client.get(SETTINGS).status_code, 401)
+        self.assertEqual(self.post({}, TEST).status_code, 401)
+
+    def test_defaults(self):
+        self.assertEqual(self.client.get(SETTINGS).json(), {
+            'engine': 'builtin', 'url': '', 'model': '', 'language': '', 'parallel': 3,
+            'has_token': False, 'builtin_available': False})
+        self.assertFalse(TranscriptionSettings.objects.exists())
+
+    @override_settings(WHISPER_URL='http://whisper:8000/v1')
+    def test_builtin_available_follows_the_environment_setting(self):
+        self.assertTrue(self.client.get(SETTINGS).json()['builtin_available'])
+
+    def test_roundtrip(self):
+        r = self.post({'engine': 'custom', 'url': 'https://whisper.example.org/v1/', 'model': 'large-v3',
+                       'language': 'de', 'parallel': 5, 'token': 'sk-secret'})
+        self.assertEqual(r.status_code, 200)
+        want = {'engine': 'custom', 'url': 'https://whisper.example.org/v1', 'model': 'large-v3',
+                'language': 'de', 'parallel': 5, 'has_token': True, 'builtin_available': False}
+        self.assertEqual(r.json(), want)
+        self.assertEqual(self.client.get(SETTINGS).json(), want)
+
+    def test_token_is_write_only_and_stored_encrypted(self):
+        self.post({'engine': 'custom', 'url': 'https://w.example.org/v1', 'token': 'sk-secret'})
+        conf = TranscriptionSettings.objects.get(user=self.user)
+        self.assertTrue(conf.token_encrypted)
+        self.assertNotIn('sk-secret', conf.token_encrypted)
+        self.assertEqual(transcription.token_of(conf), 'sk-secret')
+        for body in (self.client.get(SETTINGS).content, self.post({'model': 'm'}).content):
+            self.assertNotIn(b'sk-secret', body)
+            self.assertNotIn(conf.token_encrypted.encode(), body)
+        # Not sending a token keeps it; an empty one removes it.
+        conf.refresh_from_db()
+        self.assertTrue(conf.token_encrypted)
+        self.assertFalse(self.post({'token': ''}).json()['has_token'])
+        conf.refresh_from_db()
+        self.assertEqual(conf.token_encrypted, '')
+
+    def test_validation(self):
+        bad = [
+            {'engine': 'cloud'}, {'parallel': 0}, {'parallel': 9}, {'parallel': '3'}, {'parallel': True},
+            {'url': 'ftp://x.org/v1'}, {'url': 'javascript:alert(1)'}, {'url': 'whisper.local/v1'},
+            {'language': 'Deutsch'}, {'model': 5}, {'token': 5},
+            {'engine': 'custom'},   # a custom instance needs a url
+        ]
+        for body in bad:
+            self.assertEqual(self.post(body).status_code, 400, body)
+        self.assertEqual(self.client.post(SETTINGS, 'nope', content_type='application/json').status_code, 400)
+        self.assertFalse(TranscriptionSettings.objects.exists())
+        for body in ({'parallel': 1}, {'parallel': 8}, {'language': ''}, {'language': 'pt-BR'}):
+            self.assertEqual(self.post(body).status_code, 200, body)
+
+    def test_settings_are_per_user(self):
+        self.post({'parallel': 7})
+        other = User.objects.create_user(username='bob', password='pw12345678')
+        self.client.force_login(other)
+        self.assertEqual(self.client.get(SETTINGS).json()['parallel'], 3)
+
+
+class TestEndpointTests(TestCase):
+    def setUp(self):
+        self.user = User.objects.create_user(username='alice', password='pw12345678')
+        self.client.force_login(self.user)
+
+    def configure(self, **kw):
+        TranscriptionSettings.objects.update_or_create(user=self.user, defaults=kw)
+
+    def test_custom_sends_a_second_of_silence(self):
+        self.configure(engine='custom', url='https://whisper.example.org/v1', model='large-v3', language='de',
+                       token_encrypted=logins.encrypt('sk-secret'))
+        calls = []
+
+        def post(url, **kw):
+            calls.append((url, kw))
+            return fake_response()
+
+        with patch('podcasts.chapters.assert_safe_url'), patch('podcasts.chapters.requests.post', post):
+            r = self.client.post(TEST)
+        self.assertEqual(r.json(), {'ok': True, 'model': 'large-v3', 'error': ''})
+        url, kw = calls[0]
+        self.assertEqual(url, 'https://whisper.example.org/v1/audio/transcriptions')
+        self.assertEqual(kw['headers']['Authorization'], 'Bearer sk-secret')
+        self.assertEqual(kw['data']['model'], 'large-v3')
+        self.assertEqual(kw['data']['language'], 'de')
+        name, wav, mime = kw['files']['file']
+        self.assertEqual(mime, 'audio/wav')
+        with wave.open(io.BytesIO(wav)) as w:
+            self.assertEqual((w.getframerate(), w.getnchannels(), w.getnframes()), (16000, 1, 16000))
+
+    def test_custom_private_address_is_refused(self):
+        self.configure(engine='custom', url='http://127.0.0.1:9000/v1')
+        with patch('podcasts.chapters.requests.post') as post:
+            r = self.client.post(TEST)
+        self.assertFalse(r.json()['ok'])
+        self.assertTrue(r.json()['error'])
+        post.assert_not_called()
+
+    def test_custom_http_error_is_reported(self):
+        self.configure(engine='custom', url='https://whisper.example.org/v1')
+        with patch('podcasts.chapters.assert_safe_url'), \
+                patch('podcasts.chapters.requests.post', return_value=fake_response(401)):
+            r = self.client.post(TEST)
+        self.assertEqual((r.json()['ok'], r.json()['error']), (False, 'HTTP 401'))
+
+    @override_settings(WHISPER_URL='http://whisper:8000/v1')
+    def test_builtin_is_exempt_from_the_address_check(self):
+        self.configure(engine='builtin')
+        with patch('podcasts.transcription.requests.post', return_value=fake_response()) as post, \
+                patch('podcasts.chapters.assert_safe_url', side_effect=chapters.WebDAVInputError('private')):
+            r = self.client.post(TEST)
+        self.assertTrue(r.json()['ok'])
+        self.assertEqual(post.call_args.args[0], 'http://whisper:8000/v1/audio/transcriptions')
+        self.assertNotIn('Authorization', post.call_args.kwargs['headers'])
+
+    def test_builtin_without_instance(self):
+        r = self.client.post(TEST)
+        self.assertFalse(r.json()['ok'])
+        self.assertTrue(r.json()['error'])

podcasts/tests.py

@@ -837,7 +837,7 @@ class EpisodeListTests(TestCase):
         a2 = dict(eps[self.a2.id])
         self.assertTrue(a2.pop('discovered_at'))   # when diora found it (the Strom's mark)
         self.assertEqual(a2, {
-            'id': self.a2.id, 'title': 'A2', 'audio_url': 'https://example.org/a2.mp3', 'duration_seconds': 0,
+            'id': self.a2.id, 'title': 'A2', 'audio_url': 'https://example.org/a2.mp3', 'link': '', 'duration_seconds': 0,
             'pub_date': '2026-01-02T09:00:00+00:00', 'artwork_url': '',
             'feed__id': self.fa.id, 'feed__title': 'Alpha', 'feed__artwork_url': 'https://example.org/a.jpg',
             'position_seconds': 100, 'played': False, 'favorite': False, 'in_queue': True, 'in_later': False,

podcasts/transcribe.py

@@ -0,0 +1,431 @@
+"""Transcribe an episode with a Whisper instance (docs/transcription.md, "Transcribing").
+
+The server downloads the episode (as silence.py does, with the same size limit), decodes it
+once to 16 kHz mono, cuts it into pieces of about eight minutes where there is a pause (so no
+word is cut), sends the pieces in parallel to the instance's OpenAI-compatible
+`/audio/transcriptions` and joins the answers into one transcript, which is kept on the job
+(TranscriptionJob.data).
+
+A job runs in a daemon thread, one at a time per server; the others wait as `queued` and are
+taken oldest first (by id). A job found `downloading` or `transcribing` when the process has
+just started (a restart killed it) is queued again. Cancelling deletes the job's row: the
+worker notices when it next writes progress and stops.
+"""
+import datetime
+import logging
+import os
+import subprocess
+import tempfile
+import threading
+import time
+from concurrent.futures import FIRST_COMPLETED, ThreadPoolExecutor, wait
+
+import requests
+from django.conf import settings
+from django.db import close_old_connections
+from django.utils import timezone
+
+from . import chapters, silence, transcripts
+from .models import TranscriptionJob, TranscriptionSettings
+from .transcription import DEFAULT_MODEL, for_user, token_of
+
+log = logging.getLogger(__name__)
+
+CHUNK_SECONDS = 8 * 60      # about this long a piece
+CUT_WINDOW = 30             # a cut moves to the longest pause within this many seconds
+MIN_TAIL = 60               # a last piece shorter than this is joined to the one before
+RATE = 16000                # Hz, mono, 16-bit
+WINDOW = silence.WINDOW
+RETRIES = 3
+RETRY_PAUSES = (3, 10, 30)  # seconds before each retry
+UPLOAD_TIMEOUT = (10, 3600)  # a CPU instance may need many minutes for a piece
+DECODE_TIMEOUT = 3600
+OPUS_BITRATE = '32k'
+MAX_PROMPT = 500            # characters
+ALIVE_CHECK = 5             # seconds between looks whether the job was cancelled
+WAIT_RETRY = 600            # seconds between tries while the built-in instance can't be reached
+WAIT_MAX = 48 * 3600        # after this long since the button was pressed a waiting job gives up
+WAIT_IDLE = 30              # seconds the runner sleeps while every queued job is waiting
+
+_lock = threading.Lock()
+_runner = None              # the thread working through the queue, or None
+_resumed = False            # whether jobs left over from before a restart were queued again
+_not_before = {}            # job id -> time.time() before which a waiting job isn't tried again
+WAIT = 'wait'               # _next_job: jobs are queued, but none is due yet
+
+
+class Cancelled(Exception):
+    """The job's row is gone: the user cancelled it."""
+
+
+class Failed(Exception):
+    """The job can't succeed; the message is shown to the user."""
+
+
+class Unreachable(Exception):
+    """The built-in instance can't be reached (e.g. the machine it runs on is off): the job waits
+    and tries again later instead of failing."""
+
+
+class Permanent(Exception):
+    """A request the instance refused for good (a client error): not worth a retry."""
+
+
+# ---------------------------------------------------------------------------
+# Queue
+# ---------------------------------------------------------------------------
+
+def kick():
+    """Make sure queued jobs are being worked on. Called by the endpoints, so after a restart
+    the first request brings back what was running."""
+    global _runner, _resumed
+    with _lock:
+        if not _resumed:
+            _resumed = True
+            TranscriptionJob.objects.filter(
+                state__in=[TranscriptionJob.DOWNLOADING, TranscriptionJob.TRANSCRIBING],
+            ).update(state=TranscriptionJob.QUEUED, chunks_done=0, chunks_total=0)
+        if _runner is not None:
+            return
+        if not TranscriptionJob.objects.filter(state=TranscriptionJob.QUEUED).exists():
+            return
+        _runner = threading.Thread(target=_loop, name='transcribe', daemon=True)
+        _runner.start()
+
+
+def _next_job():
+    global _runner
+    with _lock:
+        queued = list(TranscriptionJob.objects.filter(state=TranscriptionJob.QUEUED)
+                      .order_by('id').values_list('id', flat=True))
+        if not queued:
+            _runner = None   # decided under the lock, so a job queued right now starts a new runner
+            return None
+        now = time.time()
+        due = next((i for i in queued if _not_before.get(i, 0) <= now), None)
+        return due if due is not None else WAIT
+
+
+def _loop():
+    global _runner
+    try:
+        while True:
+            close_old_connections()
+            job_id = _next_job()
+            if job_id is None:
+                return
+            if job_id == WAIT:
+                time.sleep(WAIT_IDLE)
+                continue
+            try:
+                run(job_id)
+            except Exception:  # run() reports its own failures; this keeps the queue going
+                log.exception('transcription job %s crashed', job_id)
+    except BaseException:
+        with _lock:
+            _runner = None
+        raise
+    finally:
+        close_old_connections()
+
+
+def run(job_id):
+    """Work one job through to `done` or `error`, in this thread."""
+    started = TranscriptionJob.objects.filter(pk=job_id, state=TranscriptionJob.QUEUED).update(
+        state=TranscriptionJob.DOWNLOADING, started_at=timezone.now(), finished_at=None,
+        error='', chunks_done=0, chunks_total=0)
+    if not started:
+        return   # cancelled while it waited
+    try:
+        data = _transcribe(job_id)
+        _write(job_id, state=TranscriptionJob.DONE, data=data, error='', finished_at=timezone.now())
+    except Cancelled:
+        log.info('transcription job %s cancelled', job_id)
+    except Unreachable as e:
+        _wait(job_id, str(e))
+    except Failed as e:
+        _fail(job_id, str(e))
+    except Exception as e:
+        log.exception('transcription job %s failed', job_id)
+        _fail(job_id, f'{type(e).__name__}: {e}')
+
+
+def _wait(job_id, reason):
+    """Back into the queue to try again in WAIT_RETRY seconds, or give up when it waited too long."""
+    job = TranscriptionJob.objects.filter(pk=job_id).only('created_at').first()
+    if job is None:
+        return
+    if (timezone.now() - job.created_at).total_seconds() > WAIT_MAX:
+        _not_before.pop(job_id, None)
+        _fail(job_id, 'Whisper-Instanz zwei Tage lang nicht erreichbar')
+        return
+    _not_before[job_id] = time.time() + WAIT_RETRY
+    again = timezone.localtime(timezone.now() + datetime.timedelta(seconds=WAIT_RETRY)).strftime('%H:%M')
+    log.info('transcription job %s waits for the instance (%s)', job_id, reason)
+    TranscriptionJob.objects.filter(pk=job_id).update(
+        state=TranscriptionJob.QUEUED, chunks_done=0, chunks_total=0, started_at=None,
+        error=f'Whisper-Instanz nicht erreichbar, neuer Versuch um {again}')
+
+
+def _probe(url):
+    """Unreachable unless something answers at the built-in instance's address."""
+    try:
+        requests.get(url.rsplit('/audio/transcriptions', 1)[0], timeout=(5, 10)).close()
+    except (requests.ConnectionError, requests.Timeout) as e:
+        raise Unreachable(e.__class__.__name__)
+
+
+def _fail(job_id, message):
+    TranscriptionJob.objects.filter(pk=job_id).update(
+        state=TranscriptionJob.ERROR, error=(message or 'failed')[:500], finished_at=timezone.now())
+
+
+def _write(job_id, **fields):
+    """Update the job; Cancelled when its row is gone."""
+    if not TranscriptionJob.objects.filter(pk=job_id).update(**fields):
+        raise Cancelled()
+
+
+def _alive(job_id):
+    if not TranscriptionJob.objects.filter(pk=job_id).exists():
+        raise Cancelled()
+
+
+# ---------------------------------------------------------------------------
+# The work
+# ---------------------------------------------------------------------------
+
+def _target(conf):
+    """(url, headers, builtin) of the instance the settings point at."""
+    builtin = conf.engine == TranscriptionSettings.BUILTIN
+    base = settings.WHISPER_URL if builtin else conf.url
+    if not base:
+        raise Failed('Keine eingebaute Instanz eingerichtet' if builtin else 'Keine Adresse eingetragen')
+    headers = {'User-Agent': 'diora transcription'}
+    if not builtin and conf.token_encrypted:
+        headers['Authorization'] = 'Bearer ' + token_of(conf)
+    return base.rstrip('/') + '/audio/transcriptions', headers, builtin
+
+
+def _prompt(episode):
+    parts = [episode.feed.title, episode.title, episode.feed.author]
+    return '. '.join(p.strip() for p in parts if p and p.strip())[:MAX_PROMPT]
+
+
+def _transcribe(job_id):
+    job = TranscriptionJob.objects.select_related('episode__feed').get(pk=job_id)
+    episode = job.episode
+    conf = for_user(episode.feed.user)
+    url, headers, builtin = _target(conf)
+    if builtin:
+        _probe(url)   # before downloading: is the instance there at all?
+    options = {'model': conf.model or DEFAULT_MODEL, 'response_format': 'verbose_json'}
+    if conf.language:
+        options['language'] = conf.language
+    prompt = _prompt(episode)
+    if prompt:
+        options['prompt'] = prompt
+
+    with tempfile.TemporaryDirectory(prefix='diora-transcribe-') as tmp:
+        source = os.path.join(tmp, 'episode.audio')
+        _download(episode.audio_url, source, job_id)
+        pcm = os.path.join(tmp, 'episode.pcm')
+        _decode(source, pcm)
+        os.remove(source)
+        total = os.path.getsize(pcm) // 2 / RATE
+        if total <= 0:
+            raise Failed('Kein Ton in der Datei')
+        pieces = plan(total, pauses_of(pcm))
+        _write(job_id, state=TranscriptionJob.TRANSCRIBING, chunks_total=len(pieces), chunks_done=0)
+
+        results = {}
+        stop = threading.Event()   # set when the job ends early: no piece starts after that
+        suffix, mime = ('.flac', 'audio/flac') if builtin else ('.ogg', 'audio/ogg')
+
+        def work(index):
+            # Only the HTTP side runs here; the database is touched by this thread alone.
+            if stop.is_set():
+                return index, None
+            start, end = pieces[index]
+            path = _encode(pcm, start, end, tmp, index, opus=not builtin)
+            try:
+                with open(path, 'rb') as f:
+                    body = f.read()
+            finally:
+                os.remove(path)
+            return index, _send(url, headers, builtin, f'piece-{index}{suffix}', mime, body, options, stop)
+
+        pool = ThreadPoolExecutor(max_workers=max(1, min(conf.parallel, 8)))
+        try:
+            pending = {pool.submit(work, i) for i in range(len(pieces))}
+            while pending:
+                finished, pending = wait(pending, timeout=ALIVE_CHECK, return_when=FIRST_COMPLETED)
+                _alive(job_id)   # also while a long piece is out
+                for f in finished:
+                    index, answer = f.result()   # the first failure ends the job
+                    results[index] = answer
+                if finished:
+                    _write(job_id, chunks_done=len(results))
+        finally:
+            stop.set()
+            pool.shutdown(wait=True, cancel_futures=True)
+    return build(pieces, [results[i] for i in range(len(pieces))], options['model'], conf.language)
+
+
+def _download(url, path, job_id):
+    size = 0
+    with open(path, 'wb') as out, requests.get(
+            url, stream=True, timeout=(10, 60), headers={'User-Agent': 'diora transcription'}) as r:
+        if r.status_code >= 400:
+            raise Failed(f'Download: HTTP {r.status_code}')
+        if int(r.headers.get('Content-Length') or 0) > silence.MAX_BYTES:
+            raise Failed('too large')
+        for n, chunk in enumerate(r.iter_content(1 << 20)):
+            size += len(chunk)
+            if size > silence.MAX_BYTES:
+                raise Failed('too large')
+            out.write(chunk)
+            if n % 8 == 0:
+                _alive(job_id)
+    if size == 0:
+        raise Failed('Download: leere Datei')
+
+
+def _run_ffmpeg(args, input=None, timeout=DECODE_TIMEOUT):
+    import imageio_ffmpeg
+    proc = subprocess.run(
+        [imageio_ffmpeg.get_ffmpeg_exe(), '-nostdin', '-hide_banner', '-loglevel', 'error', *args],
+        input=input, stdout=subprocess.PIPE, stderr=subprocess.PIPE, timeout=timeout)
+    if proc.returncode != 0:
+        err = proc.stderr.decode('utf-8', 'replace').strip().splitlines()
+        raise Failed('ffmpeg: ' + (err[-1] if err else f'exit {proc.returncode}')[:200])
+
+
+def _decode(source, pcm):
+    """The audio at `source` as raw 16 kHz mono 16-bit samples in the file `pcm`."""
+    _run_ffmpeg(['-vn', '-sn', '-dn', '-i', source, '-ac', '1', '-ar', str(RATE), '-f', 's16le', '-y', pcm])
+
+
+def pauses_of(pcm):
+    """The pauses of the decoded audio as [(middle, length)] in seconds, found with
+    silence.py's rules (what it cuts out of a pause is centred on the pause)."""
+    size = int(RATE * WINDOW) * 2
+    levels = []
+    with open(pcm, 'rb') as f:
+        while True:
+            data = f.read(size * 100000)
+            if not data:
+                break
+            for i in range(0, len(data) - size + 1, size):
+                levels.append(silence._level_db(data[i:i + size]))
+    return [((a + b) / 2, b - a) for a, b in silence.cuts(levels)]
+
+
+def plan(total, pauses):
+    """[(start, end)] in seconds covering 0..total: pieces of about CHUNK_SECONDS, each cut at
+    the longest pause within CUT_WINDOW of where it would fall (right there when there is none)."""
+    cuts = []
+    last = 0.0
+    k = 1
+    while k * CHUNK_SECONDS < total - MIN_TAIL:
+        target = k * CHUNK_SECONDS
+        k += 1
+        near = [p for p in pauses if abs(p[0] - target) <= CUT_WINDOW and p[0] > last + 1]
+        at = max(near, key=lambda p: p[1])[0] if near else target
+        if at <= last or at >= total:
+            continue
+        cuts.append(at)
+        last = at
+    edges = [0.0, *cuts, total]
+    return [(edges[i], edges[i + 1]) for i in range(len(edges) - 1)]
+
+
+def _encode(pcm, start, end, tmp, index, opus):
+    """The piece [start, end) of the decoded audio as FLAC or Opus (ogg), in a file."""
+    with open(pcm, 'rb') as f:
+        f.seek(int(round(start * RATE)) * 2)
+        data = f.read(int(round((end - start) * RATE)) * 2)
+    out = os.path.join(tmp, f'piece-{index}.' + ('ogg' if opus else 'flac'))
+    codec = ['-c:a', 'libopus', '-b:a', OPUS_BITRATE, '-f', 'ogg'] if opus else ['-c:a', 'flac', '-f', 'flac']
+    _run_ffmpeg(['-f', 's16le', '-ar', str(RATE), '-ac', '1', '-i', 'pipe:0', *codec, '-y', out], input=data)
+    return out
+
+
+def _post(url, headers, builtin, name, mime, body, options):
+    """One request; the parsed answer. Permanent for a refusal that won't change, RuntimeError
+    for trouble that may pass."""
+    files = {'file': (name, body, mime)}
+    try:
+        if builtin:
+            # The operator's own instance (DIORA_WHISPER_URL) may sit on a private network.
+            try:
+                r = requests.post(url, headers=headers, files=files, data=options, timeout=UPLOAD_TIMEOUT)
+            except requests.ConnectionError as e:
+                raise Unreachable(e.__class__.__name__)
+            if r.status_code >= 400:
+                if r.status_code < 500 and r.status_code not in (408, 429):
+                    raise Permanent(f'HTTP {r.status_code}')
+                raise RuntimeError(f'HTTP {r.status_code}')
+        else:
+            try:
+                r = chapters._open(url, headers, method='POST', timeout=UPLOAD_TIMEOUT, files=files, data=options)
+            except chapters.ChaptersInvalid as e:
+                raise Permanent(str(e))
+            except chapters.ChaptersUnavailable as e:
+                raise RuntimeError(str(e))
+        try:
+            return r.json()
+        finally:
+            r.close()
+    except requests.RequestException as e:
+        raise RuntimeError(e.__class__.__name__)
+    except ValueError:
+        raise RuntimeError('Antwort ist kein JSON')
+
+
+def _send(url, headers, builtin, name, mime, body, options, stop):
+    """A piece to the instance, retried up to RETRIES times with a growing pause."""
+    for attempt in range(RETRIES + 1):
+        if stop.is_set():
+            raise Cancelled()
+        try:
+            return _post(url, headers, builtin, name, mime, body, options)
+        except Permanent as e:
+            raise Failed(f'Whisper: {e}')
+        except Unreachable:
+            if attempt >= RETRIES:
+                raise
+            time.sleep(RETRY_PAUSES[min(attempt, len(RETRY_PAUSES) - 1)])
+        except RuntimeError as e:
+            if attempt >= RETRIES:
+                raise Failed(f'Whisper: {e}')
+            log.info('whisper piece %s failed (%s), retrying', name, e)
+            time.sleep(RETRY_PAUSES[min(attempt, len(RETRY_PAUSES) - 1)])
+
+
+def build(pieces, answers, model, language):
+    """{segments, source} from the pieces' places and the instance's verbose_json answers."""
+    fragments = []
+    detected = ''
+    for (start, end), answer in zip(pieces, answers):
+        if not isinstance(answer, dict):
+            continue
+        detected = detected or (answer.get('language') or '')
+        segments = answer.get('segments')
+        if not isinstance(segments, list):   # a plain {"text": …} answer: one segment for the piece
+            text = (answer.get('text') or '').strip()
+            segments = [{'start': 0, 'end': end - start, 'text': text}] if text else []
+        for s in segments:
+            if not isinstance(s, dict):
+                continue
+            text = ' '.join(str(s.get('text') or '').split())
+            try:
+                a, b = float(s.get('start', 0)), float(s.get('end', 0))
+            except (TypeError, ValueError):
+                continue
+            if text:
+                fragments.append((start + a, start + b, '', text))
+    merged = transcripts.merge(fragments)
+    if not merged:
+        raise Failed('Kein Text erkannt')
+    return {'segments': merged, 'source': {'type': 'whisper', 'model': model, 'language': language or detected}}

podcasts/transcription.py

@@ -0,0 +1,297 @@
+"""Transcription settings and the test of the configured Whisper instance
+(docs/transcription.md). The job itself is podcasts/transcribe.py."""
+
+import io
+import json
+import re
+import wave
+from urllib.parse import urlsplit
+
+import requests
+from django.conf import settings
+from django.http import JsonResponse
+from django.utils import timezone
+from django.views.decorators.csrf import csrf_exempt
+from django.views.decorators.http import require_http_methods
+
+from feeds import logins
+
+from . import chapters
+from .models import EpisodeArticle, PodcastEpisode, TranscriptionJob, TranscriptionSettings
+
+DEFAULT_MODEL = 'whisper-1'
+TEST_TIMEOUT = 60            # a cold CPU instance may load its model first
+LANGUAGE_RE = re.compile(r'^[a-z]{2,3}(-[A-Za-z]{2,4})?$')
+
+
+def _auth(request):
+    if not request.user.is_authenticated:
+        return JsonResponse({'error': 'authentication required'}, status=401)
+    return None
+
+
+def _body(request):
+    try:
+        body = json.loads(request.body or b'{}')
+    except (ValueError, UnicodeDecodeError):
+        return None
+    return body if isinstance(body, dict) else None
+
+
+def for_user(user):
+    """The user's settings (unsaved defaults when there are none)."""
+    return TranscriptionSettings.objects.filter(user=user).first() or TranscriptionSettings(user=user)
+
+
+def token_of(conf):
+    """The plain API token ('' when there is none or it can't be read)."""
+    return (logins.decrypt(conf.token_encrypted) or '') if conf.token_encrypted else ''
+
+
+def _data(conf):
+    return {
+        'engine': conf.engine, 'url': conf.url, 'model': conf.model, 'language': conf.language,
+        'parallel': conf.parallel, 'has_token': bool(conf.token_encrypted),
+        'builtin_available': bool(settings.WHISPER_URL),
+    }
+
+
+@csrf_exempt
+@require_http_methods(['GET', 'POST'])
+def transcription_settings(request):
+    if (err := _auth(request)):
+        return err
+    conf = for_user(request.user)
+    if request.method == 'GET':
+        return JsonResponse(_data(conf))
+    body = _body(request)
+    if body is None:
+        return JsonResponse({'error': 'invalid JSON'}, status=400)
+    if 'engine' in body:
+        if body['engine'] not in dict(TranscriptionSettings.ENGINES):
+            return JsonResponse({'error': 'engine must be builtin or custom'}, status=400)
+        conf.engine = body['engine']
+    if 'url' in body:
+        url = body['url']
+        if not isinstance(url, str):
+            return JsonResponse({'error': 'url must be text'}, status=400)
+        url = url.strip().rstrip('/')
+        if url and (urlsplit(url).scheme not in ('http', 'https') or not urlsplit(url).hostname):
+            return JsonResponse({'error': 'url must start with http:// or https://'}, status=400)
+        if len(url) > 500:
+            return JsonResponse({'error': 'url too long'}, status=400)
+        conf.url = url
+    if 'model' in body:
+        if not isinstance(body['model'], str) or len(body['model'].strip()) > 100:
+            return JsonResponse({'error': 'invalid model'}, status=400)
+        conf.model = body['model'].strip()
+    if 'language' in body:
+        lang = body['language']
+        if not isinstance(lang, str) or (lang.strip() and not LANGUAGE_RE.match(lang.strip())):
+            return JsonResponse({'error': 'language must be empty or an ISO code'}, status=400)
+        conf.language = lang.strip()
+    if 'parallel' in body:
+        n = body['parallel']
+        if isinstance(n, bool) or not isinstance(n, int) or not 1 <= n <= 8:
+            return JsonResponse({'error': 'parallel must be 1 to 8'}, status=400)
+        conf.parallel = n
+    if 'token' in body:
+        if not isinstance(body['token'], str):
+            return JsonResponse({'error': 'token must be text'}, status=400)
+        conf.token_encrypted = logins.encrypt(body['token']) if body['token'] else ''
+    if conf.engine == TranscriptionSettings.CUSTOM and not conf.url:
+        return JsonResponse({'error': 'a custom instance needs a url'}, status=400)
+    conf.save()
+    return JsonResponse(_data(conf))
+
+
+def silent_wav(seconds=1, rate=16000):
+    """`seconds` of 16-bit mono silence as a WAV file, made in memory."""
+    buf = io.BytesIO()
+    with wave.open(buf, 'wb') as w:
+        w.setnchannels(1)
+        w.setsampwidth(2)
+        w.setframerate(rate)
+        w.writeframes(b'\0\0' * rate * seconds)
+    return buf.getvalue()
+
+
+@csrf_exempt
+@require_http_methods(['POST'])
+def transcription_test(request):
+    """One second of silence to the configured instance: {ok, model, error}."""
+    if (err := _auth(request)):
+        return err
+    conf = for_user(request.user)
+    builtin = conf.engine == TranscriptionSettings.BUILTIN
+    base = settings.WHISPER_URL if builtin else conf.url
+    model = conf.model or DEFAULT_MODEL
+    if not base:
+        error = 'Keine eingebaute Instanz eingerichtet' if builtin else 'Keine Adresse eingetragen'
+        return JsonResponse({'ok': False, 'model': model, 'error': error})
+    url = base.rstrip('/') + '/audio/transcriptions'
+    headers = {'User-Agent': 'diora transcription test'}
+    if not builtin and conf.token_encrypted:
+        headers['Authorization'] = 'Bearer ' + token_of(conf)
+    files = {'file': ('silence.wav', silent_wav(), 'audio/wav')}
+    data = {'model': model, 'response_format': 'json'}
+    if conf.language:
+        data['language'] = conf.language
+    try:
+        if builtin:
+            # The operator's own instance (DIORA_WHISPER_URL) may sit on a private network.
+            r = requests.post(url, headers=headers, files=files, data=data, timeout=TEST_TIMEOUT)
+            if r.status_code >= 400:
+                raise chapters.ChaptersInvalid(f'HTTP {r.status_code}')
+        else:
+            r = chapters._open(url, headers, method='POST', timeout=TEST_TIMEOUT, files=files, data=data)
+        r.close()
+    except (chapters.ChaptersInvalid, chapters.ChaptersUnavailable) as e:
+        return JsonResponse({'ok': False, 'model': model, 'error': str(e)})
+    except requests.RequestException as e:
+        return JsonResponse({'ok': False, 'model': model, 'error': e.__class__.__name__})
+    return JsonResponse({'ok': True, 'model': model, 'error': ''})
+
+
+# ---------------------------------------------------------------------------
+# Jobs
+# ---------------------------------------------------------------------------
+
+CLAIM_MINUTES = 15
+ACTIVE = (TranscriptionJob.QUEUED, TranscriptionJob.DOWNLOADING, TranscriptionJob.TRANSCRIBING)
+
+
+def job_state(job):
+    """The job as the API shows it; state `none` when there is no job."""
+    if job is None:
+        return {'state': TranscriptionJob.NONE, 'chunks_done': 0, 'chunks_total': 0, 'error': '',
+                'want_article': False, 'started_at': None, 'finished_at': None}
+    return {
+        'state': job.state, 'chunks_done': job.chunks_done, 'chunks_total': job.chunks_total,
+        'error': job.error, 'want_article': job.want_article,
+        'started_at': job.started_at.isoformat() if job.started_at else None,
+        'finished_at': job.finished_at.isoformat() if job.finished_at else None,
+    }
+
+
+def job_of(episode):
+    return TranscriptionJob.objects.filter(episode=episode).first()
+
+
+def _too_large(url):
+    """Whether the server says the file at `url` is over the size limit (a failed look doesn't count)."""
+    from . import silence
+    try:
+        r = requests.head(url, timeout=5, allow_redirects=True, headers={'User-Agent': 'diora transcription'})
+        return int(r.headers.get('Content-Length') or 0) > silence.MAX_BYTES
+    except (requests.RequestException, ValueError):
+        return False
+
+
+@csrf_exempt
+@require_http_methods(['GET', 'POST', 'DELETE'])
+def episode_transcribe(request, pk):
+    """POST starts the episode's transcription (or just raises `want_article` on one that exists),
+    GET tells how far it is, DELETE cancels a queued or running one."""
+    from . import transcribe, views
+    from .youtube import is_video
+    if (err := _auth(request)):
+        return err
+    ep = PodcastEpisode.objects.select_related('feed').filter(pk=pk, feed__user=request.user).first()
+    if ep is None:
+        return JsonResponse({'error': 'not found'}, status=404)
+    job = job_of(ep)
+
+    if request.method == 'GET':
+        transcribe.kick()   # after a restart the first request brings the work back
+        return JsonResponse(job_state(job))
+
+    if request.method == 'DELETE':
+        if job is not None and job.state != TranscriptionJob.DONE:
+            job.delete()    # a running worker notices its row is gone and stops
+            job = None
+        return JsonResponse(job_state(job))
+
+    body = _body(request)
+    if body is None:
+        return JsonResponse({'error': 'invalid JSON'}, status=400)
+    article = body.get('article') is True
+    if is_video(ep.audio_url):
+        return JsonResponse({'error': 'youtube'}, status=400)
+
+    has_feed = views._has_transcript(ep)
+    if job is not None and (job.state in ACTIVE or (job.state == TranscriptionJob.DONE and (job.data or has_feed))):
+        if article and not job.want_article:
+            job.want_article = True
+            job.save(update_fields=['want_article'])
+        return JsonResponse(job_state(job))
+    if has_feed:   # the feed's own transcript is there: only the article is wanted
+        if job is None:
+            job = TranscriptionJob(episode=ep)
+        job.state, job.want_article, job.finished_at = TranscriptionJob.DONE, article or job.want_article, timezone.now()
+        job.save()
+        return JsonResponse(job_state(job))
+
+    if not ep.audio_url:
+        return JsonResponse({'error': 'no audio'}, status=400)
+    if _too_large(ep.audio_url):
+        return JsonResponse({'error': 'too large'}, status=400)
+    if job is not None:
+        job.delete()   # a failed or hollow earlier run: the new one starts at the back of the queue
+    job = TranscriptionJob.objects.create(episode=ep, state=TranscriptionJob.QUEUED, want_article=article)
+    transcribe.kick()
+    return JsonResponse(job_state(job))
+
+
+def _wishes(user):
+    """The user's jobs whose transcript is there and whose article is wanted but not written."""
+    return (TranscriptionJob.objects
+            .filter(episode__feed__user=user, state=TranscriptionJob.DONE, want_article=True)
+            .exclude(episode__in=EpisodeArticle.objects.values('episode')))
+
+
+@csrf_exempt
+@require_http_methods(['GET'])
+def transcription_pending(request):
+    """The open wishes for an article that no device holds right now."""
+    from . import transcribe
+    if (err := _auth(request)):
+        return err
+    transcribe.kick()
+    jobs = (_wishes(request.user).exclude(claimed_until__gt=timezone.now())
+            .select_related('episode__feed').order_by('id'))
+    return JsonResponse([{
+        'episode_id': j.episode_id, 'title': j.episode.title, 'podcast': j.episode.feed.title,
+        'duration_seconds': j.episode.duration_seconds,
+    } for j in jobs], safe=False)
+
+
+@csrf_exempt
+@require_http_methods(['POST', 'DELETE'])
+def article_claim(request, pk):
+    """Reserve the wish for an article for the calling device for CLAIM_MINUTES (POST), or give
+    it up (DELETE). The device names itself with `device` in the JSON body or the
+    X-Diora-Device header. 409 when another device holds it."""
+    from datetime import timedelta
+    if (err := _auth(request)):
+        return err
+    body = _body(request)
+    if body is None:
+        return JsonResponse({'error': 'invalid JSON'}, status=400)
+    device = body.get('device') or request.META.get('HTTP_X_DIORA_DEVICE') or ''
+    if not isinstance(device, str) or not device.strip():
+        return JsonResponse({'error': 'device missing'}, status=400)
+    device = device.strip()[:60]
+    job = _wishes(request.user).filter(episode_id=pk).first()
+    if job is None:
+        return JsonResponse({'error': 'no open wish'}, status=404)
+    now = timezone.now()
+    if job.claimed_until and job.claimed_until > now and job.claimed_by != device:
+        return JsonResponse({'error': 'claimed', 'claimed_until': job.claimed_until.isoformat()}, status=409)
+    if request.method == 'DELETE':
+        job.claimed_by, job.claimed_until = '', None
+    else:
+        job.claimed_by, job.claimed_until = device, now + timedelta(minutes=CLAIM_MINUTES)
+    job.save(update_fields=['claimed_by', 'claimed_until'])
+    return JsonResponse({'claimed_by': job.claimed_by,
+                         'claimed_until': job.claimed_until.isoformat() if job.claimed_until else None})

podcasts/urls.py

@@ -1,5 +1,5 @@
 from django.urls import path
-from . import audiobooks, discover, folders, radio, views
+from . import article, audiobooks, discover, folders, radio, transcription, views
 
 urlpatterns = [
     path('search/', views.podcast_search, name='podcast_search'),
@@ -28,9 +28,15 @@ urlpatterns = [
     path('episodes/discard/', views.discard_episodes, name='podcast_discard_episodes'),
     path('episodes/<int:pk>/', views.episode_detail, name='podcast_episode_detail'),
     path('episodes/<int:pk>/transcript/', views.episode_transcript, name='podcast_episode_transcript'),
+    path('episodes/<int:pk>/article/', article.episode_article, name='podcast_episode_article'),
     path('episodes/<int:pk>/silences/', views.episode_silences, name='podcast_episode_silences'),
     path('episodes/<int:pk>/sponsor/', views.episode_sponsor, name='podcast_episode_sponsor'),
     path('episodes/<int:pk>/favorite/', views.set_favorite, name='podcast_set_favorite'),
+    path('transcription/settings/', transcription.transcription_settings, name='podcast_transcription_settings'),
+    path('transcription/test/', transcription.transcription_test, name='podcast_transcription_test'),
+    path('transcription/pending/', transcription.transcription_pending, name='podcast_transcription_pending'),
+    path('episodes/<int:pk>/transcribe/', transcription.episode_transcribe, name='podcast_episode_transcribe'),
+    path('episodes/<int:pk>/article/claim/', transcription.article_claim, name='podcast_article_claim'),
     path('stats/', views.podcast_stats, name='podcast_stats'),
     path('queue/', views.queue_get, name='podcast_queue_get'),
     path('queue/add/', views.queue_add, name='podcast_queue_add'),

podcasts/views.py

@@ -29,11 +29,12 @@ from . import chapters as chapters_mod
 from . import silence as silence_mod
 from . import sponsor as sponsor_mod
 from . import feedext
+from . import transcription as transcription_mod
 from . import transcripts as transcripts_mod
 from . import youtube as youtube_mod
 from . import folders as folders_mod
 from .models import (
-    EpisodeProgress, EpisodeTranscript, PlayerDevice, PlayerState, PodcastEpisode, PodcastFeed, PodcastLater,
+    EpisodeArticle, EpisodeProgress, EpisodeTranscript, PlayerDevice, PlayerState, PodcastEpisode, PodcastFeed, PodcastLater,
     PodcastQueue, PodcastRadioHeard, MAX_SPEED, MIN_SPEED, PodcastSettings, counts_as_played,
 )
 from .shownotes import extract_timestamps, sanitize_html
@@ -231,7 +232,7 @@ def _later_data(user):
         .order_by(F('episode__pub_date').desc(nulls_last=True), '-id')
         .values(
             'id', 'added_at', 'auto',
-            'episode__id', 'episode__title', 'episode__audio_url',
+            'episode__id', 'episode__title', 'episode__audio_url', 'episode__link',
             'episode__duration_seconds', 'episode__artwork_url',
             'episode__feed__id', 'episode__feed__title', 'episode__feed__artwork_url',
             'episode__pub_date',
@@ -383,7 +384,7 @@ def _apply_extras(episode, description, chapters_url, psc_chapters, transcripts,
     return bool(changed)
 
 
-EXTRA_FIELDS = ['description', 'transcripts', 'chapters_url', 'chapters', 'chapters_fetched_at',
+EXTRA_FIELDS = ['link', 'description', 'transcripts', 'chapters_url', 'chapters', 'chapters_fetched_at',
                 'title', 'pub_date', 'episode_number']
 
 
@@ -642,6 +643,12 @@ def _move_to_permanent_url(feed_obj, new_url):
     feed_obj.rss_url = new_url
 
 
+def _entry_link(entry):
+    """The item's <link>, the episode's own page; '' unless it is an http(s) address."""
+    link = (entry.get('link') or '').strip()
+    return link[:1000] if re.match(r'^https?://', link, re.I) else ''
+
+
 def _entry_media_url(entry):
     """The episode's file: its audio enclosure, else a video one (the players
     play only its sound). '' when there is neither."""
@@ -713,6 +720,7 @@ def _apply_feed(feed_obj, fetched, full=False):
         audio_url = _entry_media_url(entry)
         if not audio_url:
             continue
+        link = _entry_link(entry)
 
         guid = entry.get('id') or entry.get('guid') or audio_url
         guid = str(guid)[:1000]
@@ -737,7 +745,7 @@ def _apply_feed(feed_obj, fetched, full=False):
 
         known = existing.get(guid)
         if known is not None:
-            renamed = {'title': title, 'episode_number': ep_number}
+            renamed = {'title': title, 'episode_number': ep_number, 'link': link}
             if pub_date is not None:
                 renamed['pub_date'] = pub_date
             if _apply_extras(known, description, chapters_url, psc_chapters, transcripts, renamed):
@@ -762,6 +770,7 @@ def _apply_feed(feed_obj, fetched, full=False):
                 'title': title,
                 'description': description,
                 'audio_url': audio_url[:1000],
+                'link': link,
                 'duration_seconds': duration_seconds,
                 'pub_date': pub_date,
                 'episode_number': ep_number,
@@ -1324,7 +1333,7 @@ def feed_episodes(request, pk):
 
     episodes = list(
         feed.episodes.values(
-            'id', 'title', 'description', 'audio_url', 'duration_seconds',
+            'id', 'title', 'description', 'audio_url', 'link', 'duration_seconds',
             'pub_date', 'artwork_url', 'episode_number', 'season_number',
         )
     )
@@ -1424,6 +1433,9 @@ def episode_detail(request, pk):
         chapters = _load_chapters(ep)
 
     progress = EpisodeProgress.objects.filter(user=request.user, episode=ep).first()
+    job = transcription_mod.job_of(ep)
+    origin = _transcript_origin(ep, job)
+    article = EpisodeArticle.objects.filter(episode=ep).first()
     return JsonResponse({
         'id': ep.id,
         'title': ep.title,
@@ -1431,9 +1443,13 @@ def episode_detail(request, pk):
         'timestamps': extract_timestamps(ep.description, ep.duration_seconds),
         'chapters': chapters,
         'transcripts': ep.transcripts,
-        'has_transcript': _has_transcript(ep),
+        'has_transcript': origin is not None,
+        'transcript_origin': origin,
+        'article_item_id': article.item_id if article else None,
+        'transcription': transcription_mod.job_state(job) if job else None,
         'feed': {'id': ep.feed.id, 'title': ep.feed.title, 'artwork_url': ep.feed.artwork_url},
         'audio_url': ep.audio_url,
+        'link': ep.link,
         'artwork_url': ep.artwork_url,
         'duration_seconds': ep.duration_seconds,
         'pub_date': ep.pub_date.isoformat() if ep.pub_date else None,
@@ -1461,6 +1477,13 @@ def _has_transcript(ep):
     return not (cached and cached[0] == sources and cached[1])
 
 
+def _transcript_origin(ep, job=None):
+    """"feed" when the feed has a usable transcript, else "whisper" when we made one, else None."""
+    if _has_transcript(ep):
+        return 'feed'
+    return 'whisper' if job is not None and job.state == 'done' and job.data else None
+
+
 @csrf_exempt
 @require_http_methods(['GET'])
 def episode_silences(request, pk):
@@ -1503,20 +1526,28 @@ def episode_sponsor(request, pk):
 def episode_transcript(request, pk):
     """The episode's transcript as {"segments": [{start, end, speaker, text}],
     "source": {url, type, language}}, fetched and parsed on first use, then
-    kept. 404 {"error": "no transcript" | "unsupported" | "invalid"} when
+    kept. When the feed has none we can use, the one Whisper made (a finished
-    there's none to show; 502 {"error": "unavailable"} when it couldn't be
+    TranscriptionJob, docs/transcription.md). 404 {"error": "no transcript" |
-    loaded right now."""
+    "unsupported" | "invalid"} when there's none to show; 502
+    {"error": "unavailable"} when it couldn't be loaded right now."""
     if not request.user.is_authenticated:
         return JsonResponse({'error': 'authentication required'}, status=401)
     try:
         ep = PodcastEpisode.objects.get(pk=pk, feed__user=request.user)
     except PodcastEpisode.DoesNotExist:
         return JsonResponse({'error': 'not found'}, status=404)
+
+    def fallback(error, status=404):
+        job = transcription_mod.job_of(ep)
+        if job is not None and job.state == 'done' and job.data:
+            return JsonResponse(job.data)
+        return JsonResponse({'error': error}, status=status)
+
     if not ep.transcripts:
-        return JsonResponse({'error': 'no transcript'}, status=404)
+        return fallback('no transcript')
     sources = _transcript_sources(ep)
     if not sources:
-        return JsonResponse({'error': 'unsupported'}, status=404)
+        return fallback('unsupported')
 
     cached = EpisodeTranscript.objects.filter(episode=ep).first()
     if cached is not None and cached.sources == sources:
@@ -1527,7 +1558,7 @@ def episode_transcript(request, pk):
             error = ''
         except chapters_mod.FetchUnavailable as e:
             log.info('transcript of episode %s not available: %s', ep.id, e)
-            return JsonResponse({'error': 'unavailable'}, status=502)
+            return fallback('unavailable', 502)
         except transcripts_mod.Unsupported as e:
             log.info('transcript of episode %s unsupported: %s', ep.id, e)
             data, error, final = None, 'unsupported', True
@@ -1539,7 +1570,7 @@ def episode_transcript(request, pk):
                 episode=ep, defaults={'sources': sources, 'data': data, 'error': error},
             )
     if error:
-        return JsonResponse({'error': error}, status=404)
+        return fallback(error)
     return JsonResponse(data)
 
 
@@ -1645,7 +1676,7 @@ EPISODE_IDS_MAX = 1000
 EPISODE_LIST_LIMIT = 100
 EPISODE_LIST_MAX = 500
 EPISODE_LIST_FIELDS = (
-    'id', 'title', 'audio_url', 'duration_seconds', 'pub_date', 'artwork_url',
+    'id', 'title', 'audio_url', 'link', 'duration_seconds', 'pub_date', 'artwork_url',
     'feed__id', 'feed__title', 'feed__artwork_url',
     'position_seconds', 'played', 'favorite', 'in_queue', 'in_later', 'last_played_at', 'discovered_at',
 )
@@ -2010,7 +2041,7 @@ def _queue_data(user):
         .select_related('episode__feed')
         .values(
             'id', 'position',
-            'episode__id', 'episode__title', 'episode__audio_url',
+            'episode__id', 'episode__title', 'episode__audio_url', 'episode__link',
             'episode__duration_seconds', 'episode__artwork_url',
             'episode__feed__id', 'episode__feed__title', 'episode__feed__artwork_url',
             'episode__pub_date',
@@ -2506,7 +2537,7 @@ def inbox(request):
 
 
 def _inbox_fields(with_description=True):
-    fields = ['id', 'title', 'audio_url', 'duration_seconds', 'pub_date', 'artwork_url',
+    fields = ['id', 'title', 'audio_url', 'link', 'duration_seconds', 'pub_date', 'artwork_url',
               'feed__id', 'feed__title', 'feed__artwork_url']
     if with_description:
         fields.insert(2, 'description')

static/css/app.css

@@ -734,6 +734,19 @@ html[data-theme="light"] .form-errors { background: #fdecee; border-color: #e7a3
   font-size: 0.75rem;
 }
 
+.auth-alt { margin-top: 1rem; }
+.auth-or {
+  display: flex; align-items: center; gap: 0.6rem;
+  margin: 0 0 1rem; font-size: 0.8rem; color: var(--text-muted);
+}
+.auth-or::before, .auth-or::after { content: ''; flex: 1; border-top: 1px solid var(--border, #333); }
+.auth-alt .btn-full { display: block; text-align: center; text-decoration: none; }
+
+/* Unlocking the books on a device without the key (js/app/33-unlock.js) */
+.unlock-code { font: 600 1.6rem/1.2 ui-monospace, monospace; letter-spacing: 0.15em; text-align: center; margin: 4px 0 14px; }
+.unlock-or { font-size: 0.85rem; color: var(--text-muted); margin: 0 0 8px; }
+.unlock-status { font-size: 0.85rem; color: var(--text-muted); min-height: 1.2em; margin: 0 0 10px; }
+
 .auth-switch {
   margin-top: 1.25rem;
   font-size: 0.85rem;
@@ -1792,14 +1805,25 @@ body.dnd-mode .timer-display {
 .pdf-page-inner { position:relative; display:inline-block; line-height:0; }
 
 /* PDF text layer — sits flush over the canvas inside pdf-page-inner */
-.pdf-text-layer { position:absolute; top:0; left:0; width:100%; height:100%; overflow:hidden; pointer-events:none; }
+.pdf-text-layer { position:absolute; top:0; left:0; width:100%; height:100%; overflow:hidden; pointer-events:none; line-height:1; -webkit-text-size-adjust:none; text-size-adjust:none; z-index:1; }
-.pdf-text-layer span { position:absolute; color:transparent !important; background:transparent; white-space:pre; cursor:text; pointer-events:auto; user-select:text; -webkit-user-select:text; line-height:1; }
+.pdf-text-layer span, .pdf-text-layer br { position:absolute; color:transparent !important; background:transparent; white-space:pre; cursor:text; pointer-events:auto; user-select:text; -webkit-user-select:text; line-height:1; transform-origin:0 0; }
-.pdf-text-layer span::selection { background:rgba(0,120,255,0.3); }
+.pdf-text-layer span::selection, .pdf-text-layer mark::selection { background:rgba(0,120,255,0.3); }
-.pdf-text-layer span.reader-search-match { background:rgba(241,196,15,.5); }
+.pdf-text-layer br::selection { background:transparent; }
-.pdf-text-layer span.reader-search-match.active { background:rgba(230,57,70,.6); }
+.pdf-text-layer mark.reader-search-match { color:transparent; background:rgba(241,196,15,.5); padding:0; border-radius:2px; }
+.pdf-text-layer mark.reader-search-match.active { background:rgba(230,57,70,.6); }
+/* PDF links — over the text, so a tap on one follows it */
+.pdf-link-layer { position:absolute; top:0; left:0; width:100%; height:100%; pointer-events:none; z-index:2; }
+.pdf-link { position:absolute; display:block; pointer-events:auto; border-radius:2px; }
+.pdf-link:hover, .pdf-link:focus-visible { background:rgba(0,120,255,0.15); outline:none; }
+.pdf-back-btn { position:absolute; left:50%; bottom:calc(var(--bar-h, 0px) + 24px); transform:translateX(-50%); z-index:20; box-shadow:0 2px 8px rgba(0,0,0,.25); }
 
 /* PDF invert */
-#reader-overlay.pdf-inverted .pdf-page { filter:invert(1); }
+/* Dark pages: inverted, and the hues turned back, so photos and coloured figures aren't negatives. */
+#reader-overlay.pdf-inverted .pdf-page { filter:invert(1) hue-rotate(180deg); }
+/* Leserfarben (58-pdf.js _pdfColorFilter): the page recoloured, its pictures over it as printed */
+.pdf-page.pdf-recolored { filter:url(#pdf-reader-colors); }
+.pdf-image-layer { position:absolute; top:0; left:0; width:100%; height:100%; pointer-events:none; }
+.pdf-image-layer canvas { position:absolute; display:block; }
 
 /* ---- Immersive reader mode ---- */
 .navbar         { transition: transform 0.35s ease; }
@@ -1824,8 +1848,17 @@ html.nui-tts-on body.reader-immersive .now-playing-bar { transform: none; }
 html.nui-tts-on body.reader-immersive .reader-overlay  { bottom: var(--bar-h); }
 
 /* PDF paginated */
-.reader-content.pdf-paginated { overflow:hidden !important; display:flex; align-items:center; justify-content:center; }
+/* Page by page the page fills the whole reading area, without its padding: 66-pdf-paginated.js
+   measures the area and places the page by its text box itself. */
+.reader-content.pdf-paginated { overflow:hidden !important; display:block; padding:0 !important; }
+/* The pages are as wide as they're drawn, not the text column's line length (--reader-max-width);
+   zoomed wider than the screen, scrolling reaches their left edge too (not centred past it). */
+.reader-content > #pdf-viewport { max-width:none; }
+.reader-content:not(.pdf-paginated) > #pdf-viewport { width:max-content; min-width:100%; }
+.reader-content.pdf-paginated > #pdf-viewport { width:100%; height:100%; max-width:none; margin:0; }
 .pdf-paginated .pdf-page-wrapper { margin:0; }
+.pdf-paginated .pdf-page-wrapper { position:relative; }
+.pdf-paginated .pdf-page-inner { position:absolute; left:0; top:0; }
 
 /* PDF loading overlay */
 .pdf-loading-overlay {
@@ -1866,6 +1899,57 @@ html.nui-tts-on body.reader-immersive .reader-overlay  { bottom: var(--bar-h); }
 .footnote-popover.reader-theme-sepia  { background:#ede0c4; color:#3b2a1a; border-color:#c8b89a; }
 .footnote-popover.reader-theme-bright { background:#f0f0f0; color:#111;    border-color:#bbb; }
 .footnote-popover-close { position:absolute; top:4px; right:6px; background:none; border:none; color:inherit; opacity:0.5; cursor:pointer; font-size:14px; padding:2px 4px; line-height:1; }
+.footnote-popover > :first-child { margin-top:0; }
+.footnote-popover > :last-child { margin-bottom:0; }
+
+/* ---- Reading in pages (63-pages.js) ----
+   Only the current chapter is laid out; the others stay in the DOM (block
+   indices, search, notes, anchors). No free scrolling: the pager moves. */
+#reader-content.nui-window > [data-epub-src]:not(.nui-ch-current) { display:none; }
+/* Room below the chapter, so its last page can show just the remaining text
+   (then the page's colour) instead of scrolling back over lines already read. */
+#reader-content.nui-window::after { content:""; display:block; height:100vh; }
+#reader-content.nui-window { overflow-y:hidden; -ms-overflow-style:none; scrollbar-width:none; }
+#reader-content.nui-window::-webkit-scrollbar { display:none; }
+/* Masks: the line cut at the page's bottom and whatever sits in the top gap. */
+.nui-page-mask { position:fixed; z-index:250; pointer-events:none; display:none; }
+/* Footnotes at the foot of the page, like a printed book. */
+.nui-page-notes { position:fixed; z-index:255; overflow:hidden; padding:0 0 4px; box-sizing:border-box;
+  line-height:1.4; font-weight:400; cursor:pointer; }
+.nui-pn-rule { width:30%; border-top:1px solid currentColor; margin:2px 0 6px; }
+.nui-pn-item { margin:0 0 4px; padding-left:1.6em; text-indent:-1.6em; }
+.nui-pn-item p, .nui-pn-item div, .nui-pn-item aside, .nui-pn-item li { display:inline; margin:0; padding:0; text-indent:0; }
+.nui-pn-item p + p::before { content:" "; }
+.nui-pn-label { display:inline-block; min-width:1.6em; text-indent:0; font-weight:700; }
+.nui-pn-item a { color:inherit; text-decoration:none; }
+.nui-pn-item img { display:none; }
+/* Cut off: say so, a tap shows the whole note. */
+.nui-page-notes.nui-pn-cut::after { content:"… mehr"; position:absolute; right:0; bottom:2px; padding-left:8px;
+  background:inherit; font-weight:700; }
+/* Two pages side by side (a spread, 63-pages.js): the right one is a copy of
+   the chapter, scrolled to the page after the left one. */
+#reader-spread-page { display:none; }
+#reader-overlay.reader-spread #reader-content,
+#reader-overlay.reader-spread #reader-spread-page { flex:1 1 0; min-width:0; }
+#reader-overlay.reader-spread #reader-spread-page { display:block; overflow:hidden; box-shadow:inset 1px 0 0 var(--border);
+  user-select:none; -webkit-user-select:none; cursor:default; }
+#reader-spread-page::after { content:""; display:block; height:100vh; }
+/* An article in the reader (65-article.js): what only books have stays out. */
+#reader-overlay.reader-article-mode .reader-progress-wrap,
+#reader-overlay.reader-article-mode #reader-bookmark-btn,
+#reader-overlay.reader-article-mode #reader-bm-list-btn,
+#reader-overlay.reader-article-mode #reader-toc-btn,
+#reader-overlay.reader-article-mode #reader-margin-btn,
+#reader-overlay.reader-article-mode #reader-reset-pos-btn,
+#reader-overlay.reader-article-mode .marker-mode-btn,
+#reader-overlay.reader-article-mode #reader-margin,
+#reader-overlay.reader-article-mode .nui-toc { display:none !important; }
+.reader-article-actions { display:inline-flex; gap:10px; align-items:center; }
+/* The browser's line under the page (the apps have their own footer). */
+#reader-page-footer { display:flex; justify-content:space-between; gap:12px; flex-shrink:0; padding:4px 16px 6px;
+  font-size:12px; color:var(--muted,#888); font-variant-numeric:tabular-nums; }
+.reader-theme-sepia #reader-page-footer { background:#f5e6c8; color:#7a6650; }
+.reader-theme-bright #reader-page-footer { background:#fff; color:#777; }
 
 /* Page numbers of the printed edition (epub:type="pagebreak") */
 .reader-content .epub-pagenum { display:inline-block; font-size:0.62em; line-height:1.3; font-weight:400; font-style:normal; letter-spacing:0; text-transform:none; vertical-align:0.35em; padding:0 0.35em; margin-right:0.4em; border:1px solid currentColor; border-radius:3px; opacity:0.45; user-select:none; }

static/js/ai.js

@@ -7,7 +7,7 @@
 // server sees neither the key nor what is sent; Anthropic sees what a feature needs (titles,
 // authors, folder names …). Nothing is changed without the user ticking it in review().
 //
-// window.DioraAI = {available(), hasKey(), settings(), ask({system, prompt, schema}), review(…)}
+// window.DioraAI = {available(), hasKey(), settings(), ask({system, prompt, schema, effort, maxTokens, onProgress, signal}), review(…)}
 // ---------------------------------------------------------------------------
 
 import Anthropic from '/static/vendor/anthropic-sdk-0.133.0.mjs';
@@ -116,12 +116,17 @@ function parseAnswer(text) {
   }
 }
 
+// Words in a text so far (for onProgress).
+const countWords = t => { const m = String(t || '').match(/\S+/g); return m ? m.length : 0; };
+
 // One request with a JSON answer of `schema` (structured outputs).
-async function ask({system, prompt, schema, effort = 'low', maxTokens = 64000}) {
+// Optional: onProgress(words) while the answer is written (throttled; OpenRouter is not streamed,
+// it reports once at the end), signal (an AbortSignal) to give up.
+async function ask({system, prompt, schema, effort = 'low', maxTokens = 64000, onProgress, signal}) {
   const v = await loadVault();
   const key = keyOf(v);
   if (!key) throw fail('Kein API-Key hinterlegt.');
-  if (providerOf(v) === 'openrouter') return askOpenRouter(key, {system, prompt, schema, effort, maxTokens});
+  if (providerOf(v) === 'openrouter') return askOpenRouter(key, {system, prompt, schema, effort, maxTokens, onProgress, signal});
   // Anthropic: streamed (long lists), a declined request re-runs on Anthropic's fallback model.
   const stream = client(key).beta.messages.stream({
     model: PROVIDERS.anthropic.model,
@@ -131,8 +136,18 @@ async function ask({system, prompt, schema, effort = 'low', maxTokens = 64000})
     system,
     messages: [{role: 'user', content: prompt}],
     output_config: {effort, format: {type: 'json_schema', schema}},
+  }, signal ? {signal} : undefined);
+  if (onProgress) {
+    let last = 0;
+    stream.on('text', (delta, snapshot) => {
+      const now = Date.now();
+      if (now - last < 400) return;
+      last = now;
+      try { onProgress(countWords(snapshot)); } catch (e) { /* the caller's display must not break the call */ }
     });
+  }
   const msg = await stream.finalMessage();
+  if (onProgress) { try { onProgress(countWords(msg.content.filter(b => b.type === 'text').map(b => b.text).join(''))); } catch (e) { /* see above */ } }
   if (msg.stop_reason === 'refusal') throw fail('Claude hat diese Anfrage abgelehnt.');
   if (msg.stop_reason === 'max_tokens') throw fail('Die Antwort wurde zu lang – bitte einen kleineren Teil ordnen.');
   return parseAnswer(msg.content.filter(b => b.type === 'text').map(b => b.text).join(''));
@@ -149,8 +164,8 @@ async function openRouterFetch(key, path, init) {
   return r.json();
 }
 
-async function askOpenRouter(key, {system, prompt, schema, effort, maxTokens}) {
+async function askOpenRouter(key, {system, prompt, schema, effort, maxTokens, onProgress, signal}) {
-  const d = await openRouterFetch(key, '/chat/completions', {method: 'POST', body: JSON.stringify({
+  const d = await openRouterFetch(key, '/chat/completions', {method: 'POST', signal, body: JSON.stringify({
     model: PROVIDERS.openrouter.model,
     max_tokens: maxTokens,
     reasoning: {effort},
@@ -161,7 +176,9 @@ async function askOpenRouter(key, {system, prompt, schema, effort, maxTokens}) {
   if (!choice) throw fail('Die Antwort ließ sich nicht lesen.');
   if (choice.message && choice.message.refusal) throw fail('Claude hat diese Anfrage abgelehnt.');
   if (choice.finish_reason === 'length') throw fail('Die Antwort wurde zu lang – bitte einen kleineren Teil ordnen.');
-  return parseAnswer((choice.message && choice.message.content) || '');
+  const text = (choice.message && choice.message.content) || '';
+  if (onProgress) { try { onProgress(countWords(text)); } catch (e) { /* the caller's display must not break the call */ } }
+  return parseAnswer(text);
 }
 
 // Is this key accepted? (Costs nothing: Anthropic's model info, OpenRouter's key info.)
@@ -361,6 +378,8 @@ window.__dioraBack = function () {
   return back ? back.apply(this, arguments) : false;
 };
 
-window.DioraAI = {available, hasKey: async () => !!(await loadKey()), settings, withKey, ask, review, progress, showError,
+const modelName = () => PROVIDERS[providerOf(vault)].model;
+
+window.DioraAI = {available, hasKey: async () => !!(await loadKey()), modelName, settings, withKey, ask, review, progress, showError,
   close, toast, explain};
 window.dispatchEvent(new Event('diora-ai'));

static/js/app/26-sidebar.js

@@ -37,9 +37,12 @@ document.addEventListener('keydown', e => {
       if (e.key === 'ArrowDown') { e.preventDefault(); readerSearchNext(); }
       if (e.key === 'ArrowUp')   { e.preventDefault(); readerSearchPrev(); }
     }
-    if (readerSettings.pdfPaginated && currentPdfDoc) {
+    // PDF page by page: arrows, Space, PageUp/PageDown turn (scrolling has its own keys below).
-      if (e.key === 'ArrowRight') { e.preventDefault(); pdfGoToPage(pdfCurrentPage + 1); }
+    if (readerSettings.pdfPaginated && currentPdfDoc && !readerSearchOpen
-      if (e.key === 'ArrowLeft')  { e.preventDefault(); pdfGoToPage(pdfCurrentPage - 1); }
+        && !/^(INPUT|TEXTAREA|SELECT)$/.test((document.activeElement && document.activeElement.tagName) || '')) {
+      const turn = {ArrowRight: 1, ArrowDown: 1, PageDown: 1, ' ': e.shiftKey ? -1 : 1,
+                    ArrowLeft: -1, ArrowUp: -1, PageUp: -1}[e.key];
+      if (turn) { e.preventDefault(); pdfGoToPage(pdfCurrentPage + turn); return; }
     }
 
     // Vim-style scroll — ignore when typing in an input

static/js/app/32-crypto.js

@@ -41,10 +41,19 @@ async function getOrCreateEncKey() {
       return _encKey;
     } catch (e) { /* fall through, generate new */ }
   }
-  // No key found — generate one and store it
+  // Just signed in with the password: the login form left the derived key (99-init.js
-  _encKey = await crypto.subtle.generateKey({name: 'AES-GCM', length: 256}, true, ['encrypt', 'decrypt']);
+  // moves it too, but this may run first).
-  const raw = await crypto.subtle.exportKey('raw', _encKey);
+  const pending = window.USER_ID ? localStorage.getItem('diora_pending_enc_key') : null;
-  localStorage.setItem(storageKey, bytesToBase64(new Uint8Array(raw)));
+  if (pending) {
+    localStorage.setItem(storageKey, pending);
+    localStorage.removeItem('diora_pending_enc_key');
+    return getOrCreateEncKey();
+  }
+  // No key on this device (signed in through Authelia, or the storage was cleared): never
+  // make a new one — that would lock the existing books away. Ask for the real one
+  // (33-unlock.js); throws DioraLocked when the user puts it off.
+  const raw = await dioraUnlock();
+  _encKey = await crypto.subtle.importKey('raw', raw, {name: 'AES-GCM'}, false, ['encrypt', 'decrypt']);
   return _encKey;
 }
 

static/js/app/33-unlock.js

@@ -0,0 +1,254 @@
+'use strict';
+
+// ---------------------------------------------------------------------------
+// Unlocking the books on a device without the key (accounts/keyshare.py)
+//
+// A password login derives the book key on the device (login.html). A device signed in
+// another way (Authelia, accounts/oidc.py) has none: getOrCreateEncKey() then asks here
+// instead of making a new key. Two ways in:
+//  - an unlocked device of the account hands it over: this device posts an ECDH P-256
+//    public key, the other one shows the same 6-digit code (made from that key), the user
+//    approves there and the book key comes back encrypted for this device only;
+//  - or the diora password, the same derivation as at login.
+// The server only ever sees public keys and ciphertext.
+// ---------------------------------------------------------------------------
+
+const KEYSHARE_INFO = new TextEncoder().encode('diora keyshare v1');
+let _unlockPromise = null;
+let _unlockDismissed = false;
+
+function dioraLockedError() {
+  const e = new Error('Bücher sind auf diesem Gerät noch nicht entsperrt');
+  e.name = 'DioraLocked';
+  return e;
+}
+
+function _csrf() {
+  return document.cookie.split('; ').find(r => r.startsWith('csrftoken='))?.split('=')[1] || '';
+}
+
+async function _ksFetch(url, body) {
+  const opts = body === undefined ? {credentials: 'same-origin'} : {
+    method: 'POST', credentials: 'same-origin',
+    headers: {'Content-Type': 'application/json', 'X-CSRFToken': _csrf()},
+    body: JSON.stringify(body),
+  };
+  const res = await fetch(url, opts);
+  if (!res.ok) throw new Error('HTTP ' + res.status);
+  return res.json();
+}
+
+// 6-digit code from the asking device's public key, shown on both devices.
+async function keyshareCode(pubB64) {
+  const h = new Uint8Array(await crypto.subtle.digest('SHA-256', base64ToBytes(pubB64)));
+  const n = ((h[0] << 24) | (h[1] << 16) | (h[2] << 8) | h[3]) >>> 0;
+  const s = String(n % 1000000).padStart(6, '0');
+  return s.slice(0, 3) + ' ' + s.slice(3);
+}
+
+async function _sharedAesKey(privateKey, otherPubB64) {
+  const otherPub = await crypto.subtle.importKey('raw', base64ToBytes(otherPubB64),
+    {name: 'ECDH', namedCurve: 'P-256'}, false, []);
+  const bits = await crypto.subtle.deriveBits({name: 'ECDH', public: otherPub}, privateKey, 256);
+  const hk = await crypto.subtle.importKey('raw', bits, 'HKDF', false, ['deriveKey']);
+  return crypto.subtle.deriveKey({name: 'HKDF', hash: 'SHA-256', salt: new Uint8Array(0), info: KEYSHARE_INFO},
+    hk, {name: 'AES-GCM', length: 256}, false, ['encrypt', 'decrypt']);
+}
+
+function _deviceName() {
+  const ua = navigator.userAgent || '';
+  const app = /diora-desktop|Electron/i.test(ua) ? 'diora Desktop'
+    : /; wv\)|diora/i.test(ua) && /Android/i.test(ua) ? 'diora App'
+    : /Firefox\//.test(ua) ? 'Firefox' : /Edg\//.test(ua) ? 'Edge'
+    : /Chrome\//.test(ua) ? 'Chrome' : /Safari\//.test(ua) ? 'Safari' : 'Browser';
+  const os = /Android/i.test(ua) ? 'Android' : /iPhone|iPad/i.test(ua) ? 'iOS'
+    : /Windows/i.test(ua) ? 'Windows' : /Mac OS/i.test(ua) ? 'macOS' : /Linux/i.test(ua) ? 'Linux' : '';
+  return os ? `${app} auf ${os}` : app;
+}
+
+// Is this raw key the account's book key? Tried on one book's metadata; without books
+// there is nothing to compare with, so it is taken.
+async function _keyFits(raw) {
+  let books = [];
+  try { books = await _ksFetch('/books/'); } catch (e) { return true; }
+  const b = Array.isArray(books) ? books.find(x => x && x.meta_ct && x.meta_iv) : null;
+  if (!b) return true;
+  try {
+    const key = await crypto.subtle.importKey('raw', raw, {name: 'AES-GCM'}, false, ['decrypt']);
+    await decryptBytes(key, b.meta_iv, b.meta_ct);
+    return true;
+  } catch (e) { return false; }
+}
+
+function _storeRawKey(raw) {
+  localStorage.setItem(`diora_enc_key_${window.USER_ID || 0}`, bytesToBase64(new Uint8Array(raw)));
+}
+
+function _el(tag, attrs, text) {
+  const el = document.createElement(tag);
+  for (const k in (attrs || {})) el.setAttribute(k, attrs[k]);
+  if (text) el.textContent = text;
+  return el;
+}
+
+function _openDialog(build) {
+  const overlay = _el('div', {class: 'modal-overlay'});
+  const dialog = _el('div', {class: 'modal-dialog', role: 'dialog', 'aria-modal': 'true'});
+  document.body.append(overlay, dialog);
+  const close = () => { overlay.remove(); dialog.remove(); };
+  build(dialog, close);
+  return close;
+}
+
+// Ask for the key. Resolves with the raw key (already stored), rejects with DioraLocked.
+function dioraUnlock(opts) {
+  opts = opts || {};
+  if (_unlockPromise) return _unlockPromise;
+  if (_unlockDismissed && !opts.force) return Promise.reject(dioraLockedError());
+  _unlockPromise = new Promise((resolve, reject) => {
+    let stopped = false;
+    let pair = null, token = null;
+    const finish = (raw) => {
+      stopped = true; close(); _unlockPromise = null; _unlockDismissed = false;
+      _storeRawKey(raw);
+      resolve(raw);
+      document.dispatchEvent(new CustomEvent('diora:unlocked'));
+    };
+    const close = _openDialog((d, closeFn) => {
+      d.append(_el('h2', {style: 'margin:0 0 10px;font-size:1.1rem;'}, 'Bücher entsperren'));
+      d.append(_el('p', {class: 'modal-message'},
+        'Deine Bücher sind mit deinem diora-Passwort verschlüsselt. Öffne diora auf einem Gerät, '
+        + 'auf dem sie schon offen sind, und bestätige dort diesen Code:'));
+      const code = _el('div', {class: 'unlock-code'}, '··· ···');
+      const status = _el('p', {class: 'unlock-status'}, 'Frage an deine anderen Geräte …');
+      d.append(code, status);
+      d.append(_el('p', {class: 'unlock-or'}, 'Oder gib hier dein diora-Passwort ein:'));
+      const pw = _el('input', {type: 'password', class: 'modal-input', autocomplete: 'current-password',
+        placeholder: 'diora-Passwort'});
+      d.append(pw);
+      const actions = _el('div', {class: 'modal-actions'});
+      const later = _el('button', {class: 'btn', type: 'button'}, 'Später');
+      const ok = _el('button', {class: 'btn btn-primary', type: 'button'}, 'Entsperren');
+      actions.append(later, ok);
+      d.append(actions);
+
+      const tryPassword = async () => {
+        if (!pw.value) { pw.focus(); return; }
+        ok.disabled = true;
+        status.textContent = 'Schlüssel wird abgeleitet …';
+        try {
+          const enc = new TextEncoder();
+          const username = document.querySelector('meta[name="username"]')?.content || '';
+          const mat = await crypto.subtle.importKey('raw', enc.encode(pw.value), 'PBKDF2', false, ['deriveKey']);
+          const key = await crypto.subtle.deriveKey(
+            {name: 'PBKDF2', salt: enc.encode('diora:' + username), iterations: 200000, hash: 'SHA-256'},
+            mat, {name: 'AES-GCM', length: 256}, true, ['encrypt', 'decrypt']);
+          const raw = new Uint8Array(await crypto.subtle.exportKey('raw', key));
+          if (!(await _keyFits(raw))) { status.textContent = 'Das Passwort passt nicht zu deinen Büchern.'; ok.disabled = false; return; }
+          finish(raw);
+        } catch (e) {
+          status.textContent = 'Fehler: ' + e.message; ok.disabled = false;
+        }
+      };
+      ok.addEventListener('click', tryPassword);
+      pw.addEventListener('keydown', (e) => { if (e.key === 'Enter') tryPassword(); });
+      later.addEventListener('click', () => {
+        stopped = true; closeFn(); _unlockPromise = null; _unlockDismissed = true;
+        reject(dioraLockedError());
+      });
+
+      // Ask the other devices.
+      (async () => {
+        try {
+          pair = await crypto.subtle.generateKey({name: 'ECDH', namedCurve: 'P-256'}, false, ['deriveBits']);
+          const pub = bytesToBase64(new Uint8Array(await crypto.subtle.exportKey('raw', pair.publicKey)));
+          token = (await _ksFetch('/accounts/keyshare/', {pub, device: _deviceName()})).token;
+          code.textContent = await keyshareCode(pub);
+          status.textContent = 'Warte auf Freigabe …';
+          while (!stopped) {
+            await new Promise(r => setTimeout(r, 3000));
+            if (stopped) break;
+            let st;
+            try { st = await _ksFetch(`/accounts/keyshare/${token}/`); } catch (e) { continue; }
+            if (st.state === 'pending') continue;
+            if (st.state === 'declined') { status.textContent = 'Auf dem anderen Gerät abgelehnt.'; break; }
+            if (st.state === 'expired') { status.textContent = 'Die Anfrage ist abgelaufen. Lade die Seite neu, um es noch einmal zu versuchen.'; break; }
+            const aes = await _sharedAesKey(pair.privateKey, st.pub);
+            const raw = new Uint8Array(await crypto.subtle.decrypt({name: 'AES-GCM', iv: base64ToBytes(st.iv)}, aes, base64ToBytes(st.ct)));
+            if (raw.length !== 32 || !(await _keyFits(raw))) { status.textContent = 'Der übergebene Schlüssel passt nicht.'; break; }
+            finish(raw);
+            break;
+          }
+        } catch (e) {
+          if (!stopped) status.textContent = 'Übergabe nicht möglich (' + e.message + '). Das Passwort geht immer.';
+        }
+      })();
+    });
+  });
+  return _unlockPromise;
+}
+
+// --- The other side: an unlocked device approves new ones -------------------------------
+
+const _seenShares = new Set();
+let _approving = false;
+
+async function _approveShare(req, rawKey) {
+  const codeText = await keyshareCode(req.pub);
+  return new Promise((resolve) => {
+    _openDialog((d, close) => {
+      d.append(_el('h2', {style: 'margin:0 0 10px;font-size:1.1rem;'}, 'Neues Gerät entsperren?'));
+      d.append(_el('p', {class: 'modal-message'},
+        `„${req.device || 'Ein Gerät'}“ möchte deine Bücher öffnen. Gib nur frei, wenn dort derselbe Code steht:`));
+      d.append(_el('div', {class: 'unlock-code'}, codeText));
+      const status = _el('p', {class: 'unlock-status'});
+      d.append(status);
+      const actions = _el('div', {class: 'modal-actions'});
+      const no = _el('button', {class: 'btn', type: 'button'}, 'Ablehnen');
+      const yes = _el('button', {class: 'btn btn-primary', type: 'button'}, 'Freigeben');
+      actions.append(no, yes);
+      d.append(actions);
+      no.addEventListener('click', async () => {
+        try { await _ksFetch(`/accounts/keyshare/${req.token}/answer/`, {decline: true}); } catch (e) {}
+        close(); resolve(false);
+      });
+      yes.addEventListener('click', async () => {
+        yes.disabled = no.disabled = true;
+        try {
+          const pair = await crypto.subtle.generateKey({name: 'ECDH', namedCurve: 'P-256'}, false, ['deriveBits']);
+          const aes = await _sharedAesKey(pair.privateKey, req.pub);
+          const iv = crypto.getRandomValues(new Uint8Array(12));
+          const ct = await crypto.subtle.encrypt({name: 'AES-GCM', iv}, aes, rawKey);
+          await _ksFetch(`/accounts/keyshare/${req.token}/answer/`, {
+            pub: bytesToBase64(new Uint8Array(await crypto.subtle.exportKey('raw', pair.publicKey))),
+            iv: bytesToBase64(iv), ct: bytesToBase64(new Uint8Array(ct)),
+          });
+          close(); resolve(true);
+        } catch (e) {
+          status.textContent = 'Freigabe fehlgeschlagen: ' + e.message;
+          yes.disabled = no.disabled = false;
+        }
+      });
+    });
+  });
+}
+
+async function checkKeyShareRequests() {
+  if (_approving || !window.USER_ID || document.hidden) return;
+  const stored = localStorage.getItem(`diora_enc_key_${window.USER_ID}`);
+  if (!stored) return;
+  let data;
+  try { data = await _ksFetch('/accounts/keyshare/pending/'); } catch (e) { return; }
+  const req = (data.requests || []).find(r => !_seenShares.has(r.token));
+  if (!req) return;
+  _seenShares.add(req.token);
+  _approving = true;
+  try { await _approveShare(req, base64ToBytes(stored)); } finally { _approving = false; }
+}
+
+(function () {
+  if (!window.USER_ID || !window.crypto || !crypto.subtle) return;
+  setTimeout(checkKeyShareRequests, 2000);
+  setInterval(checkKeyShareRequests, 15000);
+  document.addEventListener('visibilitychange', () => { if (!document.hidden) checkKeyShareRequests(); });
+})();

static/js/app/40-epub.js

@@ -106,10 +106,12 @@ async function parseEpub(arrayBuffer) {
 // spine text by CHARS_PER_BOOK_PAGE (as the reader estimates it). The library's length sorts.
 async function bookPagesOfFile(buf, type) {
   if (type === 'pdf') {
-    const pdf = await pdfjsLib.getDocument({data: new Uint8Array(buf)}).promise;
+    const pdf = await _pdfLoad(new Uint8Array(buf)).promise;
-    const n = pdf.numPages;
+    try {
-    pdf.destroy();
+      return pdf.numPages;
-    return n;
+    } finally {
+      try { pdf.destroy(); } catch (e) {}
+    }
   }
   const zip = await JSZip.loadAsync(buf);
   const containerXml = await zip.file('META-INF/container.xml').async('text');
@@ -228,14 +230,41 @@ function _injectImageBlobs(html, chapterDir, imageMap) {
 // Calibre/zeno.org conversions) use neither and just wrap a plain link to
 // another chapter file's fragment in a <sup> — in either direction
 // (<sup><a>…</a></sup> or <a><sup>…</sup></a>). Recognize all of these
-// rather than only the single hardcoded class name.
+// rather than only the single hardcoded class name. Also role="doc-noteref",
+// more note classes, and a bare note number as the link's text ("3", "[12]", "*").
 function _looksLikeFootnoteLink(el, href) {
   if (!href.includes('#')) return false;
   const cls = (el.getAttribute('class') || '').toLowerCase();
   if (/\bnote|\bfootnote|\bfn\b/.test(cls)) return true;
+  if (/\b(fnref|noteref|footnote-?ref|endnote-?ref)\b/.test(cls)) return true;
   const epubType = (el.getAttribute('epub:type') || '').toLowerCase();
   if (epubType.includes('noteref')) return true;
-  return !!(el.closest('sup') || el.querySelector('sup'));
+  if ((el.getAttribute('role') || '').toLowerCase() === 'doc-noteref') return true;
+  if (el.closest('sup') || el.querySelector('sup')) return true;
+  return /^\s*[\[(]?\s*(\d{1,4}|[ivxlc]{1,6}|[a-z]|[*†‡§¶]+)\s*[\])]?\s*$/i.test(el.textContent || '');
+}
+
+// The number at the start of a note that links back to the text ("3 Endnote…")
+// is a way back, not a reference to another note.
+const _NOTE_BOX_SELECTOR = '.footnote, .endnote, aside, [role="doc-footnote"], [role="doc-endnote"], ' +
+  '[epub\\:type~="footnote"], [epub\\:type~="endnote"], [epub\\:type~="rearnote"]';
+function _isNoteBacklink(el) {
+  const role = (el.getAttribute('role') || '').toLowerCase();
+  const type = (el.getAttribute('epub:type') || '').toLowerCase();
+  if (role === 'doc-backlink' || /backlink/.test(type)) return true;
+  // Inside a note's box. Not the link itself: a reference can carry the class
+  // "footnote" too (<a class="footnote" href="#ftn.fn1">, Suhrkamp/DocBook).
+  if (el.parentElement && el.parentElement.closest(_NOTE_BOX_SELECTOR)) return true;
+  const block = el.closest('p, li, div, dd');
+  if (!block) return false;
+  // A note: the block is a link target itself, or holds an empty anchor that is
+  // (zeno.org / e-artnow: <p><a id="F258"></a><a href="…#N258">1</a> Note…</p>).
+  const anchored = block.id || el.id || Array.prototype.some.call(block.querySelectorAll('[id]'),
+    x => x !== el && !(x.textContent || '').trim());
+  if (!anchored) return false;
+  const text = (block.textContent || '').trim();
+  const firstLink = Array.prototype.find.call(block.querySelectorAll('a'), a => (a.textContent || '').trim());
+  return firstLink === el && text.startsWith((el.textContent || '').trim());
 }
 
 // Chapters are XHTML, but parsed as HTML (forgiving with broken books). HTML
@@ -272,15 +301,18 @@ function sanitizeEpubHtml(html) {
       if (href.startsWith('http://') || href.startsWith('https://')) {
         el.setAttribute('target', '_blank');
         el.setAttribute('rel', 'noopener noreferrer');
-      } else if (_looksLikeFootnoteLink(el, href)) {
+      } else if (!(href.includes('#') && _isNoteBacklink(el)) && _looksLikeFootnoteLink(el, href)) {
-        // The whole book is one concatenated DOM, so a fragment id resolves
+        // data-footnote-ref (the id) marks the link as a reference; the href
-        // via getElementById regardless of which chapter file it originally
+        // as written (data-footnote-href) finds the note in the chapter file it
-        // pointed into — strip any "../Text/other.html" path prefix and keep
+        // points into (78-footnotes.js) — ids like "fn1" repeat across files.
-        // just the id after the last '#'.
         el.dataset.footnoteRef = href.slice(href.indexOf('#') + 1);
+        el.dataset.footnoteHref = href;
         el.removeAttribute('href');
         el.style.cursor = 'pointer';
       } else {
+        // A link within the book (a cross-reference, a note's way back): kept
+        // for the apps, which follow it (diora-android native-ui.js).
+        if (href && !/^[a-z]+:/i.test(href)) el.dataset.internalHref = href;
         el.removeAttribute('href');
         el.style.cursor = 'default';
       }

static/js/app/48-book-cache.js

@@ -41,9 +41,9 @@ let _parseCodeSig = null;
 
 function _parsePreparationSig() {
   if (_parseCodeSig) return _parseCodeSig;
-  // window.__dioraPrepareSig: what an app layer adds to the preparation (the
+  // window.__dioraPrepareSig: what an app layer adds to the preparation — a
-  // footnote detection in diora-android's native-ui.js) — a change there too.
+  // change there too.
-  const src = [parseEpub, sanitizeEpubHtml, _injectImageBlobs, _resolveImageBlob, _parseEpubToc]
+  const src = [parseEpub, sanitizeEpubHtml, _looksLikeFootnoteLink, _isNoteBacklink, _injectImageBlobs, _resolveImageBlob, _parseEpubToc]
     .map(f => String(f)).join('\n') + (window.__dioraPrepareSig || '');
   let h = 5381;
   for (let i = 0; i < src.length; i++) h = ((h << 5) + h + src.charCodeAt(i)) | 0;

static/js/app/50-reader-state.js

@@ -3,7 +3,7 @@
 // Reader settings
 let readerSettings = { fontSize: 16, lineHeight: 1.8, maxWidth: 65, theme: 'dark',
                         fontFamily: 'serif', noBold: false, fontWeight: null,
-                        pdfZoom: 100, pdfInverted: false, pdfPaginated: false, pdfSpread: false };
+                        pdfZoom: 100, pdfInverted: false, pdfPageColors: false, pdfPaginated: false, pdfSpread: false };
 let readerSettingsPanelOpen = false;
 let currentPdfDoc = null;
 let currentPdfBuffer = null;
@@ -45,9 +45,6 @@ function _readerScrollTick() {
   _readerScrollLastTs = now;
   $('reader-content')?.scrollBy({top: Math.round(_readerScrollDir * READER_SCROLL_SPEED * dt / 1000)});
 }
-let _pinchStartDist = 0;
-let _pinchStartZoom = 100;
-let _isPinching = false;
 
 if (typeof pdfjsLib !== 'undefined') {
   pdfjsLib.GlobalWorkerOptions.workerSrc = '/static/js/pdf.worker.min.js';

static/js/app/52-book-list.js

@@ -41,6 +41,13 @@ async function loadBookList() {
     try {
       key = await getOrCreateEncKey();
     } catch (e) {
+      if (e.name === 'DioraLocked') {
+        listEl.innerHTML = '<p class="muted">Deine Bücher sind auf diesem Gerät noch gesperrt. <button class="btn" type="button" data-unlock-books>Entsperren</button></p>';
+        listEl.querySelector('[data-unlock-books]')?.addEventListener('click', () => {
+          dioraUnlock({force: true}).then(() => loadBookList(), () => {});
+        });
+        return;
+      }
       listEl.innerHTML = `<p class="muted">Verschlüsselung nicht verfügbar: ${e.message}. Die Seite muss über HTTPS laufen.</p>`;
       return;
     }
@@ -126,10 +133,14 @@ function repairBook(bookId) {
       let title = file.name.replace(/\.(epub|pdf)$/i, ''), author = '', isbn = '';
       try {
         if (isPdf) {
-          const pdfDoc = await pdfjsLib.getDocument({data: new Uint8Array(buf.slice(0))}).promise;
+          const pdfDoc = await _pdfLoad(new Uint8Array(buf.slice(0))).promise;
+          try {
             const meta = await pdfDoc.getMetadata();
             title = meta.info?.Title?.trim() || title;
             author = meta.info?.Author?.trim() || '';
+          } finally {
+            try { pdfDoc.destroy(); } catch (e) {}
+          }
         } else {
           const zip = await JSZip.loadAsync(buf.slice(0));
           const containerXml = await zip.file('META-INF/container.xml').async('text');
@@ -778,10 +789,14 @@ async function _importEbookBuffer(buf, filename, statusEl, opts) {
 
     if (isPdf) {
       try {
-        const pdfDoc = await pdfjsLib.getDocument({data: new Uint8Array(buf.slice(0))}).promise;
+        const pdfDoc = await _pdfLoad(new Uint8Array(buf.slice(0))).promise;
+        try {
           const meta = await pdfDoc.getMetadata();
           title = meta.info?.Title?.trim() || title;
           author = meta.info?.Author?.trim() || '';
+        } finally {
+          try { pdfDoc.destroy(); } catch (e) {}
+        }
       } catch (e) { /* use filename as title */ }
     } else {
       try {

static/js/app/53-library.js

@@ -933,7 +933,7 @@ function _libAddMenu() {
 
 (function () {
   const style = document.createElement('style');
-  style.textContent = [
+  const rules = [
     'html.lib-v2 #tab-books .book-drop-zone, html.lib-v2 #tab-books .book-list-filter,',
     'html.lib-v2 #book-upload-area > button.btn, html.lib-v2 #book-upload-area > .nui-add-row { display: none !important; }',
     '#book-list .lib-head { display: flex; align-items: center; gap: 6px; margin: 4px 0 8px; }',
@@ -1026,6 +1026,15 @@ function _libAddMenu() {
     '.cloud-chosen:empty { display: none; }',
     '.cloud-entry .cloud-check { width: 18px; height: 18px; margin: 0 8px 0 2px; flex: none; }',
     'body.eink #book-list .lib-cover { box-shadow: none; border: 1px solid #000; }',
-  ].join('\n');
+  ];
+  // The audiobooks' shelf (audiobooks.js) is built of the same parts: the same rules for it,
+  // its covers square like an audiobook's.
+  const shelf = rules.filter(r => r.indexOf('#book-list') >= 0).map(r => r.replace(/#book-list/g, '#audiobook-shelf'));
+  style.textContent = rules.concat(shelf, [
+    '#audiobook-shelf .lib-cover { aspect-ratio: 1; }',
+    '#audiobook-shelf .lib-hero-cover { height: auto; aspect-ratio: 1; width: 120px; align-self: flex-start; }',
+    '@media (min-width: 721px) { #audiobook-shelf .lib-hero-cover { width: 160px; height: auto; } }',
+    '#audiobook-shelf button.lib-play { border: none; font: inherit; font-size: 14px; font-weight: 600; cursor: pointer; }',
+  ]).join('\n');
   document.head.appendChild(style);
 })();

static/js/app/54-issues.js

@@ -279,7 +279,7 @@ const _issueNorm = s => String(s || '').toLowerCase().replace(/ß/g, 'ss').norma
 
 async function _issueIndexPdf(bytes) {
   if (typeof pdfjsLib === 'undefined') return null;
-  const pdf = await pdfjsLib.getDocument({data: new Uint8Array(bytes)}).promise;
+  const pdf = await _pdfLoad(new Uint8Array(bytes)).promise;
   try {
     return await _issuePdfArticles(pdf);
   } finally {

static/js/app/58-pdf.js

@@ -1,49 +1,569 @@
 'use strict';
 
+// ---------------------------------------------------------------------------
+// The PDF reader: one document, shown either scrolling (every page has its
+// place, only those near the view are drawn) or page by page, cropped to the
+// page's text (66-pdf-paginated.js). Whichever, a page is drawn only through
+// _pdfDraw: one drawing at a time per canvas, sharp at the device's pixel
+// ratio but never larger than PDF_MAX_PIXELS.
+// ---------------------------------------------------------------------------
+
+// pdf.js's own viewer caps a canvas at this many pixels: beyond it phones run out of memory.
+const PDF_MAX_PIXELS = 16777216;
+
+// One pdf.js worker for every document (the reader, covers, uploads): starting one takes a
+// moment — on an e-reader a long one.
+let _pdfWorker = null;
+
+function _pdfSharedWorker() {
+  if (!_pdfWorker && typeof pdfjsLib !== 'undefined' && pdfjsLib.PDFWorker) {
+    try { _pdfWorker = new pdfjsLib.PDFWorker(); } catch (e) { _pdfWorker = null; }
+  }
+  return _pdfWorker;
+}
+
+/** pdf.js's loading task for [bytes] (a Uint8Array), on the shared worker. */
+function _pdfLoad(bytes) {
+  const worker = _pdfSharedWorker();
+  return pdfjsLib.getDocument(worker ? {data: bytes, worker} : {data: bytes});
+}
+
+/**
+ * The page (from 1) a destination leads to — a named one or an explicit array, whose page is
+ * a reference or, in some files, a plain index. 0 when it leads nowhere.
+ */
+async function _pdfDestPage(pdf, dest) {
+  try {
+    if (typeof dest === 'string') dest = await pdf.getDestination(dest);
+    if (!dest || dest[0] == null) return 0;
+    return (typeof dest[0] === 'number' ? dest[0] : await pdf.getPageIndex(dest[0])) + 1;
+  } catch (e) {
+    return 0;
+  }
+}
+
+// The table of contents: all entries resolved at once (one after the other cost a round trip
+// to the worker each).
 async function _parsePdfOutline(pdf, items, depth) {
   depth = depth || 0;
-  const result = [];
+  const parts = await Promise.all(items.map(async item => {
-  for (const item of items) {
     let href = '';
     if (item.dest) {
-      try {
+      const n = await _pdfDestPage(pdf, item.dest);
-        const dest = typeof item.dest === 'string' ? await pdf.getDestination(item.dest) : item.dest;
+      if (n) href = `#pdf-page-${n}`;
-        if (dest) {
-          const pageIndex = await pdf.getPageIndex(dest[0]);
-          href = `#pdf-page-${pageIndex + 1}`;
     }
-      } catch (e) {}
+    const kids = item.items && item.items.length ? await _parsePdfOutline(pdf, item.items, depth + 1) : [];
-    }
+    return [{label: item.title || '(untitled)', href, depth}].concat(kids);
-    result.push({label: item.title || '(untitled)', href, depth});
+  }));
-    if (item.items && item.items.length) {
+  return [].concat.apply([], parts);
-      result.push(...await _parsePdfOutline(pdf, item.items, depth + 1));
-    }
-  }
-  return result;
 }
 
 // The document being loaded (pdf.js loading task), so opening can be cancelled.
 let _pdfLoadingTask = null;
-// The pages' drawing of the last renderPdf: an observer and its queue (stopped on the next one).
+// The open document's pages and how they are drawn (renderPdf; stopped on the next one).
+let _pdfView = null;
+// Backwards name: the reader stops "the lazy drawing" when a book closes.
 let _pdfLazy = null;
 
-function _pdfStopLazy() {
+// The open document's text, page by page (pdf.js text content): for the text layer, the
-  if (_pdfLazy) {
+// search and the crop to the text box. Kept for the document, not for one layout of it.
-    _pdfLazy.stopped = true;
+let _pdfTexts = null;
-    if (_pdfLazy.observer) _pdfLazy.observer.disconnect();
+
-    _pdfLazy = null;
+function _pdfPageText(pdf, num) {
+  if (!_pdfTexts || _pdfTexts.pdf !== pdf) _pdfTexts = {pdf, pages: new Map()};
+  let p = _pdfTexts.pages.get(num);
+  if (!p) {
+    p = pdf.getPage(num).then(page => page.getTextContent()).catch(() => ({items: [], styles: {}}));
+    _pdfTexts.pages.set(num, p);
   }
+  return p;
+}
+
+/**
+ * A page's text as one string, lower case, for searching — with where each item (each span of
+ * the text layer, in order) starts in it. Lines ending inside it get a space.
+ */
+function _pdfPageString(tc) {
+  if (tc._diora) return tc._diora;
+  let str = '';
+  const starts = [];
+  for (const item of tc.items) {
+    starts.push(str.length);
+    str += item.str || '';
+    if (item.hasEOL) str += ' ';
+  }
+  tc._diora = {str: str.toLowerCase(), starts};
+  return tc._diora;
+}
+
+// Title, table of contents and page labels: asked for once per document, alongside the
+// pages (opening doesn't wait for them; a layout anew reuses them).
+let _pdfExtras = null;
+
+function _pdfDocExtras(pdf) {
+  if (_pdfExtras && _pdfExtras.pdf === pdf) return _pdfExtras;
+  const meta = pdf.getMetadata().then(m => ({
+    title: (m && m.info && m.info.Title || '').trim(),
+    author: (m && m.info && m.info.Author || '').trim(),
+  }), () => ({title: '', author: ''}));
+  const toc = pdf.getOutline().then(o => (o && o.length ? _parsePdfOutline(pdf, o) : []), () => []);
+  const labels = pdf.getPageLabels().then(l => {
+    // Labels that only say 1, 2, 3 … are no labels.
+    if (!l || !l.length || l.every((s, i) => !s || s === String(i + 1))) return null;
+    return l.map((s, i) => s || String(i + 1));
+  }, () => null);
+  _pdfExtras = {pdf, meta, toc, labels};
+  return _pdfExtras;
+}
+
+// The page labels of the open document (front matter "iv", appendix "A-3" …), null without.
+let _pdfLabels = null;
+
+/** Page [n] (from 1) as the book numbers it. */
+function pdfPageLabel(n) {
+  return (_pdfLabels && _pdfLabels[n - 1]) || String(n);
+}
+
+/** The page meant by [value] typed in: a label of the book first, else a page number. 0: none. */
+function _pdfPageFromInput(value) {
+  const v = String(value || '').trim();
+  if (!v) return 0;
+  if (_pdfLabels) {
+    let i = _pdfLabels.indexOf(v);
+    if (i < 0) i = _pdfLabels.findIndex(l => l.toLowerCase() === v.toLowerCase());
+    if (i >= 0) return i + 1;
+  }
+  return parseInt(v, 10) || 0;
+}
+
+/**
+ * The pages in the reader's colours (Leserfarben): {background, foreground} as #rrggbb, the
+ * reader's (Dunkel, Sepia …). null: as printed (also when the reader is black on white).
+ */
+function _pdfPageColors() {
+  if (!readerSettings.pdfPageColors || readerSettings.pdfInverted) return null;
+  const contentEl = $('reader-content');
+  if (!contentEl) return null;
+  let bg = null;
+  for (let el = contentEl; el && !bg; el = el.parentElement) bg = _pdfCssHex(getComputedStyle(el).backgroundColor);
+  const fg = _pdfCssHex(getComputedStyle(contentEl).color);
+  // On white the page is as printed anyway (Hell).
+  if (!bg || !fg || bg === '#ffffff') return null;
+  return {background: bg, foreground: fg};
+}
+
+/** "rgb(…)" / "rgba(…)" as #rrggbb; null for transparent. */
+function _pdfCssHex(css) {
+  const m = /rgba?\(\s*(\d+)[,\s]+(\d+)[,\s]+(\d+)(?:[,\s/]+([\d.]+))?/.exec(css || '');
+  if (!m || (m[4] !== undefined && parseFloat(m[4]) === 0)) return null;
+  return '#' + [m[1], m[2], m[3]].map(x => (+x).toString(16).padStart(2, '0')).join('');
+}
+
+// Whether the pages are drawn for Leserfarben (with their pictures laid over them, below):
+// switching that draws them anew; another theme only changes the filter's colours.
+function _pdfColorsKey() {
+  return _pdfPageColors() ? 'reader' : '';
+}
+
+/**
+ * Leserfarben: the page is drawn as printed and recoloured by an SVG filter on its canvas —
+ * its brightness from the reader's text colour (black) to its background (white). (pdf.js's
+ * own pageColors recolours the pictures too, and lets nothing be left out.) Kept up to date
+ * with the colours; true when it applies.
+ */
+function _pdfColorFilter() {
+  const c = _pdfPageColors();
+  if (!c) return false;
+  const ns = 'http://www.w3.org/2000/svg';
+  let svg = document.getElementById('pdf-reader-colors-svg');
+  if (!svg) {
+    svg = document.createElementNS(ns, 'svg');
+    svg.id = 'pdf-reader-colors-svg';
+    svg.setAttribute('width', '0');
+    svg.setAttribute('height', '0');
+    svg.setAttribute('aria-hidden', 'true');
+    svg.style.position = 'absolute';
+    svg.innerHTML = '<filter id="pdf-reader-colors" color-interpolation-filters="sRGB">' +
+      '<feColorMatrix type="matrix" values="0.2126 0.7152 0.0722 0 0 0.2126 0.7152 0.0722 0 0 ' +
+      '0.2126 0.7152 0.0722 0 0 0 0 0 1 0"/>' +
+      '<feComponentTransfer><feFuncR type="table"/><feFuncG type="table"/><feFuncB type="table"/>' +
+      '</feComponentTransfer></filter>';
+    document.body.appendChild(svg);
+  }
+  const ch = (hex, i) => (parseInt(hex.substr(1 + i * 2, 2), 16) / 255).toFixed(3);
+  ['R', 'G', 'B'].forEach((name, i) => {
+    svg.querySelector('feFunc' + name).setAttribute('tableValues', `${ch(c.foreground, i)} ${ch(c.background, i)}`);
+  });
+  return true;
+}
+
+/**
+ * Where page [page]'s pictures are, in CSS pixels of [viewport]: from its drawing operations
+ * (the unit square of each image under the transform at that point). Image masks are left
+ * out (they're drawn in a colour, like text).
+ */
+async function _pdfImageRects(page, viewport) {
+  const OPS = pdfjsLib.OPS;
+  let list;
+  try {
+    list = await page.getOperatorList({annotationMode: pdfjsLib.AnnotationMode.ENABLE_FORMS});
+  } catch (e) {
+    return [];
+  }
+  const mul = (m, n) => [m[0] * n[0] + m[2] * n[1], m[1] * n[0] + m[3] * n[1],
+                         m[0] * n[2] + m[2] * n[3], m[1] * n[2] + m[3] * n[3],
+                         m[0] * n[4] + m[2] * n[5] + m[4], m[1] * n[4] + m[3] * n[5] + m[5]];
+  let ctm = viewport.transform.slice();
+  const stack = [];
+  const rects = [];
+  for (let i = 0; i < list.fnArray.length; i++) {
+    const fn = list.fnArray[i], args = list.argsArray[i];
+    if (fn === OPS.save) stack.push(ctm);
+    else if (fn === OPS.restore) ctm = stack.pop() || ctm;
+    else if (fn === OPS.transform) ctm = mul(ctm, args);
+    else if (fn === OPS.paintFormXObjectBegin) {
+      stack.push(ctm);
+      if (Array.isArray(args[0]) && args[0].length === 6) ctm = mul(ctm, args[0]);
+    } else if (fn === OPS.paintFormXObjectEnd) ctm = stack.pop() || ctm;
+    else if (fn === OPS.paintImageXObject || fn === OPS.paintInlineImageXObject) {
+      const xs = [ctm[4], ctm[0] + ctm[4], ctm[2] + ctm[4], ctm[0] + ctm[2] + ctm[4]];
+      const ys = [ctm[5], ctm[1] + ctm[5], ctm[3] + ctm[5], ctm[1] + ctm[3] + ctm[5]];
+      const x = Math.max(0, Math.min.apply(null, xs)), y = Math.max(0, Math.min.apply(null, ys));
+      const w = Math.min(viewport.width, Math.max.apply(null, xs)) - x;
+      const h = Math.min(viewport.height, Math.max.apply(null, ys)) - y;
+      if (w > 2 && h > 2) rects.push({x, y, w, h});
+    }
+  }
+  return rects;
+}
+
+// A picture covering this much of the page is the page itself (a scan): recoloured with it.
+const PDF_SCAN_SHARE = 0.5;
+
+/** Leserfarben: page [e]'s pictures laid over it as printed (copied from its canvas). */
+async function _pdfImageLayer(e, page, viewport, ratio, token) {
+  const rects = await _pdfImageRects(page, viewport);
+  if (e.token !== token) return;
+  const area = viewport.width * viewport.height;
+  const shown = rects.filter(r => r.w * r.h < area * PDF_SCAN_SHARE);
+  if (!shown.length) return;
+  const layer = document.createElement('div');
+  layer.className = 'pdf-image-layer';
+  for (const r of shown) {
+    const c = document.createElement('canvas');
+    const sx = Math.floor(r.x * ratio), sy = Math.floor(r.y * ratio);
+    const sw = Math.min(e.canvas.width - sx, Math.ceil(r.w * ratio)), sh = Math.min(e.canvas.height - sy, Math.ceil(r.h * ratio));
+    if (sw <= 0 || sh <= 0) continue;
+    c.width = sw;
+    c.height = sh;
+    c.getContext('2d').drawImage(e.canvas, sx, sy, sw, sh, 0, 0, sw, sh);
+    c.style.left = sx / ratio + 'px';
+    c.style.top = sy / ratio + 'px';
+    c.style.width = sw / ratio + 'px';
+    c.style.height = sh / ratio + 'px';
+    layer.appendChild(c);
+  }
+  e.inner.insertBefore(layer, e.canvas.nextSibling);
+}
+
+// Internal links followed, the pages they were followed from (the back button).
+let _pdfBackStack = [];
+
+/** The document is gone (another one, or the reader closed): what belonged to it goes too. */
+function _pdfForgetDoc() {
+  _pdfTexts = null;
+  _pdfExtras = null;
+  _pdfLabels = null;
+  _pdfBackStack = [];
+  _pdfShowBack();
+  if (typeof _pdfSearchStop === 'function') _pdfSearchStop();
+}
+
+async function _pdfFollowDest(dest) {
+  const pdf = currentPdfDoc;
+  if (!pdf) return;
+  const n = await _pdfDestPage(pdf, dest);
+  if (!n || pdf !== currentPdfDoc) return;
+  const contentEl = $('reader-content');
+  const from = readerSettings.pdfPaginated || !contentEl ? pdfCurrentPage : _pdfTopPage(contentEl);
+  if (n === from) return;
+  _pdfBackStack.push(from);
+  pdfGoToPage(n);
+  _pdfShowBack();
+}
+
+/** The button back to where the last link was followed from (gone when there's none). */
+function _pdfShowBack() {
+  let btn = document.getElementById('pdf-back-btn');
+  const from = _pdfBackStack[_pdfBackStack.length - 1];
+  if (!from) {
+    if (btn) btn.remove();
+    return;
+  }
+  if (!btn) {
+    const overlay = $('reader-overlay');
+    if (!overlay) return;
+    btn = document.createElement('button');
+    btn.id = 'pdf-back-btn';
+    btn.type = 'button';
+    btn.className = 'btn btn-sm pdf-back-btn';
+    btn.addEventListener('click', ev => {
+      ev.stopPropagation();
+      const page = _pdfBackStack.pop();
+      if (page) pdfGoToPage(page);
+      _pdfShowBack();
+    });
+    overlay.appendChild(btn);
+  }
+  btn.textContent = `← Zurück zu S. ${pdfPageLabel(from)}`;
+}
+
+/** A page's text layer and links taken off (another scale, or the page let go). */
+function _pdfDropLayers(e) {
+  if (e.textTask) { try { e.textTask.cancel(); } catch (err) {} }
+  e.textTask = null;
+  e.textDivs = null;
+  e.textInfo = null;
+  e.marked = null;
+  e.inner.querySelectorAll('.pdf-text-layer, .pdf-link-layer, .pdf-image-layer').forEach(el => el.remove());
+}
+
+/**
+ * Over a drawn page: its text (transparent, to select and copy, and where the search marks
+ * go) and its links. After the pixels, so a turn isn't held up by them.
+ */
+async function _pdfLayers(e, page, viewport, token) {
+  const pdf = currentPdfDoc;
+  if (!pdf) return;
+  let annots = [];
+  try { annots = await page.getAnnotations({intent: 'display'}); } catch (err) {}
+  if (e.token !== token) return;
+  _pdfLinks(e, viewport, annots);
+  if (typeof pdfjsLib.renderTextLayer !== 'function') return;
+  const text = await _pdfPageText(pdf, e.num);
+  if (e.token !== token || !text.items.length) return;
+  const layer = document.createElement('div');
+  layer.className = 'pdf-text-layer';
+  layer.style.setProperty('--scale-factor', viewport.scale);
+  e.inner.appendChild(layer);
+  const divs = [];
+  const task = pdfjsLib.renderTextLayer({textContentSource: text, container: layer, viewport, textDivs: divs});
+  e.textTask = task;
+  try {
+    await task.promise;
+  } catch (err) {
+    return;
+  }
+  if (e.token !== token || e.textTask !== task) return;
+  e.textTask = null;
+  e.textDivs = divs;
+  e.textInfo = {items: text.items, string: _pdfPageString(text)};
+  if (typeof _pdfSearchMarkPage === 'function') _pdfSearchMarkPage(e);
+}
+
+// Links: ones into the document go there (and back, _pdfShowBack); web addresses open outside.
+function _pdfLinks(e, viewport, annots) {
+  const links = annots.filter(a => a.subtype === 'Link' && (a.url || a.dest));
+  if (!links.length) return;
+  const layer = document.createElement('div');
+  layer.className = 'pdf-link-layer';
+  for (const a of links) {
+    const r = viewport.convertToViewportRectangle(a.rect);
+    const link = document.createElement('a');
+    link.className = 'pdf-link';
+    link.style.left = Math.min(r[0], r[2]) + 'px';
+    link.style.top = Math.min(r[1], r[3]) + 'px';
+    link.style.width = Math.abs(r[2] - r[0]) + 'px';
+    link.style.height = Math.abs(r[3] - r[1]) + 'px';
+    if (a.url) {
+      link.href = a.url;
+      link.target = '_blank';
+      link.rel = 'noopener noreferrer';
+      link.title = a.url;
+    } else {
+      const dest = a.dest;
+      link.href = '#';
+      link.addEventListener('click', ev => {
+        ev.preventDefault();
+        ev.stopPropagation();
+        _pdfFollowDest(dest);
+      });
+    }
+    layer.appendChild(link);
+  }
+  e.inner.appendChild(layer);
+}
+
+function _pdfStopLazy() {
+  const v = _pdfView;
+  if (!v) return;
+  v.stopped = true;
+  if (v.observer) v.observer.disconnect();
+  if (v.resize) v.resize.disconnect();
+  clearTimeout(v.resizeTimer);
+  v.entries.forEach(e => { e.token++; if (e.task) { try { e.task.cancel(); } catch (err) {} } });
+  _pdfView = _pdfLazy = null;
+}
+
+/**
+ * Draw page [page] of entry [e] at [cssScale] (CSS pixels per PDF point). Cancels a drawing
+ * of the same canvas still under way (pdf.js refuses two at once) and waits for it to let go.
+ * Returns false when a newer drawing (or letting go) came first.
+ */
+function _pdfDraw(e, page, cssScale) {
+  // The same drawing asked for again while it's under way (the page shown, then laid out
+  // again as the bars go): that one, not a new one cancelling it.
+  if (e.drawing && e.drawing.scale === cssScale) return e.drawing.promise;
+  const drawing = {scale: cssScale, promise: null};
+  drawing.promise = _pdfDrawNow(e, page, cssScale).then(ok => {
+    if (e.drawing === drawing) e.drawing = null;
+    return ok;
+  }, err => {
+    if (e.drawing === drawing) e.drawing = null;
+    throw err;
+  });
+  e.drawing = drawing;
+  return drawing.promise;
+}
+
+async function _pdfDrawNow(e, page, cssScale) {
+  const token = ++e.token;
+  _pdfDropLayers(e);
+  const prev = e.task;
+  if (prev) {
+    try { prev.cancel(); } catch (err) {}
+    try { await prev.promise; } catch (err) {}
+    if (e.token !== token) return false;
+  }
+  const viewport = page.getViewport({scale: cssScale});
+  let ratio = window.devicePixelRatio || 1;
+  const area = viewport.width * viewport.height;
+  if (area * ratio * ratio > PDF_MAX_PIXELS) ratio = Math.max(0.5, Math.sqrt(PDF_MAX_PIXELS / area));
+  // Drawn on a canvas of its own and put in place when it's done: the page shown so far stays
+  // until then (after a zoom larger or smaller and blurred for a moment, never blank).
+  const canvas = document.createElement('canvas');
+  canvas.className = 'pdf-page';
+  canvas.width = Math.round(viewport.width * ratio);
+  canvas.height = Math.round(viewport.height * ratio);
+  canvas.style.width = viewport.width + 'px';
+  canvas.style.height = viewport.height + 'px';
+  const task = page.render({
+    canvasContext: canvas.getContext('2d'),
+    viewport,
+    transform: ratio !== 1 ? [ratio, 0, 0, ratio, 0, 0] : null,
+  });
+  e.task = task;
+  const discard = () => { canvas.width = 0; canvas.height = 0; };
+  try {
+    await task.promise;
+  } catch (err) {
+    if (e.task === task) e.task = null;
+    discard();
+    if (err && err.name !== 'RenderingCancelledException') console.warn('PDF page', e.num, err);
+    return false;
+  }
+  if (e.task === task) e.task = null;
+  if (e.token !== token) { discard(); return false; }
+  const old = e.canvas;
+  if (old.parentNode) old.parentNode.replaceChild(canvas, old);
+  old.width = 0;
+  old.height = 0;
+  e.canvas = canvas;
+  e.drawn = cssScale;
+  e.stale = false;
+  const recolor = _pdfColorFilter();
+  e.canvas.classList.toggle('pdf-recolored', recolor);
+  if (recolor) _pdfImageLayer(e, page, viewport, ratio, token).catch(err => console.warn('PDF page pictures', e.num, err));
+  _pdfLayers(e, page, viewport, token).catch(err => console.warn('PDF page layers', e.num, err));
+  return true;
+}
+
+/** Let a page's pixels go (its place stays) and pdf.js forget what it parsed of it. */
+function _pdfLetGo(e, pdf) {
+  e.token++;
+  e.drawing = null;
+  if (e.task) { try { e.task.cancel(); } catch (err) {} }
+  _pdfDropLayers(e);
+  e.drawn = 0;
+  e.canvas.width = 0;
+  e.canvas.height = 0;
+  if (pdf) pdf.getPage(e.num).then(p => { if (!e.task) p.cleanup(); }, () => {});
+}
+
+/**
+ * What of which page is at (x, y) on the screen (client pixels): {e, x, y, rx, ry}, rx/ry the
+ * place on the page as a share of its width and height.
+ */
+function _pdfPointAt(x, y) {
+  const v = _pdfView;
+  if (!v || !v.entries.length) return null;
+  let lo = 0, hi = v.entries.length - 1;
+  while (lo < hi) {
+    const mid = (lo + hi) >> 1;
+    if (v.entries[mid].wrapper.getBoundingClientRect().bottom > y) hi = mid; else lo = mid + 1;
+  }
+  const e = v.entries[lo];
+  const r = e.canvas.getBoundingClientRect();
+  return {e, x, y, rx: r.width ? (x - r.left) / r.width : 0, ry: r.height ? (y - r.top) / r.height : 0};
+}
+
+/**
+ * Scrolling, zoomed by [f]: every page's place grows or shrinks at once (the ones drawn are
+ * shown larger or smaller until drawn anew, those near the view first), and [anchor]'s point
+ * (_pdfPointAt) stays where it is on the screen. false when there's no scrolling view.
+ */
+function _pdfRezoomScroll(f, anchor) {
+  const v = _pdfView;
+  if (!v || v.stopped || readerSettings.pdfPaginated || !v.work) return false;
+  for (const e of v.entries) {
+    e.canvas.style.width = (parseFloat(e.canvas.style.width) || 0) * f + 'px';
+    e.canvas.style.height = (parseFloat(e.canvas.style.height) || 0) * f + 'px';
+    if (e.task) {
+      e.token++;
+      e.drawing = null;
+      try { e.task.cancel(); } catch (err) {}
+    }
+    if (e.drawn || e.task) {
+      _pdfDropLayers(e);
+      e.stale = true;
+    }
+    if (e.near) v.queue.add(e);
+  }
+  if (anchor) {
+    const contentEl = v.contentEl;
+    const r = anchor.e.canvas.getBoundingClientRect();
+    if (typeof _suppressScrollJumpDetect === 'function') _suppressScrollJumpDetect();
+    contentEl.scrollLeft += r.left + anchor.rx * r.width - anchor.x;
+    contentEl.scrollTop += r.top + anchor.ry * r.height - anchor.y;
+  }
+  v.work();
+  return true;
+}
+
+/** The page shown at the top of the scrolling view (binary search, the pages are in order). */
+function _pdfTopPage(contentEl) {
+  const v = _pdfView;
+  if (!v || !v.entries.length) return pdfCurrentPage || 1;
+  const top = contentEl.getBoundingClientRect().top + 20;
+  let lo = 0, hi = v.entries.length - 1;
+  while (lo < hi) {
+    const mid = (lo + hi) >> 1;
+    if (v.entries[mid].wrapper.getBoundingClientRect().bottom > top) hi = mid; else lo = mid + 1;
+  }
+  return v.entries[lo].num;
 }
 
 // Every page gets its place at once (sized like the first page until it is drawn), but only the
 // pages near the view are drawn; ones far away are let go again. A book of 1000 scanned pages
-// opens as fast as one of ten, and doesn't hold 1000 canvases.
+// opens as fast as one of ten, and doesn't hold 1000 canvases. Page by page (pdfPaginated),
+// 66-pdf-paginated.js shows and draws the pages itself.
 async function renderPdf(arrayBuffer, contentEl, scaleOverride, pivotPage) {
   const myGen = ++_pdfRenderGen;
   _pdfStopLazy();
   let pdf = currentPdfDoc;
   if (!pdf) {
-    _pdfLoadingTask = pdfjsLib.getDocument({data: new Uint8Array(arrayBuffer.slice(0))});
+    _pdfForgetDoc();
+    _pdfLoadingTask = _pdfLoad(new Uint8Array(arrayBuffer.slice(0)));
     try {
       pdf = await _pdfLoadingTask.promise;
     } finally {
@@ -52,20 +572,15 @@ async function renderPdf(arrayBuffer, contentEl, scaleOverride, pivotPage) {
   }
   if (_pdfRenderGen !== myGen) return null;
   currentPdfDoc = pdf;
+  // [pivotPage] may be given as what of the document it is (the place saved, a fraction).
+  if (typeof pivotPage === 'function') pivotPage = pivotPage(pdf.numPages);
 
-  let pdfTitle = '', pdfAuthor = '';
+  const extras = _pdfDocExtras(pdf);
-  try {
+  extras.labels.then(labels => {
-    const meta = await pdf.getMetadata();
+    if (pdf !== currentPdfDoc) return;
-    pdfTitle = meta.info?.Title?.trim() || '';
+    _pdfLabels = labels;
-    pdfAuthor = meta.info?.Author?.trim() || '';
+    if (labels) _pdfSetCurrentPage(pdfCurrentPage);
-  } catch (e) {}
+  });
-
-  let toc = [];
-  try {
-    const outline = await pdf.getOutline();
-    if (outline && outline.length) toc = await _parsePdfOutline(pdf, outline);
-  } catch (e) {}
-  if (_pdfRenderGen !== myGen) return null;
 
   contentEl.innerHTML = '';
 
@@ -77,7 +592,6 @@ async function renderPdf(arrayBuffer, contentEl, scaleOverride, pivotPage) {
     ? contentEl.clientWidth - 32
     : Math.min(contentEl.clientWidth - 32, 900);
   const pageWidth = readerSettings.pdfSpread ? (containerWidth - 8) / 2 : containerWidth;
-  const dpr = window.devicePixelRatio || 1;
   const scaleFor = vp => (scaleOverride != null ? scaleOverride
     : Math.max(0.5, pageWidth / vp.width) * (readerSettings.pdfZoom / 100));
 
@@ -102,7 +616,7 @@ async function renderPdf(arrayBuffer, contentEl, scaleOverride, pivotPage) {
     canvas.style.height = firstVp.height + 'px';
     inner.appendChild(canvas);
     wrapper.appendChild(inner);
-    if (!readerSettings.pdfSpread) {
+    if (!readerSettings.pdfSpread || readerSettings.pdfPaginated) {
       pdfVp.appendChild(wrapper);
     } else if (pageNum === 1) {
       wrapper.classList.add('pdf-spread-cover');
@@ -117,78 +631,93 @@ async function renderPdf(arrayBuffer, contentEl, scaleOverride, pivotPage) {
       if (spreadContainer) spreadContainer.appendChild(wrapper);
       else pdfVp.appendChild(wrapper);
     }
-    entries.push({num: pageNum, wrapper, canvas, drawn: false, near: false, task: null});
+    entries.push({num: pageNum, wrapper, inner, canvas, drawn: 0, near: false, task: null, token: 0});
   }
 
   pdfTotalPages = pdf.numPages;
-
   const pivot = Math.max(1, Math.min(pdf.numPages, pivotPage || 1));
-  const pivotEl = document.getElementById(`pdf-page-${pivot}`);
-  if (pivotEl && pivot > 1) pivotEl.scrollIntoView({block: 'start'});
 
-  const lazy = {stopped: false, observer: null, queue: new Set(), working: false};
+  const view = {pdf, contentEl, entries, gen: myGen, stopped: false, observer: null, resize: null,
-  _pdfLazy = lazy;
+                resizeTimer: 0, queue: new Set(), working: false, scaleFor, showToken: 0,
-  const gone = () => lazy.stopped || _pdfRenderGen !== myGen;
+                colorsKey: _pdfColorsKey()};
+  _pdfView = _pdfLazy = view;
+  const gone = () => view.stopped || _pdfRenderGen !== myGen;
 
   async function draw(e) {
     const page = await pdf.getPage(e.num);
-    if (gone() || !e.near) return;
+    if (gone() || !e.near || readerSettings.pdfPaginated) return;
-    const viewport = page.getViewport({scale: scaleFor(page.getViewport({scale: 1}))});
+    await _pdfDraw(e, page, scaleFor(page.getViewport({scale: 1})));
-    e.canvas.style.width = viewport.width + 'px';
-    e.canvas.style.height = viewport.height + 'px';
-    e.canvas.width = Math.round(viewport.width * dpr);
-    e.canvas.height = Math.round(viewport.height * dpr);
-    const ctx = e.canvas.getContext('2d');
-    ctx.scale(dpr, dpr);
-    e.task = page.render({canvasContext: ctx, viewport});
-    try {
-      await e.task.promise;
-      e.drawn = true;
-    } catch (err) { /* cancelled: let go below */ }
-    e.task = null;
-  }
-
-  function letGo(e) {
-    if (e.task) { try { e.task.cancel(); } catch (err) {} }
-    if (!e.drawn) return;
-    e.drawn = false;
-    e.canvas.width = 0;   // frees the pixels; the CSS size keeps the page's place
-    e.canvas.height = 0;
   }
 
   // One page at a time, the nearest to the current one first.
   async function work() {
-    if (lazy.working) return;
+    if (view.working) return;
-    lazy.working = true;
+    view.working = true;
     try {
-      while (!gone() && lazy.queue.size) {
+      while (!gone() && view.queue.size && !readerSettings.pdfPaginated) {
-        const cur = typeof _currentScrollPage === 'function' && !readerSettings.pdfPaginated
+        const cur = _pdfTopPage(contentEl);
-          ? _currentScrollPage(contentEl) : (pdfCurrentPage || pivot);
+        const next = Array.from(view.queue).sort((a, b) => Math.abs(a.num - cur) - Math.abs(b.num - cur))[0];
-        const next = Array.from(lazy.queue).sort((a, b) => Math.abs(a.num - cur) - Math.abs(b.num - cur))[0];
+        view.queue.delete(next);
-        lazy.queue.delete(next);
+        if (next.near && (!next.drawn || next.stale)) await draw(next);
-        if (next.near && !next.drawn) await draw(next);
       }
     } finally {
-      lazy.working = false;
+      view.working = false;
     }
   }
 
+  view.work = work;
+
+  // Turning the device, a narrower window, the bars coming or going: laid out again at the same page.
+  let lastW = contentEl.clientWidth, lastH = contentEl.clientHeight;
+  if (typeof ResizeObserver === 'function') {
+    view.resize = new ResizeObserver(() => {
+      const w = contentEl.clientWidth, h = contentEl.clientHeight;
+      if (!w || !h || (w === lastW && h === lastH)) return;
+      const widthChanged = Math.abs(w - lastW) > 2;
+      lastW = w;
+      lastH = h;
+      clearTimeout(view.resizeTimer);
+      view.resizeTimer = setTimeout(() => {
+        if (gone()) return;
+        if (readerSettings.pdfPaginated) _pdfShowPage(pdfCurrentPage);
+        else if (widthChanged) reRenderPdf(_pdfTopPage(contentEl));
+      }, readerSettings.pdfPaginated ? 150 : 300);
+    });
+    view.resize.observe(contentEl);
+  }
+
+  if (readerSettings.pdfPaginated) {
+    // Page by page: the pivot page cropped to its text, at once (66-pdf-paginated.js).
+    pdfCurrentPage = pivot;
+    await _pdfShowPage(pivot);
+    if (gone()) return null;
+    return _pdfResult(pdf, extras);
+  }
+
+  // Opening at the saved page is no "scrolled far" (44-books.js).
+  if (pivot > 1) {
+    if (typeof _suppressScrollJumpDetect === 'function') _suppressScrollJumpDetect();
+    entries[pivot - 1].wrapper.scrollIntoView({block: 'start'});
+  }
+  pdfCurrentPage = pivot;
+
   const byEl = new Map(entries.map(e => [e.wrapper, e]));
   if (typeof IntersectionObserver === 'function') {
-    lazy.observer = new IntersectionObserver(items => {
+    view.observer = new IntersectionObserver(items => {
+      if (readerSettings.pdfPaginated) return;
       for (const it of items) {
         const e = byEl.get(it.target);
         if (!e) continue;
         e.near = it.isIntersecting;
-        if (e.near) { if (!e.drawn) lazy.queue.add(e); }
+        if (e.near) { if (!e.drawn) view.queue.add(e); }
-        else { lazy.queue.delete(e); letGo(e); }
+        else { view.queue.delete(e); if (e.drawn || e.task) _pdfLetGo(e, pdf); }
       }
       work();
     }, {root: contentEl, rootMargin: '150% 0px'});
-    entries.forEach(e => lazy.observer.observe(e.wrapper));
+    entries.forEach(e => view.observer.observe(e.wrapper));
   } else {
     // Very old WebViews: every page, from the pivot on (as before).
-    entries.forEach(e => { e.near = true; lazy.queue.add(e); });
+    entries.forEach(e => { e.near = true; view.queue.add(e); });
     work();
   }
 
@@ -197,5 +726,12 @@ async function renderPdf(arrayBuffer, contentEl, scaleOverride, pivotPage) {
   pe.near = true;
   if (!pe.drawn) await draw(pe);
   if (gone()) return null;
-  return {title: pdfTitle, author: pdfAuthor, toc, numPages: pdf.numPages};
+  return _pdfResult(pdf, extras);
+}
+
+// What openBook needs of the document: the title (asked for while the page was drawn) and
+// the table of contents to come (a long one resolves its pages after the book is shown).
+async function _pdfResult(pdf, extras) {
+  const meta = await extras.meta;
+  return {title: meta.title, author: meta.author, toc: [], tocLater: extras.toc, numPages: pdf.numPages};
 }

static/js/app/62-reader.js

@@ -29,6 +29,65 @@ function _immHandleTap(e) {
   }
 }
 
+// A double click on the text (with a mouse): full screen, the bars out of the
+// way; again (or Esc): back as it was. Fingers keep their double tap; beside
+// the text a click turns the page (63-pages.js), so only on the text itself.
+let _readerFsHidBars = false;
+
+// What openBook hangs on the content and the page number for this book, taken off again in
+// closeReader — else every book opened adds its own, and a PDF's handlers run under the next
+// EPUB. (Not addEventListener's signal: the tolino's Chrome 61 ignores it.)
+let _readerListeners = [];
+
+function _readerListen(el, type, fn, opts) {
+  if (!el) return;
+  el.addEventListener(type, fn, opts);
+  _readerListeners.push([el, type, fn, opts]);
+}
+
+function _readerUnlistenAll() {
+  for (const [el, type, fn, opts] of _readerListeners) el.removeEventListener(type, fn, opts);
+  _readerListeners = [];
+}
+
+function _readerDblClick(e) {
+  if (!matchMedia('(hover: hover) and (pointer: fine)').matches) return;
+  if (!e.target.closest('[data-epub-src], .pdf-page-wrapper')) return;
+  if (e.target.closest('button, a, input, select, textarea, label, [data-footnote-ref], mark.epub-highlight')) return;
+  // The double click also picked a word: that's not a marking.
+  const sel = window.getSelection();
+  if (sel) sel.removeAllRanges();
+  if (typeof dismissHighlightPopover === 'function') dismissHighlightPopover();
+  if (document.fullscreenElement) {
+    if (document.exitFullscreen) document.exitFullscreen().catch(() => {});
+    return;
+  }
+  const el = document.documentElement;
+  if (!el.requestFullscreen) return;
+  el.requestFullscreen().then(() => {
+    if (_immBarsVisible) {
+      _immBarsVisible = false;
+      document.body.classList.add('reader-immersive');
+      _readerFsHidBars = true;
+    }
+  }).catch(() => {});
+}
+
+document.addEventListener('fullscreenchange', () => {
+  if (document.fullscreenElement || !_readerFsHidBars) return;
+  _readerFsHidBars = false;
+  const overlay = $('reader-overlay');
+  if (overlay && overlay.style.display !== 'none' && !_immBarsVisible) {
+    _immBarsVisible = true;
+    document.body.classList.remove('reader-immersive');
+  }
+});
+
+document.addEventListener('DOMContentLoaded', () => {
+  const row = $('reader-body-row');
+  if (row) row.addEventListener('dblclick', _readerDblClick);
+});
+
 function enterReaderImmersiveMode() {
   _immBarsVisible = true;
   document.body.classList.remove('reader-immersive');
@@ -64,8 +123,10 @@ function cancelBookOpening() {
   if (overlay) overlay.style.display = 'none';
   const contentEl = $('reader-content');
   if (contentEl) contentEl.innerHTML = '';
+  if (currentPdfDoc) { try { currentPdfDoc.destroy(); } catch (e) {} }
   currentPdfDoc = null;
   currentPdfBuffer = null;
+  _readerUnlistenAll();
   _readPos = null;
   return true;
 }
@@ -86,6 +147,36 @@ async function openBook(bookId) {
   }
 }
 
+async /**
+ * Book [bookId]'s entry with its place: the server's (offline: the one kept here), or an own
+ * newer one the server doesn't have yet. null when there's none.
+ */
+async function _bookProgress(bookId) {
+  let bookData = null;
+  try {
+    const progressRes = await fetch('/books/', {cache: 'no-store'});
+    _noteServerNowHeader(progressRes);
+    const allBooks = await progressRes.json();
+    bookData = allBooks.find(b => b.id === bookId) || null;
+  } catch {
+    // offline — use cached meta
+    return _bookProgressCached(bookId);
+  }
+  // The newer of the server's place and an own one it doesn't have yet.
+  return bookData ? _withPendingProgress(bookData) : null;
+}
+
+/** The same from what's kept on this device only (no waiting for the network). */
+async function _bookProgressCached(bookId) {
+  try {
+    const cached = await _loadCachedBookMeta();
+    const bookData = cached.find(b => b.id === bookId) || null;
+    return bookData ? _withPendingProgress(bookData) : null;
+  } catch {
+    return null;
+  }
+}
+
 async function _openBookOnce(bookId) {
   // Nothing is saved for this book until its place is restored.
   _readPos = {bookId, ready: false, key: null, readAt: 0, moved: false, hold: false};
@@ -107,6 +198,9 @@ async function _openBookOnce(bookId) {
   loadingEl.querySelector('button').addEventListener('click', () => cancelBookOpening());
   const showLoading = setTimeout(() => { if (!tok.cancelled) overlay.appendChild(loadingEl); }, 400);
   const left = () => tok.cancelled;
+  // The place this book was read at: asked for at once, alongside the book's file (a PDF
+  // opens at that page instead of drawing page 1 first).
+  const progressP = _bookProgress(bookId).catch(() => null);
 
   try {
     loadReaderSettings(bookId);
@@ -151,14 +245,28 @@ async function _openBookOnce(bookId) {
     const isPdfBook = cachedMeta.type === 'pdf';
 
     if (isPdfBook) {
-      currentPdfDoc = null; // reset so renderPdf creates fresh doc
+      // A document still open (another book) goes; renderPdf loads this one.
-      const result = await renderPdf(plain, contentEl);
+      if (currentPdfDoc) { try { currentPdfDoc.destroy(); } catch (e) {} }
+      currentPdfDoc = null;
+      // The colours of the reader first: the pages may be drawn in them (Leserfarben).
+      applyReaderSettings(true);
+      // The page it opens at: the place saved, if it's known by now (else the one kept here).
+      const saved = await Promise.race([progressP, new Promise(r => setTimeout(() => r(undefined), 400))])
+        || await _bookProgressCached(bookId);
+      const startFraction = saved ? (saved.scroll_fraction || 0) : 0;
+      const result = await renderPdf(plain, contentEl, undefined,
+        n => (startFraction > 0 && n > 1 ? Math.max(1, Math.min(n, Math.round(startFraction * (n - 1)) + 1)) : 1));
       if (left() || !result) return;
       title = result.title || title;
       author = result.author || author;
       toc = result.toc;
       numPages = result.numPages;
-      currentPdfBuffer = plain;
+      // The table of contents follows when its pages are resolved.
+      result.tocLater.then(entries => {
+        if (currentBookId === bookId && currentPdfDoc && entries.length) currentBookToc = entries;
+      });
+      // pdf.js holds the document: the decrypted bytes needn't stay as well.
+      currentPdfBuffer = null;
     } else {
       currentPdfBuffer = null;
       // Prepared before (same book file, same preparation code): only the images to unpack.
@@ -219,18 +327,14 @@ async function _openBookOnce(bookId) {
     // Apply reader settings (theme, font size, etc.)
     applyReaderSettings(isPdfBook);
 
-    // If this book already has notes/highlights, surface them right away
+    // The notes margin stays closed when a book opens (also with notes in it): it's one
-    // instead of making the reader hide them behind a manual toggle.
+    // tap away (✎), and opens by itself when a note is written.
-    // Not on phones: there the margin covers the text; it stays one tap away (✎).
-    if (!isPdfBook && currentHighlights.length && !readerAnnotationsMarginOpen && !_isMobileReader()) {
-      toggleAnnotationsMargin();
-    }
 
     // Touch: swipe (paginated) + pinch-to-zoom
-    contentEl.addEventListener('touchstart',  _pdfTouchStart,  {passive: true});
+    _readerListen(contentEl, 'touchstart',  _pdfTouchStart,  {passive: true});
-    contentEl.addEventListener('touchmove',   _pdfTouchMove,   {passive: false});
+    _readerListen(contentEl, 'touchmove',   _pdfTouchMove,   {passive: false});
-    contentEl.addEventListener('touchend',    _pdfTouchEnd,    {passive: true});
+    _readerListen(contentEl, 'touchend',    _pdfTouchEnd,    {passive: true});
-    contentEl.addEventListener('touchcancel', _pdfTouchEnd,    {passive: true});
+    _readerListen(contentEl, 'touchcancel', _pdfTouchEnd,    {passive: true});
 
     // Set up progress input
     const progressInput = $('reader-progress-input');
@@ -247,16 +351,13 @@ async function _openBookOnce(bookId) {
         if (progressSuffix) progressSuffix.textContent = '%';
       }
 
-      progressInput.addEventListener('change', function () {
+      _readerListen(progressInput, 'change', function () {
         _suppressScrollJumpDetect();
         if (isPdf) {
-          const page = Math.min(numPages, Math.max(1, parseInt(this.value, 10) || 1));
+          // The book's page number ("iv", "12") or the file's.
-          this.value = page;
+          const page = Math.min(numPages, Math.max(1, _pdfPageFromInput(this.value) || pdfCurrentPage || 1));
-          const target = contentEl.querySelector(`#pdf-page-${page}`);
+          this.value = pdfPageLabel(page);
-          if (target) {
+          pdfGoToPage(page);
-            const top = target.getBoundingClientRect().top - contentEl.getBoundingClientRect().top;
-            contentEl.scrollBy({top: top - 8, behavior: 'smooth'});
-          }
         } else {
           const pct = Math.min(100, Math.max(0, _parseLocalePercent(this.value)));
           this.value = _formatBookPercent(pct / 100);
@@ -264,7 +365,7 @@ async function _openBookOnce(bookId) {
         }
       });
 
-      progressInput.addEventListener('click', function () { this.select(); });
+      _readerListen(progressInput, 'click', function () { this.select(); });
     }
 
     // Restore scroll position — must happen BEFORE auto-save timer is started
@@ -272,19 +373,7 @@ async function _openBookOnce(bookId) {
     let openedProgress = null;
     try {
       let fraction = 0, anchor = '';
-      let bookData = null;
+      const bookData = await progressP;
-      try {
-        const progressRes = await fetch('/books/', {cache: 'no-store'});
-        _noteServerNowHeader(progressRes);
-        const allBooks = await progressRes.json();
-        bookData = allBooks.find(b => b.id === bookId) || null;
-      } catch {
-        // offline — use cached meta
-        const cached = await _loadCachedBookMeta();
-        bookData = cached.find(b => b.id === bookId) || null;
-      }
-      // The newer of the server's place and an own one it doesn't have yet.
-      if (bookData) bookData = _withPendingProgress(bookData);
       fraction = bookData ? (bookData.scroll_fraction || 0) : 0;
       anchor   = bookData ? (bookData.position_anchor || '') : '';
       openedProgress = bookData;
@@ -298,6 +387,7 @@ async function _openBookOnce(bookId) {
         await new Promise(r => requestAnimationFrame(r));
         if (fraction > 0)
           contentEl.scrollTop = fraction * (contentEl.scrollHeight - contentEl.clientHeight);
+        _pdfSetCurrentPage(_pdfTopPage(contentEl));
       } else {
         // EPUB — wait for images so scrollHeight is final
         const imgs = Array.from(contentEl.querySelectorAll('img'));
@@ -330,29 +420,24 @@ async function _openBookOnce(bookId) {
     // first — that could otherwise be a later, unrelated layout correction.
     _scrollRestTop = contentEl.scrollTop;
 
-    // Update progress input on scroll
+    // PDF scrolling: the page shown at the top is the current one (header, bookmarks, the
-    contentEl.addEventListener('scroll', () => {
+    // app's footer). EPUB: shown from the position anchor, see the anchor tracking below.
-      if (!progressInput) return;
     if (isPdf) {
-        const wrappers = contentEl.querySelectorAll('.pdf-page-wrapper');
+      let pageRaf = 0;
-        const cTop = contentEl.getBoundingClientRect().top;
+      _readerListen(contentEl, 'scroll', () => {
-        let currentPage = 1;
+        if (pageRaf || readerSettings.pdfPaginated) return;
-        for (const w of wrappers) {
+        pageRaf = requestAnimationFrame(() => {
-          if (w.getBoundingClientRect().bottom > cTop + 20) {
+          pageRaf = 0;
-            currentPage = parseInt(w.id.replace('pdf-page-', ''), 10) || 1;
+          if (currentPdfDoc) _pdfSetCurrentPage(_pdfTopPage(contentEl));
-            break;
-          }
-        }
-        progressInput.value = currentPage;
-      }
-      // EPUB: shown from the position anchor, see the anchor tracking below.
         });
+      }, {passive: true});
+    }
 
     // Auto-save progress every 10s and on scroll (debounced 2s)
     // Started AFTER restore so an early visibilitychange can't overwrite with position 0
     readerScrollSaveTimer = setInterval(saveReaderProgress, 10000);
     let _scrollDebounce = null;
-    contentEl.addEventListener('scroll', () => {
+    _readerListen(contentEl, 'scroll', () => {
       clearTimeout(_scrollDebounce);
       _scrollDebounce = setTimeout(saveReaderProgress, 2000);
     }, {passive: true});
@@ -362,7 +447,7 @@ async function _openBookOnce(bookId) {
     // scrolling) would otherwise restore to a stale, pre-scroll anchor.
     if (!isPdf) {
       let _anchorTrackDebounce = null;
-      contentEl.addEventListener('scroll', () => {
+      _readerListen(contentEl, 'scroll', () => {
         clearTimeout(_anchorTrackDebounce);
         _anchorTrackDebounce = setTimeout(() => {
           _currentPositionAnchor = getPositionAnchor(contentEl);
@@ -391,7 +476,7 @@ async function _openBookOnce(bookId) {
     // fat-fingered 'g'/'G' keybind, etc.) — see _checkScrollJump above.
     _scrollRestTop = null;
     let _scrollJumpRaf = null;
-    contentEl.addEventListener('scroll', () => {
+    _readerListen(contentEl, 'scroll', () => {
       if (_scrollJumpRaf) return;
       _scrollJumpRaf = requestAnimationFrame(() => {
         _scrollJumpRaf = null;
@@ -549,6 +634,8 @@ function closeReader() {
   // Still opening: just leave it (nothing to save yet).
   if (cancelBookOpening()) return;
   if (typeof _issueReaderClosed === 'function') _issueReaderClosed();
+  // An article (65-article.js): where it was read to, before the reader empties.
+  if (typeof _articleReaderClosed === 'function') _articleReaderClosed();
   exitReaderImmersiveMode();
   stopReadAloud();
   // Save progress BEFORE hiding — scrollHeight/clientHeight return 0 once display:none
@@ -598,12 +685,12 @@ function closeReader() {
 
   // Remove touch handlers and free image blob URLs
   const contentEl = $('reader-content');
+  _readerUnlistenAll();
   if (contentEl) {
-    contentEl.removeEventListener('touchstart',  _pdfTouchStart);
+    contentEl.removeEventListener('click', _pdfPaginatedClick);
-    contentEl.removeEventListener('touchmove',   _pdfTouchMove);
+    contentEl.removeEventListener('wheel', _pdfPaginatedWheel);
-    contentEl.removeEventListener('touchend',    _pdfTouchEnd);
+    contentEl.classList.remove('pinch-active', 'pdf-paginated');
-    contentEl.removeEventListener('touchcancel', _pdfTouchEnd);
+    contentEl.style.overflow = '';
-    contentEl.classList.remove('pinch-active');
     contentEl.innerHTML = '';
   }
   for (const url of Object.values(currentImageMap)) URL.revokeObjectURL(url);
@@ -616,6 +703,7 @@ function closeReader() {
   if (typeof _pdfStopLazy === 'function') _pdfStopLazy();
   if (currentPdfDoc) { try { currentPdfDoc.destroy(); } catch (e) {} }
   currentPdfDoc = null;
+  if (typeof _pdfForgetDoc === 'function') _pdfForgetDoc();
   currentPdfBuffer = null;
   currentBookmarks = [];
   bookmarksDirty = false;
@@ -667,13 +755,9 @@ function jumpToTocEntry(href) {
     const contentEl = $('reader-content');
     if (!contentEl) return;
 
-    // PDF page jump
+    // PDF page jump (scrolling or page by page)
     if (href.startsWith('#pdf-page-')) {
-      const target = contentEl.querySelector(href);
+      pdfGoToPage(parseInt(href.slice('#pdf-page-'.length), 10) || 1);
-      if (target) {
-        const top = target.getBoundingClientRect().top - contentEl.getBoundingClientRect().top;
-        contentEl.scrollBy({top: top - 16, behavior: 'smooth'});
-      }
       return;
     }
 

static/js/app/64-reader-settings.js

@@ -15,10 +15,14 @@ function _readerTypeDefaults() {
     ? _READER_TYPE_DEFAULTS_WIDE : _READER_TYPE_DEFAULTS_NARROW;
 }
 
+// Where the settings shown are kept: the book's id, or 'article' (all articles, 65-article.js).
+let _readerSettingsKey = null;
+
 function loadReaderSettings(bookId) {
+  _readerSettingsKey = bookId;
   // Reset to defaults, then apply per-book overrides
   Object.assign(readerSettings, { theme: 'app', fontFamily: 'serif', noBold: false,
-                                   pdfZoom: 100, pdfInverted: false, pdfPaginated: false, pdfSpread: false },
+                                   pdfZoom: 100, pdfInverted: false, pdfPageColors: false, pdfPaginated: false, pdfSpread: false },
                 _readerTypeDefaults());
   try {
     const saved = JSON.parse(localStorage.getItem(`diora_reader_settings_${bookId}`) || '{}');
@@ -40,8 +44,9 @@ function loadReaderSettings(bookId) {
 }
 
 function saveReaderSettings() {
-  if (currentBookId) {
+  const key = currentBookId || _readerSettingsKey;
-    localStorage.setItem(`diora_reader_settings_${currentBookId}`, JSON.stringify({ ...readerSettings, v: 3 }));
+  if (key) {
+    localStorage.setItem(`diora_reader_settings_${key}`, JSON.stringify({ ...readerSettings, v: 3 }));
   }
 }
 
@@ -89,6 +94,12 @@ function applyReaderSettings(isPdf) {
   // PDF invert
   if (isPdf && readerSettings.pdfInverted) overlay.classList.add('pdf-inverted');
   else overlay.classList.remove('pdf-inverted');
+  // Leserfarben: another theme changes the filter's colours; switching it on or off draws
+  // the pages anew (with or without their pictures laid over them).
+  if (isPdf && currentPdfDoc) {
+    _pdfColorFilter();
+    if (_pdfView && _pdfView.colorsKey !== _pdfColorsKey()) reRenderPdf();
+  }
 
   // Annotations (marker mode + margin panel) are EPUB-only for now — PDFs have no text layer yet
   const marginBtn = $('reader-margin-btn');
@@ -113,6 +124,21 @@ function _readerWeight() {
   return Math.round(w / step) * step;
 }
 
+// "Scrollen | Seiten" and, in pages, whether the notes stand at the page's foot
+// and whether two pages lie side by side.
+// The apps' e-reader mode always reads in pages: no choice there.
+function _readerModeButtons() {
+  if (!window.DioraPages || DioraPages.isForced()) return '';
+  const pages = DioraPages.isOn();
+  return `
+      <button class="btn btn-sm ${pages ? '' : 'active'}" data-rs-mode="scroll">Scrollen</button>
+      <button class="btn btn-sm ${pages ? 'active' : ''}" data-rs-mode="pages">Seiten</button>
+      <button class="btn btn-sm ${DioraPages.pageNotes() ? 'active' : ''}" id="rs-page-notes" style="${pages ? '' : 'display:none'}"
+        title="Die Fußnoten stehen unten auf der Seite, auf der auf sie verwiesen wird">Fußnoten unten</button>
+      <button class="btn btn-sm ${DioraPages.spread() ? 'active' : ''}" id="rs-page-spread" style="${pages ? '' : 'display:none'}"
+        title="Zwei Seiten nebeneinander wie ein aufgeschlagenes Buch — ab einer Breite von 900 Pixeln">Doppelseite</button>`;
+}
+
 function toggleSettingsPanel() {
   const overlay = $('reader-overlay');
   const contentEl = $('reader-content');
@@ -148,12 +174,20 @@ function toggleSettingsPanel() {
       <button class="btn btn-sm ${readerSettings.theme === 'dark'   ? 'active' : ''}" data-rs-theme="dark">Dunkel</button>
       <button class="btn btn-sm ${readerSettings.theme === 'sepia'  ? 'active' : ''}" data-rs-theme="sepia">Sepia</button>
       <button class="btn btn-sm ${readerSettings.theme === 'bright' ? 'active' : ''}" data-rs-theme="bright">Hell</button>
+      ${_readerModeButtons()}
     `;
   } else {
     panel.innerHTML = `
       <label>Zoom <button class="btn btn-sm" id="rs-zoom-minus">−</button> <input type="range" id="rs-zoom" min="50" max="200" step="10" value="${readerSettings.pdfZoom}"> <button class="btn btn-sm" id="rs-zoom-plus">+</button> <span id="rs-zoom-val">${readerSettings.pdfZoom}%</span></label>
-      <button class="btn btn-sm ${readerSettings.pdfInverted ? 'active' : ''}" id="rs-invert">Invertieren</button>
+      <button class="btn btn-sm ${readerSettings.pdfPaginated ? '' : 'active'}" data-rs-pdfmode="scroll">Scrollen</button>
-      <button class="btn btn-sm ${readerSettings.pdfSpread ? 'active' : ''}" id="rs-spread">Doppelseite</button>
+      <button class="btn btn-sm ${readerSettings.pdfPaginated ? 'active' : ''}" data-rs-pdfmode="pages">Seiten</button>
+      <button class="btn btn-sm ${!readerSettings.pdfInverted && !readerSettings.pdfPageColors ? 'active' : ''}" data-rs-pdfcolors="original"
+        title="Die Seiten wie gedruckt">Originalfarben</button>
+      <button class="btn btn-sm ${!readerSettings.pdfInverted && readerSettings.pdfPageColors ? 'active' : ''}" data-rs-pdfcolors="reader"
+        title="Text und Hintergrund in den Farben des Lesers (Dunkel, Sepia), Bilder bleiben, wie sie sind">Leserfarben</button>
+      <button class="btn btn-sm ${readerSettings.pdfInverted ? 'active' : ''}" data-rs-pdfcolors="invert"
+        title="Die ganze Seite umgekehrt, auch Bilder (Farben bleiben ungefähr)">Invertieren</button>
+      <button class="btn btn-sm ${readerSettings.pdfSpread ? 'active' : ''}" id="rs-spread"${readerSettings.pdfPaginated ? ' style="display:none"' : ''}>Doppelseite</button>
     `;
   }
 
@@ -267,21 +301,78 @@ function toggleSettingsPanel() {
         saveReaderSettings();
       });
     });
+
+    // Scrolling or pages (63-pages.js): per device, for every book.
+    const syncMode = () => {
+      const pages = DioraPages.isOn();
+      panel.querySelectorAll('[data-rs-mode]').forEach(b => b.classList.toggle('active', (b.dataset.rsMode === 'pages') === pages));
+      const notes = panel.querySelector('#rs-page-notes');
+      if (notes) {
+        notes.style.display = pages ? '' : 'none';
+        notes.classList.toggle('active', DioraPages.pageNotes());
+      }
+      const spread = panel.querySelector('#rs-page-spread');
+      if (spread) {
+        spread.style.display = pages ? '' : 'none';
+        spread.classList.toggle('active', DioraPages.spread());
+      }
+    };
+    panel.querySelectorAll('[data-rs-mode]').forEach(btn => {
+      btn.addEventListener('click', () => { DioraPages.setOn(btn.dataset.rsMode === 'pages'); syncMode(); });
+    });
+    panel.querySelector('#rs-page-notes')?.addEventListener('click', () => { DioraPages.togglePageNotes(); syncMode(); });
+    panel.querySelector('#rs-page-spread')?.addEventListener('click', () => { DioraPages.setSpread(!DioraPages.spread()); syncMode(); });
   } else {
     const zoomRange = panel.querySelector('#rs-zoom');
 
-    zoomRange.addEventListener('input', () => applyPdfZoom(parseInt(zoomRange.value, 10)));
+    // While the slider moves only the number follows; the pages are drawn anew once it rests.
+    let zoomTimer = 0;
+    zoomRange.addEventListener('input', () => {
+      const z = parseInt(zoomRange.value, 10);
+      const val = panel.querySelector('#rs-zoom-val');
+      if (val) val.textContent = z + '%';
+      clearTimeout(zoomTimer);
+      zoomTimer = setTimeout(() => applyPdfZoom(z), 250);
+    });
     panel.querySelector('#rs-zoom-minus').addEventListener('click', () =>
       applyPdfZoom(Math.max(50,  readerSettings.pdfZoom - 10)));
     panel.querySelector('#rs-zoom-plus').addEventListener('click', () =>
       applyPdfZoom(Math.min(200, readerSettings.pdfZoom + 10)));
 
-    panel.querySelector('#rs-invert').addEventListener('click', function () {
+    // The pages' colours: as printed, in the reader's colours, or inverted.
-      readerSettings.pdfInverted = !readerSettings.pdfInverted;
+    panel.querySelectorAll('[data-rs-pdfcolors]').forEach(btn => {
-      this.classList.toggle('active', readerSettings.pdfInverted);
+      btn.addEventListener('click', () => {
+        const which = btn.dataset.rsPdfcolors;
+        readerSettings.pdfInverted = which === 'invert';
+        readerSettings.pdfPageColors = which === 'reader';
+        panel.querySelectorAll('[data-rs-pdfcolors]').forEach(b => b.classList.toggle('active', b === btn));
         applyReaderSettings(true);
         saveReaderSettings();
       });
+    });
+
+    // Scrolling or page by page (phones and e-readers start with pages).
+    panel.querySelectorAll('[data-rs-pdfmode]').forEach(btn => {
+      btn.addEventListener('click', () => {
+        const pages = btn.dataset.rsPdfmode === 'pages';
+        if (pages === !!readerSettings.pdfPaginated || !currentPdfDoc) return;
+        const contentEl = $('reader-content');
+        const page = pages ? _pdfTopPage(contentEl) : pdfCurrentPage;
+        readerSettings.pdfPaginated = pages;
+        saveReaderSettings();
+        panel.querySelectorAll('[data-rs-pdfmode]').forEach(b => b.classList.toggle('active', b === btn));
+        const spread = panel.querySelector('#rs-spread');
+        if (spread) spread.style.display = pages ? 'none' : '';
+        if (pages) {
+          pdfCurrentPage = page;
+          // Spreads are a scrolling thing: one page at a time here.
+          if (readerSettings.pdfSpread) reRenderPdf(page).then(() => enterPdfPaginatedMode());
+          else enterPdfPaginatedMode();
+        } else {
+          exitPdfPaginatedMode(page);
+        }
+      });
+    });
 
     panel.querySelector('#rs-spread').addEventListener('click', function () {
       readerSettings.pdfSpread = !readerSettings.pdfSpread;
@@ -293,37 +384,42 @@ function toggleSettingsPanel() {
   }
 }
 
-async function applyPdfZoom(newZoom) {
+/**
+ * The PDF at another zoom (%). Scrolling, [anchor] (_pdfPointAt: where the fingers were) stays
+ * in place, else the middle of the screen; page by page, the page is fitted anew (_pdfFitPage).
+ */
+async function applyPdfZoom(newZoom, anchor) {
+  const old = readerSettings.pdfZoom;
   readerSettings.pdfZoom = newZoom;
   const zoomRange = document.getElementById('rs-zoom');
   const zoomVal = document.getElementById('rs-zoom-val');
   if (zoomRange) zoomRange.value = newZoom;
   if (zoomVal) zoomVal.textContent = newZoom + '%';
   saveReaderSettings();
+  if (!currentPdfDoc) return;
   if (readerSettings.pdfPaginated) {
-    pdfSmartZoomPage(pdfCurrentPage);
+    await _pdfShowPage(pdfCurrentPage);
-  } else {
+    return;
-    await reRenderPdf();
   }
+  if (newZoom === old) return;
+  if (!anchor) {
+    const r = $('reader-content').getBoundingClientRect();
+    anchor = _pdfPointAt(r.left + r.width / 2, r.top + r.height / 2);
+  }
+  if (!_pdfRezoomScroll(newZoom / old, anchor)) await reRenderPdf();
 }
 
+// The page at the top of the view (scrolling) — 58-pdf.js keeps the pages in order.
 function _currentScrollPage(contentEl) {
-  const wrappers = contentEl.querySelectorAll('.pdf-page-wrapper');
+  return _pdfTopPage(contentEl);
-  const viewTop = contentEl.scrollTop;
-  for (const w of wrappers) {
-    if (w.offsetTop + w.offsetHeight > viewTop) {
-      const n = parseInt(w.id.replace('pdf-page-', ''), 10);
-      if (n) return n;
-    }
-  }
-  return pdfCurrentPage || 1;
 }
 
-async function reRenderPdf() {
+/** Lay the open PDF out again (zoom, spreads, another width), at [pivot] or the page shown now. */
-  if (!currentPdfBuffer) return;
+async function reRenderPdf(pivot) {
+  if (!currentPdfDoc) return;
   const contentEl = $('reader-content');
   if (!contentEl) return;
-  const pivot = readerSettings.pdfPaginated ? pdfCurrentPage : _currentScrollPage(contentEl);
+  if (!pivot) pivot = readerSettings.pdfPaginated ? pdfCurrentPage : _pdfTopPage(contentEl);
-  await renderPdf(currentPdfBuffer, contentEl, undefined, pivot);
+  await renderPdf(null, contentEl, undefined, pivot);
   if (readerSettings.pdfPaginated) enterPdfPaginatedMode();
 }

static/js/app/65-article.js

@@ -0,0 +1,174 @@
+'use strict';
+
+// ---------------------------------------------------------------------------
+// Articles in the book reader (DioraArticle) — blog posts and news (feeds.js)
+// read like books: the same overlay, typography and colours (⚙, kept for all
+// articles together), pages or scrolling (63-pages.js) with the notes at the
+// page's foot, search, read-aloud, full screen. Bookmarks, markings and the
+// notes margin stay with books (an article has nowhere to keep them).
+//
+// The article is one "chapter" in #reader-content; currentBookId stays null,
+// so nothing of a book's position or annotations is saved. Where it was read
+// to goes back to its owner (onProgress), as do clicks on its own buttons.
+// ---------------------------------------------------------------------------
+(function () {
+  let session = null;   // {title, html, startFraction, onProgress, onClick, onClose, actions}
+  let progressTimer = 0;
+
+  const overlay = () => document.getElementById('reader-overlay');
+  const content = () => document.getElementById('reader-content');
+
+  // Same-page links: the notes of a blog post ("¹" → "#fn1") become references
+  // like in books; other jumps inside the post must not change the app's
+  // address (#books, #feeds …): kept as internal links (the apps follow them).
+  function markLinks(root) {
+    root.querySelectorAll('a[href^="#"]').forEach(a => {
+      const href = a.getAttribute('href');
+      a.removeAttribute('href');
+      if (href.length < 2) return;
+      // "#t=754": a second in a podcast's article, never a note.
+      if (!/^#t=\d+$/.test(href) && !_isNoteBacklink(a) && _looksLikeFootnoteLink(a, href)) {
+        a.dataset.footnoteRef = href.slice(1);
+        a.dataset.footnoteHref = href;
+        a.style.cursor = 'pointer';
+      } else {
+        a.dataset.internalHref = href;
+      }
+    });
+    // Everything else opens outside.
+    root.querySelectorAll('a[href]').forEach(a => {
+      a.target = '_blank';
+      a.rel = 'noopener noreferrer';
+    });
+  }
+
+  // How far it is read: down to the bottom of what is on screen (0-1).
+  function fraction() {
+    const c = content();
+    const ch = c && c.querySelector(':scope > [data-epub-src]');
+    if (!ch || !ch.offsetHeight) return 0;
+    return Math.max(0, Math.min(1, (c.scrollTop + c.clientHeight - ch.offsetTop) / ch.offsetHeight));
+  }
+
+  function report() {
+    if (session && session.onProgress) session.onProgress(fraction());
+  }
+
+  function onScroll() {
+    clearTimeout(progressTimer);
+    progressTimer = setTimeout(report, 1200);
+  }
+
+  function onClick(e) {
+    if (session && session.onClick) session.onClick(e);
+  }
+
+  // The article's own buttons (star, project, original): in the browser's
+  // header; the apps put them into their reader menu (native-ui.js asks actions()).
+  function renderActions() {
+    const bar = document.querySelector('#reader-overlay .reader-header-actions');
+    if (!bar) return;
+    let box = bar.querySelector('.reader-article-actions');
+    if (!box) {
+      box = document.createElement('span');
+      box.className = 'reader-article-actions';
+      bar.insertBefore(box, bar.firstChild);
+    }
+    box.textContent = '';
+    actions().forEach(a => {
+      const b = document.createElement('button');
+      b.type = 'button';
+      b.className = 'btn-icon' + (a.active ? ' active' : '');
+      b.title = a.label;
+      b.textContent = a.icon;
+      b.addEventListener('click', e => { e.stopPropagation(); a.action(); });
+      box.appendChild(b);
+    });
+  }
+
+  function actions() {
+    return session && session.actions ? session.actions().filter(Boolean) : [];
+  }
+
+  async function open(o) {
+    const ov = overlay();
+    const c = content();
+    if (!ov || !c) return;
+    if (session) {
+      report();
+      stopReadAloud();   // the next article: the voice was reading the last one
+    }
+    else if (ov.style.display !== 'none') closeReader();   // a book was open
+    session = o;
+    loadReaderSettings('article');
+    _charIndex = null;
+    _readPos = null;
+    currentBookId = null;
+    ov.classList.add('reader-article-mode');
+    $('reader-title').textContent = o.title || '';
+    c.textContent = '';
+    const ch = document.createElement('div');
+    ch.setAttribute('data-epub-src', 'article');
+    ch.id = 'epub-chapter-0';
+    ch.className = 'fd-artview';
+    ch.innerHTML = o.html;
+    markLinks(ch);
+    c.appendChild(ch);
+    ov.style.display = '';
+    applyReaderSettings(false);
+    enterReaderImmersiveMode();
+    renderActions();
+    c.scrollTop = 0;
+    if (window.DioraPages) DioraPages.opened();
+    if (window.DioraUI && DioraUI.updateFooter) DioraUI.updateFooter();
+    const f = o.startFraction || 0;
+    if (f > 0.02 && f < 0.9) {
+      // Back where reading stopped (on any device).
+      await new Promise(r => requestAnimationFrame(r));
+      if (session !== o) return;
+      _suppressScrollJumpDetect();
+      c.scrollTop = Math.max(0, f * ch.offsetHeight + ch.offsetTop - c.clientHeight);
+    } else {
+      setTimeout(() => { if (session === o) report(); }, 1500);   // short enough to fit: read at once
+    }
+  }
+
+  // Replaces what is shown (the overview of a story arrived) without moving.
+  function update(html) {
+    const ch = content() && content().querySelector(':scope > [data-epub-src="article"]');
+    if (!ch || !session) return;
+    session.html = html;
+    ch.innerHTML = html;
+    markLinks(ch);
+  }
+
+  // closeReader (62-reader.js) calls this before it empties the reader.
+  window._articleReaderClosed = function () {
+    if (!session) return;
+    clearTimeout(progressTimer);
+    report();
+    const s = session;
+    session = null;
+    overlay().classList.remove('reader-article-mode');
+    const box = document.querySelector('#reader-overlay .reader-article-actions');
+    if (box) box.remove();
+    if (s.onClose) setTimeout(s.onClose, 0);
+  };
+
+  window.DioraArticle = {
+    open,
+    update,
+    isOpen: () => !!session,
+    close: () => { if (session) closeReader(); },
+    actions,
+    refreshActions: renderActions,
+    fraction,
+  };
+
+  document.addEventListener('DOMContentLoaded', () => {
+    const c = content();
+    if (!c) return;
+    c.addEventListener('scroll', () => { if (session) onScroll(); }, {passive: true});
+    c.addEventListener('click', e => { if (session) onClick(e); });
+  });
+})();

static/js/app/66-pdf-paginated.js

@@ -1,175 +1,386 @@
 'use strict';
 
 // ---------------------------------------------------------------------------
-// PDF Paginated Mode
+// PDF page by page (phones and e-readers by default; "Seiten" in the PDF panel):
+// one page at a time, cropped to its text so the letters are as large as the
+// screen allows, the next one drawn ahead for the turn. A tap on the left or
+// right turns (the middle shows the bars), so do a swipe, the arrow keys,
+// Space, PageUp/PageDown and the mouse wheel. Every page goes through
+// _pdfDraw (58-pdf.js).
 // ---------------------------------------------------------------------------
 
+// A page nearly empty (only its number) would be zoomed to fill the screen with one digit:
+// at most this many times the size the whole page would have.
+const PDF_CROP_MAX = 2.5;
+const PDF_CROP_PAD = 12;
+// Zoom (readerSettings.pdfZoom, %): the PDF panel's slider and pinching.
+const PDF_ZOOM_MIN = 50;
+const PDF_ZOOM_MAX = 200;
+
+let _pdfTurnSaveTimer = 0;
+let _pdfWheelSum = 0;
+let _pdfWheelAt = 0;
+let _touchStartY = 0;
+// A touch gesture under way: {kind: 'pinch' | 'pan' | 'after'} (_pdfTouchStart).
+let _pdfGesture = null;
+let _pdfPanAt = 0;   // when a page was last moved by a finger (that touch is no tap)
+
 function enterPdfPaginatedMode() {
   const contentEl = $('reader-content');
   if (!contentEl) return;
   contentEl.classList.add('pdf-paginated');
   contentEl.style.overflow = 'hidden';
   const pdfVp = document.getElementById('pdf-viewport');
-  if (pdfVp) pdfVp.style.zoom = 1;
+  if (pdfVp) { pdfVp.style.transform = ''; pdfVp.style.transformOrigin = ''; }
-
+  // From scrolling: the drawing near the view stops, the pages are this mode's now.
-  const wrappers = contentEl.querySelectorAll('.pdf-page-wrapper');
+  const v = _pdfView;
-  wrappers.forEach((w, i) => {
+  if (v && v.observer) { v.observer.disconnect(); v.observer = null; v.queue.clear(); }
-    w.style.display = (i + 1 === pdfCurrentPage) ? '' : 'none';
-  });
-
-  pdfSmartZoomPage(pdfCurrentPage);
-
-  // Tap left/right to navigate
   contentEl.addEventListener('click', _pdfPaginatedClick);
+  contentEl.addEventListener('wheel', _pdfPaginatedWheel, {passive: false});
+  _pdfShowPage(pdfCurrentPage);
 }
 
-function exitPdfPaginatedMode() {
+/** Back to scrolling, at [page] (the one shown now). */
+function exitPdfPaginatedMode(page) {
   const contentEl = $('reader-content');
   if (!contentEl) return;
   contentEl.classList.remove('pdf-paginated');
   contentEl.style.overflow = '';
   contentEl.removeEventListener('click', _pdfPaginatedClick);
-  reRenderPdf();
+  contentEl.removeEventListener('wheel', _pdfPaginatedWheel);
+  reRenderPdf(page || pdfCurrentPage);
 }
 
 function _pdfPaginatedClick(e) {
-  const w = e.currentTarget.clientWidth;
+  if (e.target.closest('a, button, input')) return;
-  if (e.clientX < w * 0.4) pdfGoToPage(pdfCurrentPage - 1);
+  // The page was just moved with a finger: no tap.
-  else if (e.clientX > w * 0.6) pdfGoToPage(pdfCurrentPage + 1);
+  if (Date.now() - _pdfPanAt < 400) return;
+  // Text just selected (the text layer): that tap wasn't a turn.
+  const sel = window.getSelection();
+  if (sel && !sel.isCollapsed) return;
+  const r = e.currentTarget.getBoundingClientRect();
+  const x = e.clientX - r.left;
+  const dir = x < r.width * 0.4 ? -1 : x > r.width * 0.6 ? 1 : 0;
+  if (!dir) return;   // the middle: the bars come or go (62-reader.js)
+  // A turn is a turn only: not also the bars.
+  e.stopPropagation();
+  pdfGoToPage(pdfCurrentPage + dir);
 }
 
+function _pdfPaginatedWheel(e) {
+  if (e.ctrlKey) return;   // the browser's zoom
+  e.preventDefault();
+  // A page taller than the screen (zoomed in) is moved first; at its end the wheel turns.
+  const pe = _pdfView && _pdfView.entries[pdfCurrentPage - 1];
+  if (pe && pe.place && pe.place.minY < pe.place.maxY) {
+    const y = Math.max(pe.place.minY, Math.min(pe.place.maxY, pe.pos.y - e.deltaY));
+    if (Math.abs(y - pe.pos.y) > 0.5) {
+      _pdfPlace(pe, pe.pos.x, y);
+      _pdfKeepFocus(pe);
+      _pdfWheelSum = 0;
+      _pdfWheelAt = Date.now();
+      return;
+    }
+  }
+  const now = Date.now();
+  if (now - _pdfWheelAt > 400) _pdfWheelSum = 0;
+  _pdfWheelAt = now;
+  _pdfWheelSum += e.deltaY;
+  if (Math.abs(_pdfWheelSum) < 60) return;
+  const dir = _pdfWheelSum > 0 ? 1 : -1;
+  _pdfWheelSum = 0;
+  _pdfWheelAt = now + 250;   // one turn per flick of the wheel
+  pdfGoToPage(pdfCurrentPage + dir);
+}
+
+// Touch (62-reader.js listens for every book; only a PDF's are handled here):
+//  - two fingers zoom around the point between them: a preview while they move (a CSS
+//    transform, the layout stays as it is), then drawn anew at the new size with that point
+//    where it was (scrolling: _pdfRezoomScroll; page by page: _pdfFitPage and v.focus);
+//  - page by page, a page larger than the screen moves with one finger, and pushed on past
+//    its edge sideways it turns; otherwise a swipe turns.
 function _pdfTouchStart(e) {
+  const v = _pdfView;
+  if (!currentPdfDoc || !v) return;
+  const contentEl = $('reader-content');
   if (e.touches.length === 2) {
-    const dx = e.touches[0].clientX - e.touches[1].clientX;
+    const a = e.touches[0], b = e.touches[1];
-    const dy = e.touches[0].clientY - e.touches[1].clientY;
+    const x = (a.clientX + b.clientX) / 2, y = (a.clientY + b.clientY) / 2;
-    _pinchStartDist = Math.hypot(dx, dy);
+    const r = contentEl.getBoundingClientRect();
-    _pinchStartZoom = readerSettings.pdfZoom;
+    const g = {kind: 'pinch', dist: Math.hypot(a.clientX - b.clientX, a.clientY - b.clientY) || 1,
-    _isPinching = true;
+               zoom: readerSettings.pdfZoom, f: 1, cx: x - r.left, cy: y - r.top};
-    const ce = $('reader-content');
+    if (readerSettings.pdfPaginated) {
-    if (ce) ce.classList.add('pinch-active');
+      g.entry = v.entries[pdfCurrentPage - 1];
+      if (!g.entry || !g.entry.pos) return;
+      g.pos = {x: g.entry.pos.x, y: g.entry.pos.y};
     } else {
-    _touchStartX = e.touches[0].clientX;
+      const vp = document.getElementById('pdf-viewport');
-    _isPinching = false;
+      if (!vp) return;
+      const vr = vp.getBoundingClientRect();
+      g.vp = vp;
+      g.ox = x - vr.left;
+      g.oy = y - vr.top;
+      g.anchor = _pdfPointAt(x, y);
+    }
+    _pdfGesture = g;
+    contentEl.classList.add('pinch-active');
+    return;
+  }
+  if (e.touches.length !== 1) return;
+  const t = e.touches[0];
+  _touchStartX = t.clientX;
+  _touchStartY = t.clientY;
+  _pdfGesture = null;
+  const pe = readerSettings.pdfPaginated && v.entries[pdfCurrentPage - 1];
+  if (pe && pe.place && (pe.place.minX < pe.place.maxX || pe.place.minY < pe.place.maxY)) {
+    _pdfGesture = {kind: 'pan', entry: pe, x: t.clientX, y: t.clientY, ox: pe.pos.x, oy: pe.pos.y,
+                   dx: 0, dy: 0, moved: false};
   }
 }
 
 function _pdfTouchMove(e) {
-  if (e.touches.length !== 2 || !_isPinching) return;
+  const g = _pdfGesture;
+  if (!g) return;
+  if (g.kind === 'pinch') {
+    if (e.touches.length !== 2) return;
     e.preventDefault();
-  const dx = e.touches[0].clientX - e.touches[1].clientX;
+    const a = e.touches[0], b = e.touches[1];
-  const dy = e.touches[0].clientY - e.touches[1].clientY;
+    const dist = Math.hypot(a.clientX - b.clientX, a.clientY - b.clientY);
-  const dist = Math.hypot(dx, dy);
+    const zoom = Math.max(PDF_ZOOM_MIN, Math.min(PDF_ZOOM_MAX, g.zoom * dist / g.dist));
-  if (_pinchStartDist === 0) return;
+    g.f = zoom / g.zoom;
-  const liveZoom = Math.max(50, Math.min(200, _pinchStartZoom * (dist / _pinchStartDist)));
+    if (g.entry) {
-  const vp = document.getElementById('pdf-viewport');
+      // The page grows around the point between the fingers (its transform's origin is 0 0).
-  if (vp) vp.style.zoom = liveZoom / 100;
+      const x = g.cx - g.f * (g.cx - g.pos.x), y = g.cy - g.f * (g.cy - g.pos.y);
+      g.entry.inner.style.transform = `translate(${x}px, ${y}px) scale(${g.f})`;
+    } else {
+      g.vp.style.transformOrigin = `${g.ox}px ${g.oy}px`;
+      g.vp.style.transform = `scale(${g.f})`;
+    }
+    return;
+  }
+  if (g.kind === 'pan' && e.touches.length === 1) {
+    const t = e.touches[0];
+    g.dx = t.clientX - g.x;
+    g.dy = t.clientY - g.y;
+    if (!g.moved && Math.hypot(g.dx, g.dy) < 8) return;
+    g.moved = true;
+    e.preventDefault();
+    _pdfPlace(g.entry, g.ox + g.dx, g.oy + g.dy);
+  }
 }
 
 function _pdfTouchEnd(e) {
-  if (_isPinching) {
+  const g = _pdfGesture;
-    _isPinching = false;
+  if (g && g.kind === 'pinch') {
-    const ce = $('reader-content');
+    // The other finger may stay down a moment: it's no swipe.
-    if (ce) ce.classList.remove('pinch-active');
+    _pdfGesture = {kind: 'after'};
-    const vp = document.getElementById('pdf-viewport');
+    const contentEl = $('reader-content');
-    const liveZoom = vp ? parseFloat(vp.style.zoom) * 100 : readerSettings.pdfZoom;
+    if (contentEl) contentEl.classList.remove('pinch-active');
-    const snapped = Math.max(50, Math.min(200, Math.round(liveZoom / 10) * 10));
+    const zoom = Math.max(PDF_ZOOM_MIN, Math.min(PDF_ZOOM_MAX, Math.round(g.zoom * g.f / 10) * 10));
-    applyPdfZoom(snapped);
+    if (g.entry) {
+      const pe = g.entry, s = pe.drawn, p = pe.place;
+      // The page's point between the fingers stays there (_pdfFitPage); the preview stays
+      // until the page is drawn at its new size.
+      if (s && p) {
+        _pdfView.focus = {x: (g.cx - g.pos.x) / s, y: (g.cy - g.pos.y) / s, sx: g.cx / p.W, sy: g.cy / p.H};
+        _pdfView.focusPage = pe.num;
+      }
+      if (zoom === readerSettings.pdfZoom) _pdfPlace(pe, g.pos.x, g.pos.y);
+    } else {
+      g.vp.style.transform = '';
+      g.vp.style.transformOrigin = '';
+    }
+    if (zoom !== readerSettings.pdfZoom) applyPdfZoom(zoom, g.anchor);
     return;
   }
-  if (!readerSettings.pdfPaginated) return;
+  if (g && g.kind === 'after') {
-  const delta = e.changedTouches[0].clientX - _touchStartX;
+    if (!e.touches || !e.touches.length) _pdfGesture = null;
-  if (delta > 50) pdfGoToPage(pdfCurrentPage - 1);
+    return;
-  else if (delta < -50) pdfGoToPage(pdfCurrentPage + 1);
+  }
+  if (g && g.kind === 'pan') {
+    _pdfGesture = null;
+    if (!g.moved) return;   // a tap: the click turns or shows the bars
+    _pdfPanAt = Date.now();
+    _pdfKeepFocus(g.entry);
+    // Already at its edge and pushed on sideways: the next or previous page (a move that only
+    // reaches the edge doesn't turn as well).
+    const p = g.entry.place;
+    const atEdge = g.dx < 0 ? g.ox <= p.minX + 1 : g.ox >= p.maxX - 1;
+    if (atEdge && Math.abs(g.dx) > 50 && Math.abs(g.dx) > Math.abs(g.dy) * 1.5) pdfGoToPage(pdfCurrentPage + (g.dx > 0 ? -1 : 1));
+    return;
+  }
+  if (!currentPdfDoc || !readerSettings.pdfPaginated || !e.changedTouches.length) return;
+  const dx = e.changedTouches[0].clientX - _touchStartX;
+  const dy = e.changedTouches[0].clientY - _touchStartY;
+  // Sideways, clearly: a slanted scroll is not a turn.
+  if (Math.abs(dx) < 50 || Math.abs(dx) < Math.abs(dy) * 1.5) return;
+  // Dragging a selection's handles (the text layer) isn't a swipe.
+  const sel = window.getSelection();
+  if (sel && !sel.isCollapsed) return;
+  pdfGoToPage(pdfCurrentPage + (dx > 0 ? -1 : 1));
 }
 
+/** The page number the reader shows (header input) follows the page. */
+function _pdfSetCurrentPage(n) {
+  pdfCurrentPage = n;
+  const input = $('reader-progress-input');
+  if (input && currentPdfDoc && document.activeElement !== input) input.value = pdfPageLabel(n);
+  // With the book's own page numbers: which page of the file it is beside them.
+  const suffix = $('reader-progress-suffix');
+  if (suffix && currentPdfDoc && pdfTotalPages) {
+    suffix.textContent = _pdfLabels ? `(${n} / ${pdfTotalPages})` : `/ ${pdfTotalPages}`;
+    if (input) input.setAttribute('inputmode', _pdfLabels ? 'text' : 'decimal');
+  }
+}
+
+/**
+ * Go to page [n], scrolling or page by page: the table of contents, the page
+ * number in the header, bookmarks, keys and the app's "Zu Seite springen".
+ */
 function pdfGoToPage(n) {
   if (!currentPdfDoc) return;
-  n = Math.max(1, Math.min(pdfTotalPages, n));
+  n = Math.max(1, Math.min(pdfTotalPages, n | 0));
-  if (n === pdfCurrentPage) return;
   const contentEl = $('reader-content');
   if (!contentEl) return;
-
+  if (!readerSettings.pdfPaginated) {
-  const oldWrapper = contentEl.querySelector(`#pdf-page-${pdfCurrentPage}`);
+    const target = contentEl.querySelector(`#pdf-page-${n}`);
-  if (oldWrapper) oldWrapper.style.display = 'none';
+    if (target) {
-
+      if (typeof _suppressScrollJumpDetect === 'function') _suppressScrollJumpDetect();
-  pdfCurrentPage = n;
+      contentEl.scrollTop += target.getBoundingClientRect().top - contentEl.getBoundingClientRect().top - 8;
-  const newWrapper = contentEl.querySelector(`#pdf-page-${pdfCurrentPage}`);
+    }
-  if (newWrapper) newWrapper.style.display = '';
+    _pdfSetCurrentPage(n);
-
+    return;
-  pdfSmartZoomPage(pdfCurrentPage);
+  }
-
+  if (n === pdfCurrentPage && _pdfView && _pdfView.shown === n) return;
-  const progressInput = $('reader-progress-input');
+  _pdfShowPage(n);
-  if (progressInput) progressInput.value = pdfCurrentPage;
+  _pdfTurned();
 }
 
-async function pdfSmartZoomPage(pageNum) {
+// A turn is reading: the place is saved soon (not only by the 10 s timer), and e-readers
-  if (!currentPdfDoc) return;
+// get their refresh (native-eink.js listens).
-  const contentEl = $('reader-content');
+function _pdfTurned() {
-  if (!contentEl) return;
+  if (typeof _noteReaderInteraction === 'function') _noteReaderInteraction();
-  const wrapper = contentEl.querySelector(`#pdf-page-${pageNum}`);
+  clearTimeout(_pdfTurnSaveTimer);
-  if (!wrapper) return;
+  _pdfTurnSaveTimer = setTimeout(() => { if (currentPdfDoc) saveReaderProgress(); }, 1500);
-  const canvas = wrapper.querySelector('canvas');
+  try { window.dispatchEvent(new CustomEvent('diora-pdf-turn', {detail: {page: pdfCurrentPage}})); } catch (e) {}
-  if (!canvas) return;
-
-  const page = await currentPdfDoc.getPage(pageNum);
-  const naturalVp = page.getViewport({scale: 1});
-  const pageW = naturalVp.width;
-  const pageH = naturalVp.height;
-
-  let bbox = _pdfPageTextBoxCache[pageNum];
-  if (!bbox) {
-    bbox = await _computePdfTextBox(page, pageW, pageH);
-    _pdfPageTextBoxCache[pageNum] = bbox;
-  }
-
-  const containerW = contentEl.clientWidth;
-  const containerH = contentEl.clientHeight;
-  const contentW = bbox.x2 - bbox.x1;
-  const contentH = bbox.y2 - bbox.y1;
-  const pad = 12;
-  const scale = Math.min(
-    (containerW - pad * 2) / contentW,
-    (containerH - pad * 2) / contentH
-  ) * (readerSettings.pdfZoom / 100);
-
-  // Re-render canvas at new scale if significantly different
-  const currentScale = canvas.width / naturalVp.width;
-  if (Math.abs(scale - currentScale) / currentScale > 0.05) {
-    const vp = page.getViewport({scale});
-    canvas.width = vp.width;
-    canvas.height = vp.height;
-    await page.render({canvasContext: canvas.getContext('2d'), viewport: vp}).promise;
-  }
-
-  // Position canvas to center the text bounding box
-  const renderedScale = canvas.width / naturalVp.width;
-  const offsetX = -renderedScale * (bbox.x1 - pad) + (containerW - renderedScale * contentW - pad * 2) / 2;
-  // PDF y-axis is bottom-up; canvas is top-down
-  const offsetY = -renderedScale * (pageH - bbox.y2 - pad) + (containerH - renderedScale * contentH - pad * 2) / 2;
-  canvas.style.transform = `translate(${offsetX}px, ${offsetY}px)`;
-  wrapper.style.overflow = 'hidden';
-  wrapper.style.width = containerW + 'px';
-  wrapper.style.height = containerH + 'px';
 }
 
-async function _computePdfTextBox(page, pageW, pageH) {
+/** Show page [n] alone, cropped; draw the next one ahead and let those far away go. */
+async function _pdfShowPage(n) {
+  const v = _pdfView;
+  if (!v || !v.entries.length) return;
+  n = Math.max(1, Math.min(v.entries.length, n));
+  _pdfSetCurrentPage(n);
+  v.shown = n;
+  const token = ++v.showToken;
+  v.entries.forEach(e => { e.wrapper.style.display = e.num === n ? '' : 'none'; });
+  try {
+    await _pdfFitPage(v, v.entries[n - 1]);
+    if (token !== v.showToken || v.stopped) return;
+    if (n < v.entries.length) await _pdfFitPage(v, v.entries[n], true);
+    if (token !== v.showToken || v.stopped) return;
+    if (n > 1 && !v.entries[n - 2].drawn) await _pdfFitPage(v, v.entries[n - 2], true);
+  } catch (err) {
+    console.warn('PDF page', n, err);
+  }
+  if (token !== v.showToken || v.stopped) return;
+  v.entries.forEach(e => { if (Math.abs(e.num - n) > 2 && (e.drawn || e.task)) _pdfLetGo(e, v.pdf); });
+}
+
+/**
+ * Page [e] sized to the screen by its text (its ink, for scans): drawn when the size differs,
+ * then placed so the text is centred. [ahead]: for the next turn, not shown yet.
+ */
+async function _pdfFitPage(v, e, ahead) {
+  const contentEl = v.contentEl;
+  const page = await v.pdf.getPage(e.num);
+  if (v.stopped) return;
+  const natural = page.getViewport({scale: 1});
+  let box = _pdfPageTextBoxCache[e.num];
+  if (!box) {
+    box = await _computePdfTextBox(page, natural);
+    _pdfPageTextBoxCache[e.num] = box;
+  }
+  if (v.stopped) return;
+  const W = contentEl.clientWidth;
+  const H = contentEl.clientHeight;
+  const bw = Math.max(1, box.x2 - box.x1);
+  const bh = Math.max(1, box.y2 - box.y1);
+  const fitBox = Math.min((W - PDF_CROP_PAD * 2) / bw, (H - PDF_CROP_PAD * 2) / bh);
+  const fitPage = Math.min((W - PDF_CROP_PAD * 2) / natural.width, (H - PDF_CROP_PAD * 2) / natural.height);
+  const scale = Math.max(0.1, Math.min(fitBox, fitPage * PDF_CROP_MAX)) * (readerSettings.pdfZoom / 100);
+  if (!e.drawn || Math.abs(scale - e.drawn) / e.drawn > 0.05) {
+    if (!await _pdfDraw(e, page, scale)) return;
+  }
+  // The text box centred where it fits (at the scale it was drawn with); larger than the
+  // screen (zoomed in), it can be moved up to its edges and starts at its top left — or
+  // where it was looked at (v.focus: a pinch, a move, the screen turned).
+  const s = e.drawn;
+  const range = (size, lo, hi) => {
+    if (hi - lo + PDF_CROP_PAD * 2 <= size + 1) {
+      const c = (size - (hi - lo)) / 2 - lo;
+      return [c, c];
+    }
+    return [size - PDF_CROP_PAD - hi, PDF_CROP_PAD - lo];
+  };
+  const [minX, maxX] = range(W, s * box.x1, s * box.x2);
+  const [minY, maxY] = range(H, s * box.y1, s * box.y2);
+  e.place = {minX, maxX, minY, maxY, W, H};
+  const f = v.focusPage === e.num ? v.focus : null;
+  e.inner.style.transformOrigin = '0 0';
+  _pdfPlace(e, f ? W * f.sx - s * f.x : maxX, f ? H * f.sy - s * f.y : maxY);
+  e.wrapper.style.overflow = 'hidden';
+  e.wrapper.style.width = W + 'px';
+  e.wrapper.style.height = H + 'px';
+  if (!ahead && v.shown === e.num) e.wrapper.style.display = '';
+}
+
+/** Page [e] moved to (x, y) (its offset in the reading area), kept within its edges (e.place). */
+function _pdfPlace(e, x, y) {
+  const p = e.place;
+  if (p) {
+    x = Math.max(p.minX, Math.min(p.maxX, x));
+    y = Math.max(p.minY, Math.min(p.maxY, y));
+  }
+  e.pos = {x, y};
+  e.inner.style.transform = `translate(${x}px, ${y}px)`;
+}
+
+/** Page [e] moved: that part of it stays in view when it's laid out again (another zoom, turned). */
+function _pdfKeepFocus(e) {
+  const v = _pdfView, p = e.place;
+  if (!v || !p || !e.drawn) return;
+  v.focus = {x: (p.W / 2 - e.pos.x) / e.drawn, y: (p.H / 2 - e.pos.y) / e.drawn, sx: 0.5, sy: 0.5};
+  v.focusPage = e.num;
+}
+
+/**
+ * Where the page's content is, in the page's own top-down coordinates at scale 1 (as
+ * getViewport gives them, so turned pages and odd media boxes come out right): its text,
+ * or — a scan without text — what isn't white on a small drawing of it.
+ */
+async function _computePdfTextBox(page, natural) {
+  const pageW = natural.width, pageH = natural.height;
   // Tier 1: text-based
   try {
-    const tc = await page.getTextContent();
+    const tc = currentPdfDoc ? await _pdfPageText(currentPdfDoc, page.pageNumber) : await page.getTextContent();
     if (tc.items && tc.items.length) {
       let x1 = Infinity, y1 = Infinity, x2 = -Infinity, y2 = -Infinity;
       for (const item of tc.items) {
-        if (!item.transform) continue;
+        if (!item.transform || !item.str || !item.str.trim()) continue;
         const tx = item.transform[4], ty = item.transform[5];
-        const iw = item.width || 0, ih = item.height || 0;
+        const iw = item.width || 0, ih = item.height || Math.abs(item.transform[3]) || 0;
-        if (tx < x1) x1 = tx;
+        for (const [px, py] of [[tx, ty], [tx + iw, ty + ih]]) {
-        if (ty < y1) y1 = ty;
+          const [vx, vy] = natural.convertToViewportPoint(px, py);
-        if (tx + iw > x2) x2 = tx + iw;
+          if (vx < x1) x1 = vx;
-        if (ty + ih > y2) y2 = ty + ih;
+          if (vx > x2) x2 = vx;
+          if (vy < y1) y1 = vy;
+          if (vy > y2) y2 = vy;
+        }
       }
       const area = (x2 - x1) * (y2 - y1);
       if (isFinite(x1) && area > pageW * pageH * 0.25) {
-        return {x1, y1, x2, y2};
+        return {x1: Math.max(0, x1), y1: Math.max(0, y1), x2: Math.min(pageW, x2), y2: Math.min(pageH, y2)};
       }
     }
   } catch (e) {}
@@ -179,17 +390,18 @@ async function _computePdfTextBox(page, pageW, pageH) {
     const lowScale = 0.3;
     const vp = page.getViewport({scale: lowScale});
     const offCanvas = document.createElement('canvas');
-    offCanvas.width = vp.width;
+    offCanvas.width = Math.ceil(vp.width);
-    offCanvas.height = vp.height;
+    offCanvas.height = Math.ceil(vp.height);
     const ctx = offCanvas.getContext('2d');
     await page.render({canvasContext: ctx, viewport: vp}).promise;
-    const {data, width, height} = ctx.getImageData(0, 0, vp.width, vp.height);
+    const {data, width, height} = ctx.getImageData(0, 0, offCanvas.width, offCanvas.height);
+    offCanvas.width = offCanvas.height = 0;
 
     let rMin = height, rMax = 0, cMin = width, cMax = 0;
     for (let r = 0; r < height; r++) {
       for (let c = 0; c < width; c++) {
         const idx = (r * width + c) * 4;
-        if (data[idx] + data[idx+1] + data[idx+2] < 720) {
+        if (data[idx] + data[idx + 1] + data[idx + 2] < 720) {
           if (r < rMin) rMin = r;
           if (r > rMax) rMax = r;
           if (c < cMin) cMin = c;
@@ -198,12 +410,7 @@ async function _computePdfTextBox(page, pageW, pageH) {
       }
     }
     if (rMin < rMax && cMin < cMax) {
-      return {
+      return {x1: cMin / lowScale, y1: rMin / lowScale, x2: cMax / lowScale, y2: rMax / lowScale};
-        x1: cMin / lowScale,
-        y1: (height - rMax) / lowScale,
-        x2: cMax / lowScale,
-        y2: (height - rMin) / lowScale,
-      };
     }
   } catch (e) {}
 

static/js/app/68-bookmarks.js

@@ -59,8 +59,8 @@ function addBookmark() {
   let label, anchor, scrollFraction;
 
   if (currentPdfDoc) {
-    const page = pdfCurrentPage || parseInt($('reader-progress-input')?.value, 10) || 1;
+    const page = pdfCurrentPage || 1;
-    label = `Seite ${page}`;
+    label = `Seite ${pdfPageLabel(page)}`;
     anchor = `pdf-page-${page}`;
     scrollFraction = (page - 1) / Math.max(1, pdfTotalPages - 1);
   } else {
@@ -143,15 +143,7 @@ function jumpToBookmark(id) {
     const contentEl = $('reader-content');
     if (!contentEl) return;
     if (bm.anchor.startsWith('pdf-page-')) {
-      if (readerSettings.pdfPaginated) {
       pdfGoToPage(parseInt(bm.anchor.replace('pdf-page-', ''), 10) || 1);
-      } else {
-        const target = contentEl.querySelector('#' + bm.anchor);
-        if (target) {
-          const top = target.getBoundingClientRect().top - contentEl.getBoundingClientRect().top;
-          contentEl.scrollBy({top: top - 16, behavior: 'smooth'});
-        }
-      }
     } else {
       const target = Array.from(contentEl.querySelectorAll('[data-epub-src]'))
         .find(el => el.getAttribute('data-epub-src') === bm.anchor);

static/js/app/70-search.js

@@ -94,15 +94,9 @@ async function doReaderSearch(query) {
       } catch (e) {}
     }
   } else {
-    // PDF: collect text layer spans
+    // PDF: the whole document, page by page (not only the pages drawn now).
-    const spans = contentEl.querySelectorAll('.pdf-text-layer > span');
+    _pdfSearchAll(query);
-    const lq = query.toLowerCase();
+    return;
-    for (const span of spans) {
-      if (span.textContent.toLowerCase().includes(lq)) {
-        span.classList.add('reader-search-match');
-        searchMatches.push(span);
-      }
-    }
   }
 
   if (countEl) countEl.textContent = searchMatches.length ? `1 / ${searchMatches.length}` : '0';
@@ -131,10 +125,7 @@ function clearReaderSearchHighlights() {
       });
     }
   } else {
-    // PDF: remove highlight class from spans
+    _pdfSearchStop();
-    document.querySelectorAll('.reader-search-match').forEach(el => {
-      el.classList.remove('reader-search-match', 'active');
-    });
   }
   searchMatches = [];
   searchMatchIndex = -1;
@@ -150,6 +141,7 @@ function clearReaderSearch() {
 
 function scrollToSearchMatch(idx) {
   if (!searchMatches.length) return;
+  if (currentPdfDoc) { _pdfShowMatch(idx); return; }
   _suppressScrollJumpDetect();
   searchMatches.forEach((m, i) => m.classList.toggle('active', i === idx));
   searchMatches[idx].scrollIntoView({behavior: 'smooth', block: 'center'});
@@ -157,6 +149,134 @@ function scrollToSearchMatch(idx) {
   if (countEl) countEl.textContent = `${idx + 1} / ${searchMatches.length}`;
 }
 
+// ---------------------------------------------------------------------------
+// PDF search: through the text of every page (58-pdf.js keeps it), in order; the matches
+// marked in the text layer of the pages drawn, and of those drawn later.
+// ---------------------------------------------------------------------------
+
+let _pdfSearchState = null;   // {query, matches: [{page, start, end}], done}
+let _pdfSearchGen = 0;
+
+function _pdfSearchCount() {
+  const countEl = document.getElementById('rs-search-count');
+  const st = _pdfSearchState;
+  if (!countEl || !st) return;
+  const more = st.done ? '' : ' …';
+  countEl.textContent = st.matches.length
+    ? `${searchMatchIndex >= 0 ? searchMatchIndex + 1 : '–'} / ${st.matches.length}${more}`
+    : (st.done ? '0' : '…');
+}
+
+async function _pdfSearchAll(query) {
+  const pdf = currentPdfDoc;
+  const gen = ++_pdfSearchGen;
+  const lq = query.toLowerCase();
+  const st = {query, matches: [], done: false};
+  _pdfSearchState = st;
+  searchMatches = st.matches;
+  searchMatchIndex = -1;
+  _pdfSearchCount();
+  // The first match from the page being read on is shown as soon as it's found.
+  const from = pdfCurrentPage || 1;
+  for (let n = 1; n <= pdf.numPages; n++) {
+    const tc = await _pdfPageText(pdf, n);
+    if (gen !== _pdfSearchGen || pdf !== currentPdfDoc) return;
+    const {str} = _pdfPageString(tc);
+    const before = st.matches.length;
+    let i = 0;
+    while (lq && (i = str.indexOf(lq, i)) !== -1) {
+      st.matches.push({page: n, start: i, end: i + lq.length});
+      i += lq.length;
+    }
+    if (st.matches.length > before) {
+      const e = _pdfView && _pdfView.entries[n - 1];
+      if (e) _pdfSearchMarkPage(e);
+      if (searchMatchIndex < 0 && n >= from) { searchMatchIndex = before; _pdfShowMatch(before); }
+    }
+    if (n % 20 === 0) _pdfSearchCount();
+  }
+  st.done = true;
+  if (searchMatchIndex < 0 && st.matches.length) { searchMatchIndex = 0; _pdfShowMatch(0); }
+  _pdfSearchCount();
+}
+
+/** Marks of search [st] in page [e]'s text layer (none when no search is on). */
+function _pdfSearchMarkPage(e) {
+  if (!e.textDivs || !e.textInfo) return;
+  if (e.marked) {
+    for (const i of e.marked) if (e.textDivs[i]) e.textDivs[i].textContent = e.textInfo.items[i].str;
+    e.marked = null;
+  }
+  const st = _pdfSearchState;
+  if (!st) return;
+  const {starts} = e.textInfo.string;
+  const items = e.textInfo.items;
+  // Per span, the pieces of it that match (one match can run over several spans).
+  const pieces = new Map();
+  st.matches.forEach((m, idx) => {
+    if (m.page !== e.num) return;
+    for (let i = 0; i < items.length; i++) {
+      const s = starts[i], len = (items[i].str || '').length;
+      if (s >= m.end) break;
+      if (s + len <= m.start || !len) continue;
+      if (!pieces.has(i)) pieces.set(i, []);
+      pieces.get(i).push([Math.max(m.start, s) - s, Math.min(m.end, s + len) - s, idx]);
+    }
+  });
+  e.marked = [];
+  pieces.forEach((list, i) => {
+    const div = e.textDivs[i];
+    if (!div) return;
+    const str = items[i].str;
+    div.textContent = '';
+    let at = 0;
+    for (const [a, b, idx] of list) {
+      if (a > at) div.appendChild(document.createTextNode(str.slice(at, a)));
+      const mark = document.createElement('mark');
+      mark.className = 'reader-search-match' + (idx === searchMatchIndex ? ' active' : '');
+      mark.dataset.m = idx;
+      mark.textContent = str.slice(a, b);
+      div.appendChild(mark);
+      at = b;
+    }
+    if (at < str.length) div.appendChild(document.createTextNode(str.slice(at)));
+    e.marked.push(i);
+  });
+}
+
+/** Go to match [idx]: its page, and (scrolling) the match itself into view. */
+async function _pdfShowMatch(idx) {
+  const st = _pdfSearchState;
+  const m = st && st.matches[idx];
+  if (!m) return;
+  const gen = _pdfSearchGen;
+  _pdfSearchCount();
+  document.querySelectorAll('.pdf-text-layer mark.reader-search-match').forEach(mk => {
+    mk.classList.toggle('active', +mk.dataset.m === idx);
+  });
+  const contentEl = $('reader-content');
+  const shown = readerSettings.pdfPaginated ? pdfCurrentPage : (contentEl ? _pdfTopPage(contentEl) : 0);
+  if (m.page !== shown || readerSettings.pdfPaginated) pdfGoToPage(m.page);
+  if (readerSettings.pdfPaginated || !contentEl) return;
+  // Scrolling: the page is drawn (and its text laid over it) once it's near the view.
+  for (let t = 0; t < 50; t++) {
+    if (gen !== _pdfSearchGen || searchMatchIndex !== idx) return;
+    const mk = contentEl.querySelector(`#pdf-page-${m.page} mark[data-m="${idx}"]`);
+    if (mk) {
+      _suppressScrollJumpDetect();
+      mk.scrollIntoView({block: 'center'});
+      return;
+    }
+    await new Promise(r => setTimeout(r, 100));
+  }
+}
+
+function _pdfSearchStop() {
+  _pdfSearchGen++;
+  _pdfSearchState = null;
+  if (_pdfView) _pdfView.entries.forEach(e => { if (e.marked) _pdfSearchMarkPage(e); });
+}
+
 function readerSearchNext() {
   if (!searchMatches.length) return;
   searchMatchIndex = (searchMatchIndex + 1) % searchMatches.length;

static/js/app/72-highlights.js

@@ -219,7 +219,9 @@ function handleReaderSelection(e) {
     return;
   }
 
-  if (!markerModeActive) return;
+  // Markings belong to books (an article has nowhere to keep them, 65-article.js), and to
+  // EPUBs only: a PDF's text layer is there to select and copy.
+  if (!markerModeActive || !currentBookId || currentPdfDoc) return;
   // Don't hijack taps on real interactive content (footnote refs, external
   // links) — but plenty of EPUBs wrap plain reading text in a bare, hrefless
   // <a id="..."> used only as an internal jump target (sanitizeEpubHtml

static/js/app/78-footnotes.js

@@ -1,41 +1,131 @@
 'use strict';
 
 // ---------------------------------------------------------------------------
-// Footnote popover
+// Footnotes: finding a reference's note, and the popover that shows it
 // ---------------------------------------------------------------------------
+// The whole book is one concatenated DOM, and ids like "fn1" repeat per
+// chapter file: a note is looked up in the chapter its link points into (the
+// link's original href, kept in data-footnote-href by sanitizeEpubHtml), and
+// only exactly the note is shown — not the parent around an id that sits on
+// the note's own <p>/<aside>/<li>, which can be a whole chapter. The same
+// lookup serves the popover here, the notes at the foot of a page
+// (63-pages.js) and the apps' note sheet (diora-android native-ui.js).
+
+const _NOTE_INLINE = /^(A|SPAN|SUP|SUB|B|I|EM|STRONG|SMALL|CODE)$/;
+
+function noteHrefOf(link) {
+  return link.dataset.footnoteHref || ('#' + (link.dataset.footnoteRef || ''));
+}
+
+function _noteChapterFor(fromEl, href) {
+  const chapter = fromEl.closest('[data-epub-src]');
+  const file = href.split('#')[0];
+  if (!file || !chapter) return chapter;
+  const src = chapter.dataset.epubSrc || '';
+  const dir = src.includes('/') ? src.slice(0, src.lastIndexOf('/') + 1) : '';
+  const target = resolveEpubPath(dir, file);
+  return document.querySelector(`#reader-content > [data-epub-src="${CSS.escape(target)}"]`) || chapter;
+}
+
+// The element `href` (as written in the book, relative to fromEl's chapter) points at.
+function resolveNoteTarget(fromEl, href) {
+  const id = decodeURIComponent(href.slice(href.indexOf('#') + 1));
+  const scope = _noteChapterFor(fromEl, href);
+  const content = document.getElementById('reader-content');
+  let el = id && scope ? scope.querySelector(`[id="${CSS.escape(id)}"]`) : null;
+  if (!el && id && content) el = content.querySelector(`[id="${CSS.escape(id)}"]`);
+  if (!el && !id) el = scope; // link to a whole chapter file
+  return el;
+}
+
+// Converters (zeno.org / e-artnow, some Calibre output) park an empty
+// <a id="F537"></a> right *before* the note's paragraph instead of inside it —
+// or at the start of it (then the paragraph is the note, not what follows).
+function _noteAfterEmptyAnchor(target) {
+  if ((target.textContent || '').trim() || target.querySelector('img')) return null;
+  const own = target.closest('p, li, dd, dt, td, blockquote');
+  const ownText = own && !own.hasAttribute('data-epub-src') ? (own.textContent || '').trim() : '';
+  if (ownText && !/^[\[(]?\d{1,4}[\])]?\.?$/.test(ownText)) return own;
+  if (ownText) {
+    // Only the number: the note (a poem, a long quote) runs on in the next
+    // paragraphs, up to the next note's anchor.
+    const box = document.createElement('div');
+    box.appendChild(own.cloneNode(true));
+    let len = 0;
+    for (let next = own.nextElementSibling; next && len < 6000; next = next.nextElementSibling) {
+      if (next.matches('h1, h2, h3, h4, h5, h6') || next.querySelector('[id]')) break;
+      if (!(next.textContent || '').trim()) continue;
+      box.appendChild(next.cloneNode(true));
+      len += next.textContent.length;
+    }
+    return box;
+  }
+  for (let el = target; el && !el.hasAttribute('data-epub-src'); el = el.parentElement) {
+    let next = el.nextElementSibling;
+    while (next && !(next.textContent || '').trim()) next = next.nextElementSibling;
+    if (next) return next.matches('h1, h2, h3, h4, h5, h6') ? null : next;
+  }
+  return null;
+}
+
+function _noteBox(target) {
+  const following = _noteAfterEmptyAnchor(target);
+  if (following) return following;
+  let box = target.closest(_NOTE_BOX_SELECTOR);   // 40-epub.js
+  if (box && (box.hasAttribute('data-epub-src') || box.id === 'reader-content')) box = null;
+  if (!box) box = _NOTE_INLINE.test(target.tagName) ? (target.closest('p, li, dd, div, td, blockquote') || target) : target;
+  // Never a whole chapter or section: fall back to the paragraph around the target.
+  if (box.hasAttribute('data-epub-src') || (box.textContent || '').length > 6000) {
+    box = target.closest('p, li, dd') || target;
+  }
+  return box;
+}
+
+// A copy of the note at `target`, ready to show outside the book.
+function noteContentFor(target, label) {
+  const clone = _noteBox(target).cloneNode(true);
+  clone.querySelectorAll('[role="doc-backlink"], [epub\\:type~="backlink"]').forEach(e => e.remove());
+  // A page of the printed edition that starts inside the note.
+  clone.querySelectorAll('.epub-pagenum, [role="doc-pagebreak"], [epub\\:type~="pagebreak"]').forEach(e => e.remove());
+  // Links back to the text ("↩", or the note's own number at its start).
+  clone.querySelectorAll('a').forEach(a => {
+    const t = (a.textContent || '').trim();
+    if (/^[↩↑⤴^]+$/.test(t)) a.remove();
+  });
+  // The note's own number at its very start ("3 Endnote…", "[4] …"): whatever
+  // shows the note labels it already.
+  const first = Array.prototype.find.call(clone.querySelectorAll('a, sup, span, b, strong'),
+    e => (e.textContent || '').trim());
+  const firstText = first ? (first.textContent || '').trim() : '';
+  const bare = t => t.replace(/[\[\]().\s]/g, '');
+  if (first && firstText && (clone.textContent || '').trim().startsWith(firstText) &&
+      (bare(firstText) === bare(label || '') || /^[\[(]?(\d{1,4}|[*†‡§]+)[\])]?\.?$/.test(firstText))) {
+    first.remove();
+    // …and the space that followed it.
+    const walker = document.createTreeWalker(clone, NodeFilter.SHOW_TEXT, null, false);
+    const t = walker.nextNode();
+    if (t) t.nodeValue = t.nodeValue.replace(/^\s+/, '');
+  }
+  // No duplicate ids in the page (getElementById must keep finding the book's).
+  clone.removeAttribute('id');
+  clone.querySelectorAll('[id]').forEach(e => e.removeAttribute('id'));
+  clone.querySelectorAll('mark.tts-current, mark.reader-search-match').forEach(m => m.replaceWith(...m.childNodes));
+  return clone;
+}
+
+// The note of a reference link, or null.
+function noteForLink(link) {
+  const target = resolveNoteTarget(link, noteHrefOf(link));
+  return target ? noteContentFor(target, (link.textContent || '').trim()) : null;
+}
 
 let currentFootnotePopover = null;
 let _footnotePopoverTimer = null;
 
 function showFootnotePopover(link) {
   dismissFootnotePopover();
-  const targetEl = document.getElementById(link.dataset.footnoteRef);
+  const note = noteForLink(link);
-  if (!targetEl) return;
+  if (!note) return;
-  // The note's paragraph, also when the id sits on an inline element in it
-  // (<p><span><a id="F400" …>143</a></span> Note…</p>).
-  const inline = /^(A|SPAN|SUP|SUB|B|I|EM|STRONG|SMALL)$/.test(targetEl.tagName);
-  const container = targetEl.closest('.footnote') ||
-    (inline && targetEl.closest('p, li, dd, blockquote')) || targetEl.parentElement;
-  if (!container) return;
-
-  const clone = container.cloneNode(true);
-  clone.querySelector('sup')?.remove();
-  clone.querySelectorAll('.epub-pagenum').forEach(e => e.remove());
-  // The note's own number at its start, a link back to the text ("1 Note…").
-  const first = Array.from(clone.querySelectorAll('a')).find(a => a.textContent.trim());
-  const num = first ? first.textContent.trim() : '';
-  if (/^[\[(]?\d{1,4}[\])]?\.?$/.test(num) && clone.textContent.trim().startsWith(num)) first.remove();
-  // Only the number: the note (a poem, a long quote) runs on in the next
-  // paragraphs, up to the next note's anchor.
-  if (!clone.textContent.trim()) {
-    let len = 0;
-    for (let next = container.nextElementSibling; next && len < 6000; next = next.nextElementSibling) {
-      if (/^H[1-6]$/.test(next.tagName) || next.querySelector('[id]')) break;
-      if (!next.textContent.trim()) continue;
-      clone.appendChild(next.cloneNode(true));
-      len += next.textContent.length;
-    }
-  }
 
   const popover = document.createElement('div');
   popover.className = 'footnote-popover';
@@ -43,7 +133,7 @@ function showFootnotePopover(link) {
   ['reader-theme-sepia', 'reader-theme-bright'].forEach(cls => {
     if (_overlay?.classList.contains(cls)) popover.classList.add(cls);
   });
-  popover.innerHTML = clone.innerHTML;
+  popover.appendChild(note);
 
   const closeBtn = document.createElement('button');
   closeBtn.className = 'footnote-popover-close';
@@ -66,7 +156,7 @@ function showFootnotePopover(link) {
 
   const pr = popover.getBoundingClientRect();
   if (pr.right  > window.innerWidth  - 8) popover.style.left = (window.innerWidth  - pr.width  - 8) + 'px';
-  if (pr.bottom > window.innerHeight - 8) popover.style.top  = (rect.top - pr.height - 8) + 'px';
+  if (pr.bottom > window.innerHeight - 8) popover.style.top  = Math.max(8, rect.top - pr.height - 8) + 'px';
 }
 
 function dismissFootnotePopover() {

static/js/audiobooks.js

@@ -31,6 +31,10 @@
     folder: null,      // the folder chip chosen (null: all, '': without a folder)
     folderOrder: [],   // the podcasts' folder order (/podcasts/folders/)
     series: null,      // the series shown (its key), inside the folder
+    q: '',             // the search in the shelf's head
+    searching: false,  // the search field open (narrow screens show a ⌕ first)
+    sort: cached('sort') || 'recent',
+    status: 'all',     // the state chip: all, started, unheard, heard
   };
 
   const shelf = document.createElement('div');
@@ -175,45 +179,6 @@
     return st.detail;
   }
 
-  // ---- the shelf in "Bücher" ------------------------------------------------------------
-
-  function renderShelf() {
-    const h = host();
-    const books = st.books || [];
-    const state = h ? podcastState() : null;
-    const pods = state && state.feeds ? state.feeds.filter(f => f.kind !== 'audiobook') : [];
-    if (!h) { shelf.innerHTML = ''; return; }
-    if (!books.length && !pods.length) {
-      shelf.innerHTML = '<p class="ab-empty">Hörbücher und Hörspiele sind hier Podcasts, die wie ein Buch gehört ' +
-        'werden: Kapitel in Reihenfolge, Lesezeichen. Erst einen Podcast abonnieren (Podcasts → Suche), dann hier markieren.</p>';
-      return;
-    }
-    let html = '';
-    const go = books.filter(b => b.next && b.last_played_at)[0];
-    if (go) {
-      html += '<div class="ab-continue" data-ab-open="' + go.feed_id + '">' +
-        cover(go.artwork_url, go.title, 'ab-cover ab-cover-s') +
-        '<div class="ab-card-text"><div class="ab-label">Weiterhören</div>' +
-        '<div class="ab-title">' + esc(go.title) + '</div>' +
-        '<div class="ab-meta">' + esc(progressText(go)) + '</div>' +
-        '<div class="ab-bar"><span style="width:' + pct(go) + '%"></span></div></div>' +
-        '<button type="button" class="ab-play" data-ab-go="' + go.feed_id + '" aria-label="Weiterhören">▶</button></div>';
-    }
-    html += '<div class="ab-head"><h3 class="ab-h">Hörbücher</h3>' +
-      (pods.length ? '<button type="button" class="ab-link" data-ab="pick">+ Hörbuch</button>' : '') + '</div>';
-    if (books.length) {
-      html += gridHtml(books);
-    } else {
-      html += '<p class="ab-empty">Podcasts, die eigentlich Hörbücher oder Hörspiele sind, lassen sich hier ' +
-        'wie ein Buch hören: in Kapitelreihenfolge, mit Lesezeichen. „+ Hörbuch“ wählt einen aus.</p>';
-    }
-    if (st.picking) html += pickHtml(pods);
-    shelf.innerHTML = html;
-    drawChips(books);
-    const input = shelf.querySelector('.ab-pick-q');
-    if (input && document.activeElement !== input && st.pickFocus) { input.focus(); st.pickFocus = false; }
-  }
-
   // ---- organising: folders (set per book) and series (found in the titles) ----------------
 
   // "Das Kapital – Erster Band", "Harry Potter: Teil 3", "Die drei ??? – Folge 12":
@@ -270,28 +235,131 @@
       x.sort.localeCompare(y.sort, 'de'));
   }
 
+  // ---- the shelf, built like the books' library (53-library.js: its lib-* parts, and its
+  // styles, which hold for #audiobook-shelf too): a head with search, sort and +, chips for
+  // the state and the folders, "Weiterhören" on top, the books as cards. ---------------------
+
+  const SORTS = [['recent', 'Zuletzt'], ['title', 'Titel'], ['author', 'Autor'],
+    ['long', 'Längste zuerst'], ['short', 'Kürzeste zuerst']];
+  const heard = b => !b.next;
+  const started = b => !!b.next && (b.heard_seconds > 0 || b.done > 0 || !!b.last_played_at);
+  const unheard = b => !!b.next && !started(b);
+  const STATES = {started, unheard, heard};
+
+  function renderShelf() {
+    const h = host();
+    const books = st.books || [];
+    const state = h ? podcastState() : null;
+    const pods = state && state.feeds ? state.feeds.filter(f => f.kind !== 'audiobook') : [];
+    if (!h) { shelf.innerHTML = ''; return; }
+    const narrow = window.matchMedia && matchMedia('(max-width: 720px)').matches;
+    let html = '<div class="lib-head"><h2 class="lib-h">Hörbücher</h2>' +
+      (!narrow || st.searching || st.q
+        ? '<input class="lib-q" type="search" placeholder="Titel, Autor …" value="' + esc(st.q) + '">'
+        : '<button type="button" class="lib-icon" data-ab="search" aria-label="Suchen">⌕</button>') +
+      '<select class="lib-sort" aria-label="Sortieren">' + SORTS.map(s =>
+        '<option value="' + s[0] + '"' + (st.sort === s[0] ? ' selected' : '') + '>' + s[1] + '</option>').join('') + '</select>' +
+      (pods.length ? '<button type="button" class="lib-icon lib-add" data-ab="pick" aria-label="Hörbuch hinzufügen" ' +
+        'title="Einen abonnierten Podcast als Hörbuch führen">+</button>' : '') +
+      '</div><div class="lib-chips"></div>';
+    if (st.picking) html += pickHtml(pods);
+    if (!books.length) {
+      html += '<p class="lib-empty">' + (pods.length
+        ? 'Podcasts, die eigentlich Hörbücher oder Hörspiele sind, lassen sich hier wie ein Buch hören: ' +
+          'in Kapitelreihenfolge, mit Lesezeichen. Mit + einen auswählen.'
+        : 'Hörbücher und Hörspiele sind hier Podcasts, die wie ein Buch gehört werden: Kapitel in Reihenfolge, ' +
+          'Lesezeichen. Erst einen Podcast abonnieren (Podcasts → Suche), dann hier mit + markieren.') + '</p>';
+    } else {
+      html += listHtml(books);
+    }
+    const q = shelf.querySelector('.lib-q');
+    const focused = q && document.activeElement === q;
+    shelf.innerHTML = html;
+    drawChips(books);
+    if (focused || (st.searching && !st.q)) {
+      const n = shelf.querySelector('.lib-q');
+      if (n) { n.focus(); n.setSelectionRange(n.value.length, n.value.length); }
+    }
+    const input = shelf.querySelector('.ab-pick-q');
+    if (input && document.activeElement !== input && st.pickFocus) { input.focus(); st.pickFocus = false; }
+  }
+
+  function shownBooks(books) {
+    const q = (st.q || '').trim().toLowerCase();
+    return books.filter(b =>
+      (st.folder == null ? true : st.folder === '' ? !foldersOf(b).length : foldersOf(b).indexOf(st.folder) >= 0) &&
+      (!STATES[st.status] || STATES[st.status](b)) &&
+      (!q || (b.title + ' ' + (b.author || '')).toLowerCase().indexOf(q) >= 0));
+  }
+
+  // The list's order (entries() comes "Zuletzt": the last heard first).
+  function sorted(list) {
+    const books = e => (e.kind === 'series' ? e.books : [e.b]);
+    const length = e => books(e).reduce((n, b) => n + (b.total_seconds || 0), 0);
+    const author = e => books(e)[0].author || '';
+    const by = {
+      title: (x, y) => x.sort.localeCompare(y.sort, 'de'),
+      author: (x, y) => author(x).localeCompare(author(y), 'de') || x.sort.localeCompare(y.sort, 'de'),
+      long: (x, y) => length(y) - length(x),
+      short: (x, y) => length(x) - length(y),
+    }[st.sort];
+    return by ? list.slice().sort(by) : list;
+  }
+
+  function listHtml(books) {
+    const here = shownBooks(books);
+    // Series as stacks, unless searching or filtering by state (as with the books).
+    const stack = !st.q && !STATES[st.status];
+    const list = sorted(stack ? entries(here) : here.map(b => ({kind: 'book', b, sort: b.title})));
+    if (st.series != null) {
+      const g = list.filter(e => e.kind === 'series' && e.key === st.series)[0];
+      if (g) {
+        return '<div class="lib-place"><button type="button" class="lib-btn" data-ab="up">‹ Hörbücher</button>' +
+          '<strong class="lib-place-name">' + esc(g.name) + '</strong></div>' +
+          '<div class="lib-grid">' + g.books.map(b => bookCard(b, b._volume)).join('') + '</div>';
+      }
+      st.series = null;
+    }
+    let html = '';
+    const go = !st.q && st.folder == null && !STATES[st.status] && books.filter(b => b.next && b.last_played_at)[0];
+    if (go) html += continueHtml(go);
+    if (!list.length) return html + '<p class="lib-empty">Nichts gefunden.</p>';
+    return html + '<div class="lib-grid">' + list.map(e => e.kind === 'series' ? seriesCard(e) : bookCard(e.b)).join('') + '</div>';
+  }
+
+  // "Weiterhören": the book last listened to, large — like "Weiterlesen".
+  function continueHtml(b) {
+    return '<section class="lib-recent" aria-label="Weiterhören"><div class="lib-continue" data-ab-open="' + b.feed_id + '">' +
+      cover(b.artwork_url, b.title, 'lib-hero-cover', b.author) +
+      '<span class="lib-text"><span class="lib-kicker">Weiterhören</span>' +
+      '<span class="lib-title lib-hero-title">' + esc(b.title) + '</span>' +
+      (b.author ? '<span class="lib-meta">' + esc(b.author) + '</span>' : '') +
+      '<span class="lib-hero-progress"><span class="lib-bar"><span style="width:' + pct(b) + '%"></span></span>' +
+      '<span class="lib-meta">' + esc(progressText(b)) + '</span></span>' +
+      '<button type="button" class="lib-play" data-ab-go="' + b.feed_id + '">▶ Weiterhören</button></span></div></section>';
+  }
+
   function bookCard(b, label) {
-    return '<button type="button" class="ab-card" data-ab-open="' + b.feed_id + '">' +
+    return '<div class="lib-card" data-ab-open="' + b.feed_id + '">' +
-      cover(b.artwork_url, b.title, 'ab-cover') +
+      '<span class="lib-coverbox">' + cover(b.artwork_url, b.title, 'lib-cover', b.author) +
-      '<span class="ab-card-text"><span class="ab-title">' + esc(label || b.title) + '</span>' +
+      (heard(b) ? '<span class="lib-badge">✓</span>' : '') + '</span>' +
-      (b.author && !label ? '<span class="ab-meta">' + esc(b.author) + '</span>' : '') +
+      '<span class="lib-bar"><span style="width:' + pct(b) + '%"></span></span>' +
-      '<span class="ab-meta">' + esc(progressText(b)) + '</span>' +
+      '<span class="lib-title">' + esc(label || b.title) + '</span>' +
-      '<span class="ab-bar"><span style="width:' + pct(b) + '%"></span></span></span></button>';
+      '<span class="lib-meta">' + esc(label ? progressText(b) : (b.author || progressText(b))) + '</span></div>';
   }
 
   function seriesCard(g) {
     const total = g.books.reduce((n, b) => n + (b.total_seconds || 0), 0);
-    const heard = g.books.reduce((n, b) => n + (b.heard_seconds || 0), 0);
+    const heardSecs = g.books.reduce((n, b) => n + (b.heard_seconds || 0), 0);
     const going = g.books.filter(b => b.next && b.last_played_at)[0];
-    const done = g.books.filter(b => !b.next).length;
+    const done = g.books.filter(heard).length;
     const meta = g.books.length + ' Bände' + (going ? ' · bei ' + going._volume
       : done === g.books.length ? ' · fertig gehört' : done ? ' · ' + done + ' gehört' : '');
-    return '<button type="button" class="ab-card ab-stack" data-ab-series="' + esc(g.key) + '">' +
+    return '<div class="lib-card lib-stack" data-ab-series="' + esc(g.key) + '">' +
-      '<span class="ab-stack-covers">' + cover(g.books[0].artwork_url, g.name, 'ab-cover') + '</span>' +
+      '<span class="lib-coverbox lib-stackbox">' + cover(g.books[0].artwork_url, g.name, 'lib-cover', g.books[0].author) + '</span>' +
-      '<span class="ab-card-text"><span class="ab-title">' + esc(g.name) + '</span>' +
+      '<span class="lib-bar"><span style="width:' + (total ? Math.round(100 * heardSecs / total) : 0) + '%"></span></span>' +
-      (g.books[0].author ? '<span class="ab-meta">' + esc(g.books[0].author) + '</span>' : '') +
+      '<span class="lib-title">' + esc(g.name) + '</span>' +
-      '<span class="ab-meta">' + esc(meta) + '</span>' +
+      '<span class="lib-meta">' + esc(meta) + '</span></div>';
-      '<span class="ab-bar"><span style="width:' + (total ? Math.round(100 * heard / total) : 0) + '%"></span></span></span></button>';
   }
 
   const foldersOf = b => (b.folders && b.folders.length ? b.folders : (b.folder ? [b.folder] : []));
@@ -311,32 +379,24 @@
     return {folders: names.map(n => ({name: n, count: counts.get(n)})), none: names.length ? none : 0};
   }
 
-  function gridHtml(books) {
-    const here = books.filter(b => st.folder == null ? true
-      : st.folder === '' ? !foldersOf(b).length : foldersOf(b).indexOf(st.folder) >= 0);
-    const list = entries(here);
-    if (st.series != null) {
-      const g = list.filter(e => e.kind === 'series' && e.key === st.series)[0];
-      if (g) {
-        return '<div class="ab-place"><button type="button" class="ab-back" data-ab="up">‹ Hörbücher</button>' +
-          '<strong>' + esc(g.name) + '</strong></div>' +
-          '<div class="ab-grid">' + g.books.map(b => bookCard(b, b._volume)).join('') + '</div>';
-      }
-      st.series = null;
-    }
-    return '<div class="ab-chips"></div><div class="ab-grid">' +
-      list.map(e => e.kind === 'series' ? seriesCard(e) : bookCard(e.b)).join('') + '</div>';
-  }
-
   function drawChips(books) {
-    const el = shelf.querySelector('.ab-chips');
+    const el = shelf.querySelector('.lib-chips');
     if (!el || !window.DioraFolders) return;
+    if (!books.length) { el.innerHTML = ''; return; }
     const c = folderChips(books);
-    if (!c.folders.length) { el.innerHTML = ''; return; }
     window.DioraFolders.chips(el, {
+      lead: [
+        {id: 'started', label: 'Angefangen', count: books.filter(started).length, on: st.status === 'started'},
+        {id: 'unheard', label: 'Ungehört', count: books.filter(unheard).length, on: st.status === 'unheard'},
+        {id: 'heard', label: 'Gehört', count: books.filter(heard).length, on: st.status === 'heard'},
+      ],
       folders: c.folders, active: st.folder, none: c.none,
-      onPick(name) { st.folder = name; renderShelf(); },
+      onPick(p) {
-      onManage: manageFolders,
+        if (p && p.lead) st.status = st.status === p.lead ? 'all' : p.lead;
+        else st.folder = p;
+        renderShelf();
+      },
+      onManage: c.folders.length ? manageFolders : null,
     });
   }
 
@@ -442,7 +502,7 @@
     const d = st.detail && st.detail.book.feed_id === st.open ? st.detail : null;
     const fromList = (st.books || []).filter(b => b.feed_id === st.open)[0];
     const b = d ? d.book : fromList;
-    if (!b) { view.innerHTML = '<div class="ab-top"><button type="button" class="ab-back" data-ab="close">‹ Bücher</button></div><p class="ab-empty">Wird geladen …</p>'; return; }
+    if (!b) { view.innerHTML = '<div class="ab-top"><button type="button" class="ab-back" data-ab="close">‹ Hörbücher</button></div><p class="ab-empty">Wird geladen …</p>'; return; }
     const now = nowInBook();
     const marks = d ? d.bookmarks : [];
     const dlMap = d ? downloads() : null;
@@ -452,7 +512,7 @@
       return ch ? 'Kapitel ' + (ch.index + 1) : '';
     };
     view.innerHTML =
-      '<div class="ab-top"><button type="button" class="ab-back" data-ab="close">‹ Bücher</button>' +
+      '<div class="ab-top"><button type="button" class="ab-back" data-ab="close">‹ Hörbücher</button>' +
       '<button type="button" class="ab-icon" data-ab="menu" aria-label="Optionen">⋮</button></div>' +
       '<div class="ab-menu" hidden>' +
       '<button type="button" data-ab="unplayed">Alles als ungehört markieren</button>' +
@@ -763,8 +823,9 @@
   }
 
   shelf.addEventListener('click', e => {
-    const t = e.target.closest('button, .ab-continue');
+    const t = e.target.closest('button, .lib-continue, .lib-card');
     if (!t) return;
+    if (t.dataset.ab === 'search') { st.searching = true; renderShelf(); return; }
     if (t.dataset.abGo != null) { e.stopPropagation(); goOn(Number(t.dataset.abGo)); return; }
     if (t.dataset.abOpen != null) { openBook(Number(t.dataset.abOpen)); return; }
     if (t.dataset.abSeries != null) { st.series = t.dataset.abSeries; renderShelf(); return; }
@@ -792,7 +853,18 @@
     if (t.dataset.ab === 'pick') { st.picking = !st.picking; st.pickFocus = st.picking; renderShelf(); return; }
     if (t.dataset.ab === 'pick-close') { st.picking = false; renderShelf(); }
   });
+  shelf.addEventListener('change', e => {
+    if (!e.target.classList.contains('lib-sort')) return;
+    st.sort = e.target.value;
+    cached('sort', st.sort);
+    renderShelf();
+  });
   shelf.addEventListener('input', e => {
+    if (e.target.classList.contains('lib-q')) {
+      st.q = e.target.value;
+      renderShelf();
+      return;
+    }
     if (!e.target.classList.contains('ab-pick-q')) return;
     st.pickQuery = e.target.value;
     const box = shelf.querySelector('.ab-pick-list');

static/js/covers.js

@@ -122,16 +122,20 @@
 
   async function pdfCover(bytes) {
     if (typeof pdfjsLib === 'undefined') return null;
-    const pdf = await pdfjsLib.getDocument({data: new Uint8Array(bytes)}).promise;
+    const pdf = await (typeof _pdfLoad === 'function' ? _pdfLoad(new Uint8Array(bytes))
+      : pdfjsLib.getDocument({data: new Uint8Array(bytes)})).promise;
+    const canvas = document.createElement('canvas');
+    try {
       const page = await pdf.getPage(1);
       const v = page.getViewport({scale: 1});
       const scale = Math.min(MAX_W / v.width, MAX_H / v.height);
       const viewport = page.getViewport({scale});
-    const canvas = document.createElement('canvas');
       canvas.width = Math.round(viewport.width);
       canvas.height = Math.round(viewport.height);
       await page.render({canvasContext: canvas.getContext('2d'), viewport}).promise;
-    pdf.destroy();
+    } finally {
+      try { pdf.destroy(); } catch (e) {}
+    }
     return new Promise(resolve => canvas.toBlob(resolve, 'image/jpeg', QUALITY));
   }
 

static/js/feeds.js

@@ -603,59 +603,35 @@
 
   // ---- article view
 
-  let reader = null;      // the overlay element
+  // ---- reading: an article opens in the book reader (65-article.js, DioraArticle) —
+  // the same typography, colours, pages, search, read-aloud and full screen as books.
+
   let readerList = [];    // what "next" walks through (blog items)
   let readerStory = null;
   let readerItem = null;
-
+  let reading = false;    // an article of ours is in the reader
-  function fontSize() {
-    let v = 18;
-    try { v = parseInt(localStorage.getItem('diora_feeds_font') || '18', 10) || 18; } catch (e) {}
-    return Math.max(14, Math.min(26, v));
-  }
-
-  function ensureReader() {
-    if (reader) return reader;
-    reader = document.createElement('div');
-    reader.id = 'fd-article';
-    reader.hidden = true;
-    reader.addEventListener('click', onReaderClick);
-    reader.addEventListener('scroll', () => {
-      clearTimeout(progressTimer);
-      progressTimer = setTimeout(saveProgress, 1200);
-    }, {passive: true});
-    document.body.appendChild(reader);
-    return reader;
-  }
 
   function readerOpen() {
-    return !!(reader && !reader.hidden);
+    return reading && !!(window.DioraArticle && DioraArticle.isOpen());
   }
 
   function closeReader() {
-    if (!reader) return;
+    if (window.DioraArticle) DioraArticle.close();
-    saveProgress();
+  }
-    reader.hidden = true;
+
-    reader.innerHTML = '';
+  function articleClosed() {
-    document.documentElement.classList.remove('fd-reading');
+    reading = false;
     readerItem = readerStory = null;
     render();
   }
 
-  // How far the open article is read (0-1); saved now and then and when leaving it.
+  // How far the open article is read (0-1, from the reader); saved now and then and when leaving it.
   let progressSent = 0;
-  let progressTimer = null;
 
-  function readProgress() {
+  function saveProgress(f) {
-    if (!reader || reader.hidden) return 0;
-    const room = reader.scrollHeight - reader.clientHeight;
-    return room <= 40 ? 1 : Math.min(1, reader.scrollTop / room);
-  }
-
-  function saveProgress() {
     const i = readerItem;
     if (!i) return;
-    const p = Math.round(readProgress() * 100) / 100;
+    const p = Math.round(f * 100) / 100;
     if (Math.abs(p - progressSent) < 0.02 && !(p >= 0.9 && !i.read)) return;
     progressSent = p;
     i.progress = p;
@@ -667,51 +643,68 @@
     api('/feeds/items/' + i.id + '/progress/', {progress: p}).catch(() => {});
   }
 
-  async function openItem(id, story) {
+  // Opens a link outside, like the article's own links (the app hands them to the browser).
-    ensureReader();
+  function openOutside(url) {
-    if (readerItem) saveProgress();
+    const a = document.createElement('a');
-    readerStory = story || null;
+    a.href = url;
-    reader.hidden = false;
+    a.target = '_blank';
-    document.documentElement.classList.add('fd-reading');
+    a.rel = 'noopener';
-    reader.innerHTML = readerShell('<p class="fd-note">Lädt …</p>');
+    document.body.appendChild(a);
-    reader.scrollTop = 0;
+    a.click();
-    try {
+    a.remove();
-      const data = await api('/feeds/items/' + id + '/');
-      readerItem = data.item;
-      reader.innerHTML = readerShell(articleHtml(readerItem));
-      reader.scrollTop = 0;
-      if (readerStory) loadOverview(readerStory);
-      progressSent = readerItem.progress || 0;
-      if (readerItem.kind === 'news') {
-        // News is read once opened; a blog post once read to 90 % (the server keeps where).
-        if (!readerItem.read) {
-          readerItem.read = true;
-          markRead([id], true);
-        }
-      } else {
-        if (progressSent > 0.02 && progressSent < 0.9) {
-          // Back where reading stopped (on any device).
-          requestAnimationFrame(() => { reader.scrollTop = progressSent * (reader.scrollHeight - reader.clientHeight); });
-        } else {
-          setTimeout(saveProgress, 1500);   // short enough to fit: read at once
-        }
-      }
-    } catch (e) {
-      reader.innerHTML = readerShell('<p class="fd-error">' + esc(e.message === 'Failed to fetch' ? 'Keine Verbindung.' : e.message) + '</p>');
-    }
   }
 
-  function readerShell(inner) {
+  // The article's own buttons, in the reader's header (the apps: in its menu).
-    const i = readerItem;
+  function articleActions(i) {
-    return '<div class="fd-art-bar">' +
+    return [
-      '<button type="button" class="fd-icon" data-r="close" aria-label="Zurück" title="Zurück">‹</button>' +
+      {icon: i.starred ? '★' : '☆', label: i.starred ? 'Gemerkt' : 'Merken', active: !!i.starred, action: () => toggleStar(i)},
-      '<span class="fd-art-src">' + esc(i ? i.source_title : '') + '</span>' +
-      '<button type="button" class="fd-icon" data-r="font" aria-label="Schriftgröße" title="Schriftgröße">Aa</button>' +
-      (i ? '<button type="button" class="fd-icon' + (i.starred ? ' on' : '') + '" data-r="star" aria-label="Merken" title="Merken">' + (i.starred ? '★' : '☆') + '</button>' : '') +
       // Projekte (projects.js): this article as a source of a project.
-      (i && window.DioraProjects ? '<button type="button" class="fd-icon" data-r="project" aria-label="Mit Projekt verknüpfen" title="Mit Projekt verknüpfen">◇</button>' : '') +
+      window.DioraProjects && {icon: '◇', label: 'Mit Projekt verknüpfen', action: () => window.DioraProjects.link('article', i.id, {
-      (i && i.url ? '<a class="fd-icon" href="' + esc(i.url) + '" target="_blank" rel="noopener" aria-label="Im Browser öffnen" title="Im Browser öffnen">↗</a>' : '') +
+        title: i.title, source: i.source_title, author: i.author, date: i.published, url: i.url, news: i.kind === 'news'})},
-      '</div><div class="fd-art" style="font-size:' + fontSize() + 'px">' + inner + '</div>';
+      i.url && {icon: '↗', label: 'Im Browser öffnen', action: () => openOutside(i.url)},
+    ];
+  }
+
+  function toggleStar(i) {
+    i.starred = !i.starred;
+    api('/feeds/items/' + i.id + '/star/', {starred: i.starred}).catch(() => {});
+    st.items.forEach(x => { if (x.id === i.id) x.starred = i.starred; });
+    if (window.DioraArticle) DioraArticle.refreshActions();
+  }
+
+  async function openItem(id, story) {
+    if (!window.DioraArticle) return;
+    readerStory = story || null;
+    let item = null;
+    let error = '';
+    try {
+      item = (await api('/feeds/items/' + id + '/')).item;
+    } catch (e) {
+      error = e.message === 'Failed to fetch' ? 'Keine Verbindung.' : e.message;
+    }
+    readerItem = item;
+    reading = true;
+    if (!item) {
+      DioraArticle.open({title: 'Artikel', html: '<p class="fd-error">' + esc(error) + '</p>',
+        onClick: onReaderClick, onClose: articleClosed});
+      return;
+    }
+    progressSent = item.progress || 0;
+    await DioraArticle.open({
+      title: item.source_title || item.title,
+      html: articleHtml(item),
+      // News is read once opened; a blog post once read to 90 % (the server keeps where).
+      startFraction: item.kind === 'news' ? 0 : progressSent,
+      onProgress: f => { if (readerItem === item) saveProgress(f); },
+      onClick: onReaderClick,
+      onClose: articleClosed,
+      actions: () => articleActions(item),
+    });
+    if (readerStory) loadOverview(readerStory);
+    if (item.kind === 'news' && !item.read) {
+      item.read = true;
+      markRead([id], true);
+    }
   }
 
   function articleHtml(i) {
@@ -727,6 +720,10 @@
     if (story) h += '<div class="fd-ov">' + (story.overview ? overviewHtml(story) : '') + '</div>';
     const content = i.content || '';
     if (i.image_url && content.indexOf(i.image_url) < 0) h += '<img class="fd-art-img" alt="" src="' + esc(i.image_url) + '">';
+    if (i.episode_id) {
+      // The article of a podcast episode (docs/transcription.md): to the episode itself.
+      h += '<p class="fd-art-listen"><button type="button" class="btn" data-r="episode" data-id="' + i.episode_id + '">▶ Folge anhören</button></p>';
+    }
     h += '<div class="fd-art-body">' + (content || '<p>' + esc(i.summary) + '</p>') + '</div>';
     if (i.paywalled) {
       h += '<p class="fd-note">Nur der freie Teil des Artikels. Mit deiner Abo-Anmeldung (⚙ → ' + esc(i.source_title) +
@@ -787,7 +784,7 @@
     } catch (e) {
       return;   // offline and never fetched: the article alone
     }
-    const box = reader && !reader.hidden && readerStory === story && reader.querySelector('.fd-ov');
+    const box = readerOpen() && readerStory === story && document.querySelector('#reader-content .fd-ov');
     if (box) box.innerHTML = overviewHtml(story);
   }
 
@@ -807,37 +804,76 @@
     });
   }
 
-  function onReaderClick(e) {
+  // ---- the episode of a podcast article (docs/transcription.md)
-    const b = e.target.closest('[data-r]');
+
-    if (!b) {
+  function podcastHost() {
-      // Links in the article open outside (the app hands them to the browser).
+    return window.DioraPodcastHost || window.DioraNative || null;
-      const a = e.target.closest('.fd-art-body a[href]');
+  }
-      if (a) { a.target = '_blank'; a.rel = 'noopener'; }
+
+  function parseJson(text, fallback) {
+    try { return JSON.parse(text); } catch (e) { return fallback; }
+  }
+
+  // Play the episode from `seconds` (null: open its page, where ▶ starts it). The browser and the
+  // diora desktop have the host of podcasts-web.js; the apps' bridge has the same calls
+  // (podcastPlayEpisodeAt is the one that starts at a place; without it the episode starts at its own place).
+  async function playEpisode(episodeId, seconds) {
+    const host = podcastHost();
+    if (!host || !host.playerState) { alert('Der Podcast-Bereich ist hier nicht verfügbar.'); return; }
+    const now = parseJson(host.playerState(), {});
+    const ms = seconds == null ? null : Math.max(0, Math.round(seconds * 1000));
+    let opened = false;
+    if (now.kind === 'podcast' && now.episodeId === episodeId) {
+      if (ms != null && host.podcastSeek) host.podcastSeek(ms);
+      if (ms != null && !now.playing && host.podcastToggle) host.podcastToggle();
+      opened = ms != null;
+    } else {
+      let d;
+      try {
+        d = await api('/podcasts/episodes/' + episodeId + '/');
+      } catch (e) {
+        alert(e.message === 'Failed to fetch' ? 'Keine Verbindung.' : 'Die Folge ist nicht mehr da.');
         return;
       }
+      const row = {id: d.id, title: d.title, audio_url: d.audio_url, duration_seconds: d.duration_seconds,
+        artwork_url: d.artwork_url, pub_date: d.pub_date, position_seconds: d.position_seconds,
+        feed__id: d.feed.id, feed__title: d.feed.title, feed__artwork_url: d.feed.artwork_url};
+      closeReader();
+      if (ms != null && host.podcastPlayEpisodeAt) {
+        host.podcastPlayEpisodeAt(JSON.stringify(row), now.episodeId || 0, ms);
+        opened = true;
+      } else if (ms == null && window.__dioraPodcastOpenEpisode && window.__dioraPodcastOpenEpisode(row)) {
+        return;
+      } else if (host.podcastPlayEpisode) {
+        host.podcastPlayEpisode(JSON.stringify(row), now.episodeId || 0);
+        opened = true;
+      }
+    }
+    closeReader();
+    if (opened && window.__dioraOpenPlayer) window.__dioraOpenPlayer('podcast');
+  }
+
+  // Clicks in the article (65-article.js hands them over): its other reports, the next article.
+  function onReaderClick(e) {
+    // A time mark in a podcast article ("▶ ab 12:34", href="#t=754"; the reader keeps it as data-internal-href).
+    const t = e.target.closest('a[data-internal-href^="#t="], a[href^="#t="]');
+    if (t && readerItem && readerItem.episode_id) {
+      const sec = parseInt((t.dataset.internalHref || t.getAttribute('href')).slice(3), 10);
+      if (sec >= 0) {
+        e.preventDefault();
+        playEpisode(readerItem.episode_id, sec);
+      }
+      return;
+    }
+    const b = e.target.closest('[data-r]');
+    if (!b) return;
     const r = b.dataset.r;
-    if (r === 'close') closeReader();
+    if (r === 'episode') playEpisode(parseInt(b.dataset.id, 10), null);
     else if (r === 'item') openItem(parseInt(b.dataset.id, 10), readerStory);
     else if (r === 'story') {
       const other = readerStory && readerStory.overview && (readerStory.overview.related || []).find(x => x.id === parseInt(b.dataset.id, 10));
       if (other) openItem((other.lead || other.items[0]).id, other);
     }
-    else if (r === 'font') {
-      let v = fontSize() + 2;
-      if (v > 26) v = 14;
-      try { localStorage.setItem('diora_feeds_font', String(v)); } catch (err) {}
-      const art = reader.querySelector('.fd-art');
-      if (art) art.style.fontSize = v + 'px';
-    } else if (r === 'project' && readerItem && window.DioraProjects) {
-      window.DioraProjects.link('article', readerItem.id, {title: readerItem.title, source: readerItem.source_title,
-        author: readerItem.author, date: readerItem.published, url: readerItem.url, news: readerItem.kind === 'news'});
-    } else if (r === 'star' && readerItem) {
-      readerItem.starred = !readerItem.starred;
-      api('/feeds/items/' + readerItem.id + '/star/', {starred: readerItem.starred}).catch(() => {});
-      b.textContent = readerItem.starred ? '★' : '☆';
-      b.classList.toggle('on', readerItem.starred);
-      st.items.forEach(i => { if (i.id === readerItem.id) i.starred = readerItem.starred; });
-    }
   }
 
   async function openStory(id) {
@@ -1354,9 +1390,6 @@ und Ordner. Schlage nur Änderungen vor, die wirklich helfen; eine Quelle ohne s
     }
     return false;
   };
-  document.addEventListener('keydown', e => {
-    if (e.key === 'Escape' && readerOpen()) closeReader();
-  });
 
   // diora Reader has no tab bar of its own: a small one. With the navigation (nav.js) it
   // holds the sections of Lesen itself — a row of them to swipe is no good on e-ink.
@@ -1436,9 +1469,9 @@ und Ordner. Schlage nur Änderungen vor, die wirklich helfen; eine Quelle ohne s
     '.fd-sources { display: flex; gap: 6px; margin: 4px -18px 10px; padding: 0 18px; align-items: center; overflow-x: auto; white-space: nowrap; scrollbar-width: none; }',
     '.fd-sources::-webkit-scrollbar { display: none; }',
     '.fd-sources .fd-chip { flex: none; }',
-    '#fd-article .fd-art-more { font-family: var(--font); margin: 30px 0 4px; font-size: 12px; font-weight: 400; text-transform: uppercase; letter-spacing: 0.08em; color: var(--nui-muted, #9a9a9a); }',
+    '.fd-artview .fd-art-more { font-family: var(--font); margin: 30px 0 4px; font-size: 12px; font-weight: 400; text-transform: uppercase; letter-spacing: 0.08em; color: var(--nui-muted, #9a9a9a); }',
-    '#fd-article .fd-other { display: flex; flex-direction: column; gap: 3px; width: 100%; padding: 10px 0; background: none; border: none; border-bottom: 1px solid var(--border, #2a2a2a); color: inherit; font: inherit; font-size: 0.85em; text-align: left; cursor: pointer; }',
+    '.fd-artview .fd-other { display: flex; flex-direction: column; gap: 3px; width: 100%; padding: 10px 0; background: none; border: none; border-bottom: 1px solid var(--border, #2a2a2a); color: inherit; font: inherit; font-size: 0.85em; text-align: left; cursor: pointer; }',
-    '#fd-article .fd-other-title { font-weight: 600; line-height: 1.35; }',
+    '.fd-artview .fd-other-title { font-weight: 600; line-height: 1.35; }',
     '.fd-chip { padding: 6px 12px; border: 1px solid var(--border, #2a2a2a); border-radius: 16px; background: none; color: inherit; font: inherit; font-size: 13px; cursor: pointer; }',
     '.fd-chip.on { border-color: var(--accent, #e63946); color: var(--accent, #e63946); }',
     '.fd-chip small { color: var(--nui-muted, #9a9a9a); font-size: 11px; }',
@@ -1491,45 +1524,39 @@ und Ordner. Schlage nur Änderungen vor, die wirklich helfen; eine Quelle ohne s
     '#tab-feeds .fd-opml-row { display: flex; justify-content: space-between; align-items: center; gap: 8px; }',
     '#tab-feeds .fd-keep { display: flex; align-items: center; gap: 8px; margin-top: 14px; }',
     '#tab-feeds .fd-file { display: inline-block; }',
-    // The article: a page over everything, like the book reader.
+    // The article, in the book reader (65-article.js).
-    '#fd-article { position: fixed; top: 0; right: 0; bottom: 0; left: 0; z-index: 900; background: var(--bg, #000); color: var(--fg, #fff); overflow-y: auto; -webkit-overflow-scrolling: touch; }',
+    '.fd-artview .fd-art-title { font-size: 1.45em; line-height: 1.25; margin: 10px 0 6px; font-weight: 700; }',
-    '#fd-article[hidden] { display: none; }',
+    '.fd-artview .fd-art-meta { font-family: var(--font); font-size: 12px; color: var(--nui-muted, #9a9a9a); margin: 0 0 14px; }',
-    '#fd-article .fd-art-bar { position: sticky; top: 0; display: flex; align-items: center; gap: 2px; padding: 4px 6px; background: var(--bg, #000); border-bottom: 1px solid var(--border, #2a2a2a); padding-top: calc(4px + env(safe-area-inset-top, 0px)); }',
+    '.fd-artview .fd-art-img, .fd-artview .fd-art-body img { display: block; max-width: 100%; height: auto; margin: 12px auto; border-radius: 4px; }',
-    '#fd-article .fd-art-src { flex: 1; min-width: 0; overflow: hidden; text-overflow: ellipsis; white-space: nowrap; font-size: 13px; color: var(--nui-muted, #9a9a9a); }',
+    '.fd-artview .fd-art-listen { margin: 0 0 12px; }',
-    '#fd-article .fd-art { max-width: 68ch; margin: 0 auto; padding: 12px 18px 48px; font-family: "Source Serif 4", Georgia, serif; font-weight: 380; line-height: 1.65; }',
+    '.fd-artview .fd-art-body a { color: var(--accent, #e63946); }',
-    '#fd-article .fd-art-title { font-size: 1.45em; line-height: 1.25; margin: 10px 0 6px; font-weight: 700; }',
+    '.fd-artview .fd-art-body figure { margin: 12px 0; }',
-    '#fd-article .fd-art-meta { font-family: var(--font); font-size: 12px; color: var(--nui-muted, #9a9a9a); margin: 0 0 14px; }',
+    '.fd-artview .fd-art-body figcaption { font-size: 0.8em; color: var(--nui-muted, #9a9a9a); }',
-    '#fd-article .fd-art-img, #fd-article .fd-art-body img { display: block; max-width: 100%; height: auto; margin: 12px auto; border-radius: 4px; }',
+    '.fd-artview .fd-art-body blockquote { margin: 12px 0; padding-left: 14px; border-left: 3px solid var(--border-strong, #444); }',
-    '#fd-article .fd-art-body a { color: var(--accent, #e63946); }',
+    '.fd-artview .fd-art-body pre { white-space: pre-wrap; font-size: 0.85em; }',
-    '#fd-article .fd-art-body figure { margin: 12px 0; }',
+    '.fd-artview .fd-art-body table { display: block; overflow-x: auto; border-collapse: collapse; font-size: 0.85em; }',
-    '#fd-article .fd-art-body figcaption { font-size: 0.8em; color: var(--nui-muted, #9a9a9a); }',
+    '.fd-artview .fd-art-body td, .fd-artview .fd-art-body th { border: 1px solid var(--border, #333); padding: 4px 6px; }',
-    '#fd-article .fd-art-body blockquote { margin: 12px 0; padding-left: 14px; border-left: 3px solid var(--border-strong, #444); }',
+    '.fd-artview .fd-sources { font-family: var(--font); }',
-    '#fd-article .fd-art-body pre { white-space: pre-wrap; font-size: 0.85em; }',
-    '#fd-article .fd-art-body table { display: block; overflow-x: auto; border-collapse: collapse; font-size: 0.85em; }',
-    '#fd-article .fd-art-body td, #fd-article .fd-art-body th { border: 1px solid var(--border, #333); padding: 4px 6px; }',
-    '#fd-article .fd-sources { font-family: var(--font); }',
     '#tab-feeds .fd-topics .fd-note { margin-right: 2px; }',
     '#tab-feeds .fd-topics small { opacity: 0.6; }',
-    '#fd-article .fd-ov { font-size: 0.9em; }',
+    '.fd-artview .fd-ov { font-size: 0.9em; }',
-    '#fd-article .fd-ov-box { margin: 0 0 18px; padding: 10px 14px; border: 1px solid var(--border, #2a2a2a); border-radius: 6px; }',
+    '.fd-artview .fd-ov-box { margin: 0 0 18px; padding: 10px 14px; border: 1px solid var(--border, #2a2a2a); border-radius: 6px; }',
-    '#fd-article .fd-ov-box summary { font-family: var(--font); font-size: 12px; text-transform: uppercase; letter-spacing: 0.08em; color: var(--nui-muted, #9a9a9a); cursor: pointer; }',
+    '.fd-artview .fd-ov-box summary { font-family: var(--font); font-size: 12px; text-transform: uppercase; letter-spacing: 0.08em; color: var(--nui-muted, #9a9a9a); cursor: pointer; }',
-    '#fd-article .fd-ov p { margin: 8px 0; line-height: 1.45; }',
+    '.fd-artview .fd-ov p { margin: 8px 0; line-height: 1.45; }',
-    '#fd-article .fd-ov h4 { font-family: var(--font); font-size: 12px; font-weight: 400; text-transform: uppercase; letter-spacing: 0.08em; color: var(--nui-muted, #9a9a9a); margin: 14px 0 4px; }',
+    '.fd-artview .fd-ov h4 { font-family: var(--font); font-size: 12px; font-weight: 400; text-transform: uppercase; letter-spacing: 0.08em; color: var(--nui-muted, #9a9a9a); margin: 14px 0 4px; }',
-    '#fd-article .fd-ov-src { font-family: var(--font); font-size: 11px; color: var(--nui-muted, #9a9a9a); white-space: nowrap; }',
+    '.fd-artview .fd-ov-src { font-family: var(--font); font-size: 11px; color: var(--nui-muted, #9a9a9a); white-space: nowrap; }',
-    '#fd-article .fd-ov-words { font-family: var(--font); font-size: 13px; }',
+    '.fd-artview .fd-ov-words { font-family: var(--font); font-size: 13px; }',
-    '#fd-article .fd-ov-rel { margin: 0 0 18px; }',
+    '.fd-artview .fd-ov-rel { margin: 0 0 18px; }',
-    '#fd-article .fd-art-foot { display: flex; flex-wrap: wrap; gap: 10px; justify-content: space-between; margin-top: 28px; font-family: var(--font); }',
+    '.fd-artview .fd-art-foot { display: flex; flex-wrap: wrap; gap: 10px; justify-content: space-between; margin-top: 28px; font-family: var(--font); }',
-    'html.fd-reading #nui-dock { display: none; }',
     '@media (hover: hover) { #tab-feeds .fd-card:hover, #tab-feeds .fd-row:hover, #tab-feeds .fd-post:hover, #tab-feeds .fd-src:hover { background: var(--hover, #111); } }',
     // Bright pages and e-ink: black on white, no grey boxes, no thumbnails to flicker.
     'body.bright-bg #tab-feeds .fd-brief, body.eink #tab-feeds .fd-brief { background: #fff; }',
     'body.eink #tab-feeds .fd-thumb { display: none; }',
     'body.eink #tab-feeds .fd-seg button.on, body.eink #tab-feeds .fd-badge { background: #000; color: #fff; }',
     'body.eink .fd-row-meta, body.eink .fd-note, body.eink #tab-feeds .fd-card-sum, body.eink #tab-feeds .fd-read .fd-card-title, body.eink #tab-feeds .fd-read .fd-row-title { color: #333; }',
-    'body.eink #fd-article, body.eink #fd-article .fd-art-bar { background: #fff; color: #000; }',
+    'body.eink .fd-artview .fd-ov-box { border-color: #000; }',
-    'body.eink #fd-article .fd-ov-box { border-color: #000; }',
+    'body.eink .fd-artview .fd-ov-src, body.eink .fd-artview .fd-ov h4, body.eink .fd-artview .fd-ov-box summary { color: #333; }',
-    'body.eink #fd-article .fd-ov-src, body.eink #fd-article .fd-ov h4, body.eink #fd-article .fd-ov-box summary { color: #333; }',
+    'body.eink .fd-artview .fd-art-body a, body.eink .fd-chip.on, body.eink .fd-icon.on, body.eink #tab-feeds .fd-brief-head h2 { color: #000; }',
-    'body.eink #fd-article .fd-art-body a, body.eink .fd-chip.on, body.eink .fd-icon.on, body.eink #tab-feeds .fd-brief-head h2 { color: #000; }',
   ].join('\n');
   document.head.appendChild(style);
 })();

static/js/folders.js

@@ -22,22 +22,45 @@
   // lead: chips before the folders (the library's Angefangen/Ungelesen …): [{id, label, count, on}],
   // picked with onPick({lead: id}). active: null = all, '' = without a folder, else a name.
 
+  // The folders fold away (a "Ordner" chip), per area (the list's element id), remembered on
+  // this device; folded, only the folder chosen stays in sight. Folded until opened.
+  function areaKey(el) {
+    const owner = el.id ? el : el.parentElement && el.parentElement.closest('[id]');
+    return 'diora_folders_folded_' + (owner ? owner.id : 'x');
+  }
+
+  function folded(el) {
+    try { return localStorage.getItem(areaKey(el)) !== '0'; } catch (e) { return true; }
+  }
+
   function chips(el, o) {
     const lead = o.lead || [];
     const leadOn = lead.some(c => c.on);
+    const folders = o.folders || [];
+    const fold = folders.length > 0 && folded(el);
     let h = '<div class="dfo-row" role="toolbar">';
     if (!lead.length) h += chip('dfo-all', 'Alle', null, o.active == null && !leadOn, 'data-dfo-all="1"');
     lead.forEach(c => { h += chip('', c.label, c.count, !!c.on, 'data-dfo-lead="' + esc(c.id) + '"'); });
-    if (lead.length && (o.folders || []).length) h += '<span class="dfo-sep" aria-hidden="true"></span>';
+    if (lead.length && folders.length) h += '<span class="dfo-sep" aria-hidden="true"></span>';
-    (o.folders || []).forEach(f => {
+    if (folders.length) {
+      h += '<button type="button" class="dfo-chip dfo-fold" data-dfo-fold="1" aria-expanded="' + !fold + '" title="' +
+        (fold ? 'Ordner zeigen' : 'Ordner einklappen') + '">Ordner ' + (fold ? '▸ <span class="dfo-count">' + folders.length + '</span>' : '▾') + '</button>';
+    }
+    folders.forEach(f => {
+      if (fold && o.active !== f.name) return;
       h += chip('', f.name, f.count, o.active === f.name, 'data-dfo-folder="' + esc(f.name) + '"');
     });
-    if (o.none) h += chip('dfo-none', 'Ohne Ordner', o.none, o.active === '', 'data-dfo-folder=""');
+    if (o.none && (!fold || o.active === '')) h += chip('dfo-none', 'Ohne Ordner', o.none, o.active === '', 'data-dfo-folder=""');
-    if (o.onManage) h += '<button type="button" class="dfo-chip dfo-manage" data-dfo-manage="1" title="Ordner verwalten" aria-label="Ordner verwalten">⋯</button>';
+    if (o.onManage && !fold) h += '<button type="button" class="dfo-chip dfo-manage" data-dfo-manage="1" title="Ordner verwalten" aria-label="Ordner verwalten">⋯</button>';
     el.innerHTML = h + '</div>';
     el.onclick = e => {
       const b = e.target.closest('button');
       if (!b) return;
+      if (b.dataset.dfoFold) {
+        try { localStorage.setItem(areaKey(el), fold ? '0' : '1'); } catch (err) {}
+        chips(el, o);
+        return;
+      }
       if (b.dataset.dfoManage) { o.onManage(); return; }
       if (b.dataset.dfoAll) { o.onPick(null); return; }
       if (b.dataset.dfoLead != null) { o.onPick({lead: b.dataset.dfoLead}); return; }
@@ -188,6 +211,7 @@
     '.dfo-count { opacity: .7; font-size: 12px; }',
     '.dfo-none { font-style: italic; }',
     '.dfo-manage { min-width: 34px; padding: 0 8px; }',
+    '.dfo-fold { border-style: dashed; }',
     '.dfo-sep { flex: none; width: 1px; height: 20px; background: var(--border-strong, #444); margin: 0 2px; }',
     // Above the book view (900), below the page's dialogs (990): a name is asked for on top.
     '.dfo-scrim { position: fixed; top: 0; right: 0; bottom: 0; left: 0; z-index: 950; background: var(--scrim, rgba(0,0,0,.5)); display: flex; align-items: flex-end; justify-content: center; }',

static/js/podcast-article.js

@@ -0,0 +1,380 @@
+// ---------------------------------------------------------------------------
+// Podcast articles on the device (docs/transcription.md): the transcript of an episode goes to
+// Claude with the user's own key (DioraAI, docs/ai.md) and comes back as an article that is saved
+// to the episode and to the blog "Podcast-Artikel". Also the picker-up of open wishes: the server
+// only remembers that an article was wanted; a device with a key writes it.
+//
+// An ES module like ai.js. window.DioraPodcastArticle = {write, pickUp, watch, ...}; the helpers
+// are exported too (tools/tests/podcast-article.test.mjs runs them under Node).
+// ---------------------------------------------------------------------------
+
+export const MAX_TOKENS_IN = 120000;     // about nine hours of speech; more stops with a message
+export const GROUP_SECONDS = 30;
+const SHOWNOTES_MAX = 6000;              // characters of the shownotes handed to Claude
+const POLL_MS = 5 * 60 * 1000;
+
+const ai = () => globalThis.DioraAI;
+
+export const SYSTEM = `You turn the transcript of a podcast episode into an article that reads as if the people
+speaking had written it themselves instead of recording the episode.
+
+The reader wants to decide whether this episode still has something to teach them. They do not
+want a summary; they want to understand what is argued here well enough to judge it for
+themselves. Write for that reader.
+
+Perspective. Write from the speakers' standpoint and state their positions directly, as their
+own claims. Never distance yourself from them: no "im Podcast wird die These aufgestellt", "der
+Sprecher meint", "in dieser Folge geht es um", "die Gäste diskutieren", "laut X". If the episode
+is one person speaking, write as that person argues, in the third person or impersonal voice,
+but without reporting verbs. Do not add your own opinion, criticism or fact checks, and do not
+soften what is said.
+
+Conversations. Interviews, conversations and discussions are condensed into their core theses;
+the form of question and answer disappears. Where people hold different positions, keep them
+apart and name who holds which ("Zuckermann hält dagegen, dass …"), but do not turn it into
+minutes. Questions from a host matter only where they shape the argument. When the transcript
+has no speaker names, infer them from the context (introductions, the shownotes, how people
+address each other); if you can't, do not invent names.
+
+Substance over summary. Give the arguments with their reasons, examples and evidence, not just
+the conclusions. Keep figures, dates, quotations that carry an argument (short, in quotation
+marks, attributed), and every work mentioned (books, articles, studies, films) with author and
+title. Make clear what is new, distinctive or surprising in the episode and where it restates
+what is commonly said, by how you weight and order it, without grading it.
+
+Leave out: advertising and sponsors, appeals for donations, subscriptions or ratings, notes on
+the podcast itself (schedules, technology, community), greetings, small talk and digressions that
+lead nowhere.
+
+Form. Write in the language of the episode. \`title\`: a title for the article, not the episode
+title repeated. \`summary\`: three to six sentences up front that state the central thesis and how
+the episode gets there. \`sections\`: as many as the content needs (usually four to ten), each
+with a heading that says something (not "Einleitung"), \`start\` = the second in the episode where
+its content begins (from the [mm:ss] marks), and \`body_markdown\` in plain paragraphs; lists only
+where the speakers list things themselves. Follow the order of the episode unless a different
+order makes the argument clearer. Length: as long as needed for the reader to judge the episode,
+roughly a tenth of the transcript, at most about 5,000 words.
+
+The transcript was made by speech recognition: fix obvious mishearings, especially of names,
+using the metadata, and don't quote garbled passages.`;
+
+export const SCHEMA = {
+  "type": "object",
+  "additionalProperties": false,
+  "required": [
+    "title",
+    "summary",
+    "kind",
+    "speakers",
+    "sections"
+  ],
+  "properties": {
+    "title": {
+      "type": "string"
+    },
+    "summary": {
+      "type": "string"
+    },
+    "kind": {
+      "type": "string",
+      "enum": [
+        "Interview",
+        "Gespräch",
+        "Monolog",
+        "Diskussion"
+      ]
+    },
+    "speakers": {
+      "type": "array",
+      "items": {
+        "type": "string"
+      }
+    },
+    "sections": {
+      "type": "array",
+      "items": {
+        "type": "object",
+        "additionalProperties": false,
+        "required": [
+          "heading",
+          "start",
+          "body_markdown"
+        ],
+        "properties": {
+          "heading": {
+            "type": "string"
+          },
+          "start": {
+            "type": "integer"
+          },
+          "body_markdown": {
+            "type": "string"
+          }
+        }
+      }
+    }
+  }
+};
+
+// ---- the input text -------------------------------------------------------------------------
+
+export const pad = n => String(n).padStart(2, '0');
+
+// mm:ss (minutes keep counting past 59: "125:03"), the form of the [mm:ss] marks Claude reads.
+export function mmss(sec) {
+  const s = Math.max(0, Math.floor(Number(sec) || 0));
+  return pad(Math.floor(s / 60)) + ':' + pad(s % 60);
+}
+
+// h:mm:ss from an hour on, else mm:ss: the label of the "▶ ab …" links.
+export function clock(sec) {
+  const s = Math.max(0, Math.floor(Number(sec) || 0));
+  return s >= 3600 ? Math.floor(s / 3600) + ':' + pad(Math.floor(s / 60) % 60) + ':' + pad(s % 60) : mmss(s);
+}
+
+// One line per group of about GROUP_SECONDS: "[mm:ss] Sprecher: text", or "[mm:ss] text" without
+// speakers. A new speaker always starts a new line.
+export function transcriptToText(segments) {
+  const lines = [];
+  let cur = null;
+  for (const seg of segments || []) {
+    const text = String((seg && seg.text) || '').replace(/\s+/g, ' ').trim();
+    if (!text) continue;
+    const start = Number(seg.start) || 0;
+    const speaker = String(seg.speaker || '').trim();
+    if (!cur || speaker !== cur.speaker || start - cur.start >= GROUP_SECONDS) {
+      cur = {start, speaker, parts: []};
+      lines.push(cur);
+    }
+    cur.parts.push(text);
+  }
+  return lines.map(l => '[' + mmss(l.start) + '] ' + (l.speaker ? l.speaker + ': ' : '') + l.parts.join(' ')).join('\n');
+}
+
+export const estimateTokens = text => Math.ceil(String(text || '').length / 4);
+
+// HTML (the shownotes) as plain text.
+export function htmlToText(html) {
+  return String(html || '')
+    .replace(/<(script|style)[\s\S]*?<\/\1>/gi, '')
+    .replace(/<\s*br\s*\/?>/gi, '\n')
+    .replace(/<\/(p|div|li|h[1-6]|blockquote|tr)\s*>/gi, '\n')
+    .replace(/<li[^>]*>/gi, '- ')
+    .replace(/<[^>]+>/g, '')
+    .replace(/&nbsp;/g, ' ').replace(/&lt;/g, '<').replace(/&gt;/g, '>').replace(/&quot;/g, '"')
+    .replace(/&#39;|&apos;/g, "'").replace(/&amp;/g, '&')
+    .replace(/[ \t]+/g, ' ').replace(/ *\n */g, '\n').replace(/\n{3,}/g, '\n\n').trim();
+}
+
+export function metaBlock(ep) {
+  const rows = [];
+  const feed = ep.feed || {};
+  if (feed.title) rows.push('Podcast: ' + feed.title);
+  if (ep.author || feed.author) rows.push('Autor: ' + (ep.author || feed.author));
+  rows.push('Folge: ' + (ep.title || ''));
+  if (ep.pub_date) rows.push('Datum: ' + String(ep.pub_date).slice(0, 10));
+  if (ep.duration_seconds) rows.push('Dauer: ' + clock(ep.duration_seconds));
+  let notes = htmlToText(ep.description);
+  if (notes.length > SHOWNOTES_MAX) notes = notes.slice(0, SHOWNOTES_MAX) + ' …';
+  if (notes) rows.push('Shownotes:\n' + notes);
+  const chapters = (ep.chapters || []).filter(c => c && c.title);
+  if (chapters.length) rows.push('Kapitel:\n' + chapters.map(c => '[' + mmss(c.start) + '] ' + c.title).join('\n'));
+  return rows.join('\n');
+}
+
+function userError(message) {
+  const e = new Error(message);
+  e.diora = true;
+  return e;
+}
+
+// The prompt (user turn), or an Error with a German message when it is too long.
+export function buildPrompt(ep, transcript) {
+  const text = transcriptToText(transcript && transcript.segments);
+  if (!text) throw userError('Das Transkript ist leer.');
+  const prompt = metaBlock(ep) + '\n\nTranskript:\n' + text;
+  const tokens = estimateTokens(prompt + SYSTEM);
+  if (tokens > MAX_TOKENS_IN) {
+    throw userError('Dieses Transkript ist für einen Artikel zu lang (etwa ' + Math.round(tokens / 1000) +
+      '.000 Tokens, mehr als ' + MAX_TOKENS_IN / 1000 + '.000 gehen nicht, das sind rund neun Stunden).');
+  }
+  return prompt;
+}
+
+// ---- markdown to HTML (paragraphs, emphasis, lists, block quotes; everything else is escaped) ---
+
+export const esc = s => String(s == null ? '' : s).replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/"/g, '&quot;');
+
+// Escaped first, then the emphasis marks: model output can't bring its own tags.
+function inline(s) {
+  return esc(s)
+    .replace(/\*\*(?=\S)(.+?)(?<=\S)\*\*/g, '<strong>$1</strong>')
+    .replace(/(^|[^*\w])\*(?=[^\s*])(.+?)(?<=[^\s*])\*(?![*\w])/g, '$1<em>$2</em>')
+    .replace(/(^|[^_\w])_(?=[^\s_])(.+?)(?<=[^\s_])_(?![_\w])/g, '$1<em>$2</em>');
+}
+
+export function markdownToHtml(md) {
+  const blocks = String(md || '').replace(/\r\n?/g, '\n').split(/\n\s*\n/);
+  const out = [];
+  for (const block of blocks) {
+    const lines = block.split('\n').map(l => l.replace(/\s+$/, '')).filter(l => l.trim());
+    if (!lines.length) continue;
+    if (lines.every(l => /^\s*>/.test(l))) {
+      out.push('<blockquote><p>' + inline(lines.map(l => l.replace(/^\s*>\s?/, '')).join(' ')) + '</p></blockquote>');
+    } else if (lines.every(l => /^\s*([-*+]|\d+[.)])\s+/.test(l))) {
+      const tag = /^\s*\d/.test(lines[0]) ? 'ol' : 'ul';
+      out.push('<' + tag + '>' + lines.map(l => '<li>' + inline(l.replace(/^\s*([-*+]|\d+[.)])\s+/, '')) + '</li>').join('') + '</' + tag + '>');
+    } else {
+      out.push('<p>' + inline(lines.map(l => l.trim()).join(' ')) + '</p>');
+    }
+  }
+  return out.join('\n');
+}
+
+// The article: the summary as a lead paragraph, per section a heading, the "▶ ab 12:34" link
+// (feeds.js plays the item's episode from `#t=`) and the body.
+export function articleHtml(answer) {
+  const parts = [];
+  if (answer.summary) parts.push(markdownToHtml(answer.summary));
+  for (const s of answer.sections || []) {
+    const t = Math.max(0, Math.floor(Number(s.start) || 0));
+    parts.push('<h2>' + esc(s.heading) + '</h2>');
+    parts.push('<p><a href="#t=' + t + '">▶ ab ' + clock(t) + '</a></p>');
+    parts.push(markdownToHtml(s.body_markdown));
+  }
+  return parts.filter(Boolean).join('\n');
+}
+
+// ---- talking to diora ---------------------------------------------------------------------------
+
+function csrf() {
+  try {
+    if (typeof getCsrfToken === 'function') return getCsrfToken();
+  } catch (e) { /* fall through */ }
+  const m = typeof document !== 'undefined' && /(?:^|; )csrftoken=([^;]+)/.exec(document.cookie || '');
+  return m ? decodeURIComponent(m[1]) : '';
+}
+
+const send = (url, method, body) => fetch(url, {
+  method, headers: {'Content-Type': 'application/json', 'X-CSRFToken': csrf()},
+  body: body === undefined ? undefined : JSON.stringify(body),
+});
+
+// This device's name for reserving a wish (/article/claim/), kept per browser or app.
+let memDevice = '';
+export function deviceId() {
+  try {
+    let id = localStorage.getItem('diora-device-id');
+    if (!id) { id = 'd-' + Math.random().toString(36).slice(2, 12); localStorage.setItem('diora-device-id', id); }
+    return id;
+  } catch (e) {
+    return memDevice || (memDevice = 'd-' + Math.random().toString(36).slice(2, 12));
+  }
+}
+export const claimWish = id => send('/podcasts/episodes/' + id + '/article/claim/', 'POST', {device: deviceId()});
+export const release = id => send('/podcasts/episodes/' + id + '/article/claim/', 'DELETE', {device: deviceId()});
+
+async function getJson(url) {
+  const r = await fetch(url, {cache: 'no-store'});
+  return {status: r.status, ok: r.ok, data: r.ok ? await r.json() : null};
+}
+
+// Write the article of an episode and save it. Resolves {item_id, title}.
+// onState({phase: 'loading'|'writing'|'saving', text, words}); signal aborts.
+export async function write(episodeId, {onState, signal} = {}) {
+  const say = (phase, text, words) => { if (onState) { try { onState({phase, text, words}); } catch (e) { /* display only */ } } };
+  const AI = ai();
+  if (!AI || !AI.available()) throw userError('Die KI-Funktionen sind auf diesem Gerät nicht verfügbar.');
+  say('loading', 'Lade Transkript …');
+  const base = '/podcasts/episodes/' + encodeURIComponent(episodeId) + '/';
+  const [detail, tr] = await Promise.all([getJson(base), getJson(base + 'transcript/')]);
+  if (!detail.ok) throw userError('Die Folge ließ sich nicht laden (' + detail.status + ').');
+  if (!tr.ok) throw userError(tr.status === 404 ? 'Zu dieser Folge gibt es noch kein Transkript.' : 'Das Transkript ließ sich nicht laden (' + tr.status + ').');
+  const prompt = buildPrompt(detail.data, tr.data);
+  const answer = await new Promise((resolve, reject) => {
+    // The key first (the settings open when there is none, as for every KI feature).
+    AI.withKey(() => AI.ask({
+      system: SYSTEM, prompt, schema: SCHEMA, effort: 'medium', maxTokens: 32000, signal,
+      onProgress: words => say('writing', 'Schreibe Artikel … ' + words.toLocaleString('de-DE') + ' Wörter', words),
+    }).then(resolve, reject));
+  });
+  if (!answer || !Array.isArray(answer.sections) || !answer.sections.length) throw userError('Die Antwort enthielt keinen Artikel.');
+  say('saving', 'Speichere Artikel …');
+  const title = answer.title || detail.data.title;
+  const r = await send(base + 'article/', 'POST', {
+    title,
+    summary: answer.summary || '',
+    body_html: articleHtml(answer),
+    model: AI.modelName ? AI.modelName() : 'claude-opus-5-5',
+    kind: answer.kind || '',
+  });
+  if (!r.ok) throw userError('Der Artikel ließ sich nicht speichern (' + r.status + ').');
+  const saved = await r.json();
+  return {item_id: saved.item_id, title};
+}
+
+// "Artikel fertig": a system notification on Android (the bridge), else a toast in the page.
+export function notifyDone(title, episodeId) {
+  const N = globalThis.DioraNative;
+  try {
+    if (N && typeof N.notify === 'function') { N.notify('Artikel fertig', title, 'episode:' + episodeId); return; }
+  } catch (e) { /* fall back to the toast */ }
+  const AI = ai();
+  if (AI && AI.toast) AI.toast('Artikel fertig: ' + title);
+}
+
+let running = false;
+const gaveUp = new Set();   // wishes that failed in this visit: not tried again until diora is reopened
+
+// The open wishes (docs/transcription.md): reserve one, write it, save it. Nothing without a key.
+export async function pickUp({onState} = {}) {
+  if (running) return {busy: true};
+  const AI = ai();
+  if (!AI || !AI.available()) return {done: 0};
+  running = true;
+  const result = {done: 0, failed: 0, skipped: 0};
+  try {
+    if (!(await AI.hasKey())) return result;
+    let pending;
+    try {
+      const r = await getJson('/podcasts/transcription/pending/');
+      if (!r.ok || !Array.isArray(r.data)) return result;
+      pending = r.data;
+    } catch (e) { return result; }
+    for (const wish of pending) {
+      if (gaveUp.has(wish.episode_id)) { result.skipped++; continue; }
+      let claim;
+      try { claim = await claimWish(wish.episode_id); } catch (e) { return result; }
+      if (!claim.ok) { result.skipped++; continue; }   // 409: another device has it
+      try {
+        const saved = await write(wish.episode_id, {onState});
+        result.done++;
+        notifyDone(saved.title || wish.title, wish.episode_id);
+      } catch (e) {
+        result.failed++;
+        gaveUp.add(wish.episode_id);
+        try { await release(wish.episode_id); } catch (e2) { /* it runs out in 15 minutes */ }
+      }
+    }
+  } finally {
+    running = false;
+  }
+  return result;
+}
+
+// While diora is open and a job of its own runs, ask every five minutes (watch(true) when one starts,
+// watch(false) when none does).
+let timer = null;
+export function watch(on) {
+  if (on && !timer) timer = setInterval(() => { pickUp(); }, POLL_MS);
+  else if (!on && timer) { clearInterval(timer); timer = null; }
+}
+
+export const _reset = () => { running = false; gaveUp.clear(); };
+
+if (typeof window !== 'undefined' && typeof document !== 'undefined') {
+  window.DioraPodcastArticle = {write, pickUp, watch, notifyDone, claim: claimWish, release, deviceId, markdownToHtml, transcriptToText};
+  window.dispatchEvent(new Event('diora-podcast-article'));
+  // At the start of diora.
+  setTimeout(() => { pickUp(); }, 4000);
+}

static/js/sw.js

@@ -2,7 +2,7 @@
  * diora service worker — caches the app shell for offline use.
  */
 
-const CACHE = 'diora-v155';
+const CACHE = 'diora-v171';
 const SHELL = [
   '/static/css/app.css',
   '/static/js/app/01-core.js',
@@ -16,6 +16,7 @@ const SHELL = [
   '/static/js/app/29-m3u.js',
   '/static/js/app/30-contrast.js',
   '/static/js/app/32-crypto.js',
+  '/static/js/app/33-unlock.js',
   '/static/js/app/34-wallpaper.js',
   '/static/js/app/40-epub.js',
   '/static/js/app/44-books.js',
@@ -31,6 +32,8 @@ const SHELL = [
   '/static/js/app/58-pdf.js',
   '/static/js/app/60-tts.js',
   '/static/js/app/62-reader.js',
+  '/static/js/app/63-pages.js',
+  '/static/js/app/65-article.js',
   '/static/js/app/64-reader-settings.js',
   '/static/js/app/66-pdf-paginated.js',
   '/static/js/app/68-bookmarks.js',

static/podcasts/native-podcast.js

@@ -1,4 +1,4 @@
-// Copied from diora-android e287f48 2026-10-25: web-native/static/native-podcast.js
+// Copied from diora-android ba1e933 2026-11-09: web-native/static/native-podcast.js
 // by tools/sync_podcast_ui.py. Do not edit here: change it in diora-android
 // and copy again. The browser host is static/js/podcasts-web.js.
 
@@ -273,6 +273,47 @@
     t.nuiTimer = setTimeout(() => t.classList.remove('nui-show'), 2600);
   }
 
+  // ---- sharing an episode: its page from the feed (`link`), else the file / the video's watch URL ----
+
+  function shareUrl(e) {
+    const r = e.raw || {};
+    const d = e.id > 0 ? storedDetails(e.id) : null;
+    return r.link || r.episode__link || (d && d.link) || r.audio_url || r.episode__audio_url || (d && d.audio_url) || '';
+  }
+
+  function copyText(text) {
+    if (navigator.clipboard && navigator.clipboard.writeText) {
+      return navigator.clipboard.writeText(text).then(() => true, () => copyTextOld(text));
+    }
+    return Promise.resolve(copyTextOld(text));
+  }
+
+  function copyTextOld(text) {
+    const t = el('textarea');
+    t.value = text;
+    t.setAttribute('readonly', '');
+    t.style.cssText = 'position:fixed;top:0;left:0;opacity:0';
+    document.body.appendChild(t);
+    t.select();
+    let ok = false;
+    try { ok = document.execCommand('copy'); } catch (err) {}
+    t.remove();
+    return ok;
+  }
+
+  // Android: the system's share sheet. Elsewhere (browser, desktop app): title and link onto the clipboard.
+  async function shareEpisode(e) {
+    let url = shareUrl(e);
+    if (!url && e.id > 0 && navigator.onLine) {   // the phone's own list data may not know the link yet
+      const d = await loadDetails(e.id);
+      url = (d && (d.link || d.audio_url)) || '';
+    }
+    if (!url) { toast('Diese Folge hat keine Adresse zum Teilen'); return; }
+    const headline = e.feed ? `${e.title} – ${e.feed}` : e.title;
+    if (N && N.shareText) { N.shareText(headline, url); return; }
+    toast(await copyText(headline + '\n' + url) ?'Link kopiert' : 'Kopieren ging nicht');
+  }
+
   // GET from our server; a status other than 2xx throws (as a network error does).
   async function getJson(url) {
     const r = await fetch(url);
@@ -581,6 +622,7 @@
       {icon: '▶', label: playing ? 'Player öffnen' : 'Abspielen', action: () => play(e)},
       !playing && {icon: '☰', label: 'Folge ansehen', hint: 'Shownotes, Kapitel, alles zu dieser Folge', action: () => openEpisode(e, opts)},
       favoriteItem(e),
+      shareItem(e),
       !playing && {icon: '⤴', label: 'Als Nächstes', hint: 'direkt nach der laufenden Folge', action: () => moveTo(e, 'next')},
       {icon: '⇈', label: 'Ganz nach oben', action: () => moveTo(e, 'top')},
       {icon: '⇊', label: 'Ganz nach unten', action: () => moveTo(e, 'bottom')},
@@ -595,6 +637,10 @@
     ]);
   }
 
+  function shareItem(e) {
+    return {icon: '⇪', label: 'Teilen', hint: N && N.shareText ? null : 'Link in die Zwischenablage', action: () => shareEpisode(e)};
+  }
+
   function episodeMenu(e, opts) {
     if (v2()) { listsMenu(e, opts); return; }
     const queued = inQueue(e.id);
@@ -602,6 +648,7 @@
       {icon: '▶', label: 'Abspielen', hint: queued ? null : 'kommt dafür in die Warteschlange', action: () => play(e)},
       {icon: '☰', label: 'Folge ansehen', hint: 'Shownotes, Kapitel, alles zu dieser Folge', action: () => openEpisode(e, opts)},
       favoriteItem(e),
+      shareItem(e),
       !queued && {icon: '⤴', label: 'Als Nächstes', action: () => enqueue(e, 'after_current')},
       !queued && {icon: '⇈', label: 'An den Anfang der Warteschlange', action: () => enqueue(e, 'front')},
       !queued && {icon: '⇊', label: 'Ans Ende der Warteschlange', action: () => enqueue(e, 'back')},
@@ -625,6 +672,7 @@
       {icon: '▶', label: 'Abspielen', hint: queued ? null : 'kommt an den Anfang der Wiedergabeliste', action: () => play(e)},
       {icon: '☰', label: 'Folge ansehen', hint: 'Shownotes, Kapitel, alles zu dieser Folge', action: () => openEpisode(e, opts)},
       favoriteItem(e),
+      shareItem(e),
       N.podcastLaterAdd && {icon: kept ? '✕' : '＋', svg: kept ? 'listx' : null, label: kept ? 'Aus der Merkliste nehmen' : 'Auf die Merkliste',
        hint: kept ? null : 'für später; wird geladen', action: () => toggleLater(e)},
       !queued && {icon: '⤴', label: 'Als Nächstes hören', hint: 'in die Wiedergabeliste nach der laufenden Folge', action: () => enqueue(e, 'after_current')},
@@ -691,6 +739,7 @@
         ...dlSettingsItems(),
       ]),
       backgroundSyncItem(),
+      {icon: '✎', label: 'Transkription', hint: 'Whisper-Instanz für Transkripte und Artikel, Modell, Sprache', action: transcriptionSettings},
       {icon: '⇆', label: 'Wischgesten', hint: 'was Wischen nach links und rechts in jeder Liste tut', action: swipeSettings},
       startPageItem(),
       group('⇅', 'Abos umziehen', 'OPML exportieren und importieren', () => [
@@ -2833,6 +2882,15 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
     return true;
   };
 
+  // A transcript/article notice was tapped (TranscriptNotices.kt): the episode by id.
+  window.__dioraOpenEpisodeById = async function (id) {
+    const d = await loadDetails(id);
+    if (!d) return false;
+    return window.__dioraPodcastOpenEpisode(Object.assign({}, d, {
+      feed__id: d.feed && d.feed.id, feed__title: d.feed && d.feed.title, feed__artwork_url: d.feed && d.feed.artwork_url,
+    }));
+  };
+
   window.__dioraPodcastOpenFeed = function (feedId) {
     const f = feeds().find(x => x.id === Number(feedId));
     if (!f) return false;
@@ -4328,6 +4386,8 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
       box.append(el('div', 'nui-pod-meta', 'Shownotes brauchen beim ersten Mal eine Verbindung.'));
       return;
     }
+    const tr = trOf(id);
+    trFill(tr, d, !!fresh);
     if (live && sheet) {
       const fav = sheet.querySelector('.nui-pod-fav');
       if (fav) { fav.textContent = d.favorite ? '★' : '☆'; fav.classList.toggle('on', !!d.favorite); }
@@ -4338,10 +4398,23 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
     if (chapBox) chapterBox(chapBox, chapters, id);
     const tabs = el('div', 'nui-pod-subtabs');
     const body = el('div', 'nui-pod-notesbody');
+    // Jump to a second of the episode (taps on the article's "▶ ab 12:34" links).
+    const seek = s => {
+      if (live) N.podcastSeek(s * 1000);
+      else if (jump) jump(s);
+      else toast('Zum Anhören die Folge öffnen');
+    };
+    let current = '';
+    let sig = '';
+    const sigOf = () => tabList().map(x => x[0]).join() + '|' +
+      (current === 'transcript' ? (tr.has ? 'has' : trSig(id)) : current === 'article' ? trSig(id) + ':' + tr.article : '');
     const show = which => {
+      current = which;
       tabs.querySelectorAll('button').forEach(b => b.classList.toggle('active', b.dataset.which === which));
       body.textContent = '';
-      if (which === 'chapters') {
+      if (which === 'article') {
+        articlePanel(body, id, seek);
+      } else if (which === 'chapters') {
         const ol = el('ol', 'nui-pod-chapters');
         chapters.forEach((c, i) => {
           const li = el('li', 'nui-pod-chapter');
@@ -4372,6 +4445,7 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
         }
         body.append(notes);
       }
+      sig = sigOf();
     };
     const tab = (which, label) => {
       const b = el('button', 'nui-pod-subtab', label);
@@ -4380,11 +4454,33 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
       b.addEventListener('click', () => show(which));
       tabs.append(b);
     };
-    if (chapters.length && !chapBox) tab('chapters', `Kapitel (${chapters.length})`);
+    // The Artikel tab comes first (when there is one); Transkript for every episode with sound
-    tab('notes', 'Shownotes');
+    // (not YouTube: the server never fetches its sound), with a button to make one when it is missing.
-    if (d.has_transcript) tab('transcript', 'Transkript');
+    const canTranscribe = !!d.audio_url && !isYouTube(d.audio_url);
+    function tabList() {
+      const list = [];
+      if (tr.article) list.push(['article', 'Artikel']);
+      if (chapters.length && !chapBox) list.push(['chapters', `Kapitel (${chapters.length})`]);
+      list.push(['notes', 'Shownotes']);
+      if (tr.has || canTranscribe) list.push(['transcript', 'Transkript']);
+      return list;
+    }
+    const draw = want => {
+      const list = tabList();
+      tabs.textContent = '';
+      list.forEach(([w, l]) => tab(w, l));
+      show(list.some(x => x[0] === want) ? want : list[0][0]);
+    };
+    // Called when the state of the episode's transcript / article changes.
+    box.trUpdate = select => {
+      if (select) draw(select);
+      else if (sigOf() !== sig) draw(current);
+    };
+    trBoxes.add(box);
     box.append(tabs, body);
-    show(chapters.length && !chapBox ? 'chapters' : 'notes');
+    draw('');
+    if (epScreen && epScreen.e.id === id) episodeScreenParts();
+    trSchedule();
   }
 
   // The chapters in the full player: which one plays, previous / next (like AntennaPod's
@@ -4455,6 +4551,8 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
   let transcriptTouched = 0;   // the user scrolled the transcript: don't pull it along for a while
 
   async function showTranscript(body, id, live) {
+    // None yet: the button to make one (or its progress).
+    if (!trOf(id).has) { transcriptPanel(body, id); return; }
     body.append(el('div', 'nui-pod-meta', 'Transkript wird geladen …'));
     let t = transcripts.get(id);
     if (!t) {
@@ -4468,6 +4566,7 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
       }
     }
     body.textContent = '';
+    if (trOf(id).origin === 'whisper') body.append(el('div', 'nui-pod-meta nui-pod-tr-origin', 'automatisch transkribiert'));
     const list = el('div', 'nui-pod-transcript');
     for (const seg of t.segments || []) {
       const p = el('p', 'nui-pod-seg');
@@ -4495,6 +4594,493 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
     }
   }
 
+  // ---- smart transcription (diora-web docs/transcription.md) ------------------------------
+  // The server transcribes (a job per episode, polled every 5 s while it is on screen); the article
+  // is written on this device with the user's own key (DioraPodcastArticle). One state per episode
+  // feeds the Transkript tab, the Artikel tab and the "✦ Artikel" button.
+
+  const trStates = new Map();   // episode id -> {job, writing, has, origin, article}
+  const trBoxes = new Set();    // the details boxes on screen (showDetails)
+  const trArticles = new Map(); // feed item id -> its item (the article's text)
+  let trTimer = null;
+
+  async function trFetch(method, url, body) {
+    const headers = {'Content-Type': 'application/json'};
+    if (typeof getCsrfToken === 'function') headers['X-CSRFToken'] = getCsrfToken();
+    const r = await fetch(url, {method, headers, cache: 'no-store', body: body === undefined ? undefined : JSON.stringify(body)});
+    let data = null;
+    try { data = await r.json(); } catch (e) {}
+    return {ok: r.ok, status: r.status, data};
+  }
+
+  function trFill(t, d, fresh) {
+    if (fresh || !t.has) t.has = !!d.has_transcript;
+    if (fresh || !t.origin) t.origin = d.transcript_origin || null;
+    if (fresh || t.article == null) t.article = d.article_item_id || null;
+    if ((fresh || !t.job) && 'transcription' in d) t.job = d.transcription || null;
+  }
+
+  function trOf(id) {
+    let t = trStates.get(id);
+    if (!t) {
+      t = {job: null, writing: null, has: false, origin: null, article: null, polling: false};
+      trStates.set(id, t);
+      const d = storedDetails(id);
+      if (d) trFill(t, d, true);
+    }
+    return t;
+  }
+
+  const trActive = j => !!j && (j.state === 'queued' || j.state === 'downloading' || j.state === 'transcribing');
+
+  // What is going on with the episode right now (a transcription on the server, or the article
+  // being written here), as the line the buttons show; null when nothing.
+  function trStatus(id) {
+    const t = trOf(id);
+    if (t.writing) return {text: t.writing.text, kind: 'writing'};
+    const j = t.job;
+    if (!trActive(j)) return null;
+    const text = j.state === 'queued' ? (j.error ? 'Wartet: ' + j.error : 'Wartet …')
+      : j.state === 'downloading' ? 'Lade Folge …'
+      : j.chunks_total ? `Transkribiere ${j.chunks_done || 0}/${j.chunks_total} …` : 'Transkribiere …';
+    return {text, kind: 'job'};
+  }
+
+  // Changes only when what a view shows changes.
+  function trSig(id) {
+    const t = trOf(id);
+    const st = trStatus(id);
+    return (st ? st.text : '') + '|' + (t.job && t.job.state === 'error' ? t.job.error || 'x' : '');
+  }
+
+  const trVisible = id => (epScreen && epScreen.e.id === id && !!notesSheet && notesSheet.isConnected) ||
+    [...trBoxes].some(b => b.isConnected && b.dataset.episode === String(id));
+
+  function trChanged(id, select) {
+    for (const b of [...trBoxes]) {
+      if (!b.isConnected) { trBoxes.delete(b); continue; }
+      if (b.dataset.episode === String(id) && b.trUpdate) b.trUpdate(select);
+    }
+    if (epScreen && epScreen.e.id === id) episodeScreenParts();
+    trSchedule();
+  }
+
+  // Poll the jobs on screen every five seconds; nothing runs when none is shown.
+  function trSchedule() {
+    const ids = [...trStates].filter(([id, t]) => trActive(t.job) && trVisible(id)).map(([id]) => id);
+    if (ids.length && !trTimer) {
+      trTimer = setInterval(trTick, 5000);
+      trTick();
+    } else if (!ids.length && trTimer) {
+      clearInterval(trTimer);
+      trTimer = null;
+    }
+  }
+
+  async function trTick() {
+    if (document.hidden) return;
+    for (const [id, t] of [...trStates]) {
+      if (!trActive(t.job) || !trVisible(id) || t.polling) continue;
+      t.polling = true;
+      try {
+        const r = await trFetch('GET', `/podcasts/episodes/${id}/transcribe/`);
+        if (r.ok && r.data) await trApply(id, r.data);
+      } catch (e) {} finally { t.polling = false; }
+    }
+    trSchedule();
+  }
+
+  async function trApply(id, job) {
+    const t = trOf(id);
+    if (!job || job.state === 'none') { t.job = null; trChanged(id); return; }
+    t.job = job;
+    if (job.state !== 'done') { trChanged(id); return; }
+    const want = !!job.want_article;
+    t.job = null;
+    t.has = true;
+    transcripts.delete(id);
+    await trFresh(id);
+    trChanged(id);
+    // diora is open: the article is written right away (when this device has a key; else it waits).
+    if (want && !t.article && !t.writing) writeArticle(id, {claim: true});
+  }
+
+  async function trFresh(id) {
+    const d = await loadDetails(id);
+    if (d) trFill(trOf(id), d, true);
+    return d;
+  }
+
+  async function startTranscription(id, article) {
+    const t = trOf(id);
+    if (!navigator.onLine) { toast('Das braucht eine Verbindung'); return; }
+    let r;
+    try { r = await trFetch('POST', `/podcasts/episodes/${id}/transcribe/`, {article: !!article}); } catch (e) { r = {ok: false, status: 0}; }
+    if (!r.ok) {
+      const err = r.data && r.data.error;
+      const msg = err === 'youtube' ? 'YouTube-Folgen lassen sich nicht transkribieren.'
+        : err === 'too large' ? 'Die Folge ist zu groß zum Transkribieren (über 800 MB).'
+        : r.status ? `Das Transkribieren ließ sich nicht starten (${r.status}).` : 'Keine Verbindung.';
+      t.job = {state: 'error', error: msg, want_article: !!article};
+      trChanged(id);
+      return;
+    }
+    const job = r.data || {};
+    if (job.state === 'done') {
+      t.job = null;
+      t.has = true;
+      await trFresh(id);
+      trChanged(id);
+      if (article) writeArticle(id);
+      return;
+    }
+    t.job = Object.assign({state: 'queued', want_article: !!article}, job);
+    if (article && window.DioraPodcastArticle) DioraPodcastArticle.watch(true);
+    trChanged(id);
+  }
+
+  async function cancelTranscription(id) {
+    const t = trOf(id);
+    try { await trFetch('DELETE', `/podcasts/episodes/${id}/transcribe/`); } catch (e) {}
+    t.job = null;
+    trChanged(id);
+  }
+
+  // The key first: without one the settings open (write() would wait for ever behind the sheet).
+  async function trKeyReady() {
+    const AI = window.DioraAI;
+    if (!AI || !AI.available()) { toast('Die KI-Funktionen gibt es hier nicht'); return false; }
+    let has = false;
+    try { has = await AI.hasKey(); } catch (e) {}
+    if (!has) {
+      toast('Für den Artikel braucht diora deinen KI-Schlüssel');
+      AI.settings();
+    }
+    return has;
+  }
+
+  async function articleAction(id) {
+    const st = trStatus(id);
+    if (st) { toast(st.text); return; }
+    if (!(await trKeyReady())) return;
+    if (trOf(id).has) writeArticle(id);
+    else startTranscription(id, true);
+  }
+
+  // Write the article here and save it. `claim`: a wish of the server (reserve it first, so that the
+  // phone and the desktop don't both write it) and no questions when there is no key.
+  async function writeArticle(id, opts) {
+    const claim = !!(opts && opts.claim);
+    const t = trOf(id);
+    if (t.writing) return;
+    const AI = window.DioraAI;
+    const PA = window.DioraPodcastArticle;
+    if (!AI || !AI.available() || !PA) { if (!claim) toast('Artikel schreiben geht hier nicht'); return; }
+    if (claim) {
+      let has = false;
+      try { has = await AI.hasKey(); } catch (e) {}
+      if (!has) return;
+    }
+    const ctrl = new AbortController();
+    t.writing = {text: 'Lade Transkript …', ctrl};
+    trChanged(id);
+    let claimed = false;
+    try {
+      if (claim) {
+        const c = await PA.claim(id);
+        if (!c.ok) { t.writing = null; trChanged(id); return; }   // another device has it
+        claimed = true;
+      }
+      let last = 0;
+      const saved = await PA.write(id, {
+        signal: ctrl.signal,
+        onState: s => {
+          if (!t.writing) return;
+          t.writing.text = s.text;
+          const now = Date.now();
+          if (s.phase !== 'writing' || now - last > 1000) { last = now; trChanged(id); }
+        },
+      });
+      if (t.article) trArticles.delete(t.article);
+      t.article = saved.item_id;
+      t.writing = null;
+      await trFresh(id);
+      if (claim && PA.notifyDone) PA.notifyDone(saved.title, id); else toast('Artikel fertig');
+      trChanged(id, 'article');
+    } catch (e) {
+      t.writing = null;
+      if (claimed) { try { await PA.release(id); } catch (e2) {} }
+      trChanged(id);
+      if (ctrl.signal.aborted) toast('Abgebrochen');
+      else if (AI.showError) AI.showError('Artikel schreiben', e);
+      else toast((e && e.message) || 'Der Artikel ließ sich nicht schreiben');
+    }
+  }
+
+  async function removeArticle(id) {
+    const t = trOf(id);
+    const ok = typeof customConfirm === 'function'
+      ? await customConfirm('Den Artikel zu dieser Folge entfernen? Er verschwindet auch aus „Podcast-Artikel“ in den Blogs.', {danger: true}) : true;
+    if (!ok) return;
+    let r;
+    try { r = await trFetch('DELETE', `/podcasts/episodes/${id}/article/`); } catch (e) { r = {ok: false}; }
+    if (!r.ok) { toast('Entfernen ging nicht'); return; }
+    if (t.article) trArticles.delete(t.article);
+    t.article = null;
+    await trFresh(id);
+    trChanged(id);
+    toast('Artikel entfernt');
+  }
+
+  // The Transkript tab of an episode without a transcript: start, progress, error.
+  function transcriptPanel(body, id) {
+    const t = trOf(id);
+    body.textContent = '';
+    const st = trStatus(id);
+    const wrap = el('div', 'nui-pod-tr');
+    if (st) {
+      wrap.append(el('div', 'nui-pod-tr-status', st.text));
+      if (st.kind === 'job') {
+        wrap.append(el('div', 'nui-pod-meta', t.job.want_article
+          ? 'Das dauert je nach Länge Minuten bis Stunden. Danach schreibt diora den Artikel, wenn diora offen ist.'
+          : 'Das dauert je nach Länge Minuten bis Stunden. Du bekommst eine Benachrichtigung, wenn es fertig ist.'));
+        wrap.append(trButton('Abbrechen', () => cancelTranscription(id)));
+      } else {
+        wrap.append(trButton('Abbrechen', () => { if (t.writing) t.writing.ctrl.abort(); }));
+      }
+    } else if (t.job && t.job.state === 'error') {
+      wrap.append(el('div', 'nui-pod-tr-error', t.job.error || 'Das Transkribieren ist fehlgeschlagen.'));
+      wrap.append(trButton('Erneut versuchen', () => startTranscription(id, !!t.job.want_article), true));
+    } else {
+      wrap.append(el('div', 'nui-pod-meta', 'Diese Folge hat kein Transkript. Der Server kann es erstellen; das dauert je nach Länge ' +
+        'Minuten bis Stunden, am Ende kommt eine Benachrichtigung.'));
+      wrap.append(trButton('Transkript erstellen', () => startTranscription(id, false), true));
+    }
+    body.append(wrap);
+  }
+
+  function trButton(label, run, primary) {
+    const b = el('button', 'btn nui-pod-trbtn' + (primary ? ' btn-primary' : ''), label);
+    b.type = 'button';
+    b.addEventListener('click', run);
+    return b;
+  }
+
+  // The Artikel tab: the article from Blogs (a feed item), its three actions, taps on "▶ ab 12:34" seek.
+  async function articlePanel(body, id, seek) {
+    const t = trOf(id);
+    const itemId = t.article;
+    body.textContent = '';
+    const bar = el('div', 'nui-pod-trbar');
+    const st = trStatus(id);
+    if (st && st.kind === 'writing') {
+      bar.append(el('span', 'nui-pod-tr-status', st.text), trButton('Abbrechen', () => { if (t.writing) t.writing.ctrl.abort(); }));
+    } else {
+      if (window.__dioraFeedsOpenItem) {
+        bar.append(trButton('In Blogs öffnen', () => {
+          closeNotes();
+          closePlayer();
+          window.__dioraFeedsOpenItem(itemId);
+        }));
+      }
+      if (window.DioraAI && DioraAI.available() && !st) bar.append(trButton('Neu erstellen', () => rewriteArticle(id)));
+      bar.append(trButton('Entfernen', () => removeArticle(id)));
+    }
+    const text = el('div', 'nui-pod-notes nui-pod-article');
+    body.append(bar, text);
+    let item = trArticles.get(itemId);
+    if (!item) {
+      text.append(el('div', 'nui-pod-meta', 'Artikel wird geladen …'));
+      try {
+        item = (await getJson(`/feeds/items/${itemId}/`)).item;
+        if (item) trArticles.set(itemId, item);
+      } catch (e) {}
+      if (!text.isConnected || t.article !== itemId) return;
+      text.textContent = '';
+    }
+    if (!item) {
+      text.append(el('div', 'nui-pod-meta', navigator.onLine ? 'Der Artikel ließ sich nicht laden.' : 'Der Artikel braucht eine Verbindung.'));
+      return;
+    }
+    // The server cleaned the HTML (feeds/fetch.py clean_html), as for every feed item.
+    if (item.content) text.innerHTML = item.content;
+    else text.append(el('p', null, item.summary || ''));
+    text.querySelectorAll('a[href]:not([href^="#t="])').forEach(a => a.setAttribute('target', '_blank'));
+    text.addEventListener('click', ev => {
+      const a = ev.target.closest('a[href^="#t="]');
+      if (!a) return;
+      ev.preventDefault();
+      const s = parseInt(a.getAttribute('href').slice(3), 10);
+      if (s >= 0) seek(s);
+    });
+  }
+
+  // "Neu erstellen" replaces the article: ask first.
+  async function rewriteArticle(id) {
+    const ok = typeof customConfirm === 'function'
+      ? await customConfirm('Den Artikel neu schreiben? Der jetzige wird ersetzt.') : true;
+    if (ok) articleAction(id);
+  }
+
+  // The "✦ Artikel" button of the episode screen, or null where it isn't offered.
+  function trArticleButton(e) {
+    const id = e.id;
+    if (!(id > 0)) return null;
+    const t = trOf(id);
+    const d = storedDetails(id) || {};
+    const url = (e.raw && (e.raw.audio_url || e.raw.episode__audio_url)) || d.audio_url || '';
+    if (e.yt || !url || isYouTube(url)) return null;
+    const st = trStatus(id);
+    if (st) return {label: st.text, run: () => toast(st.text), cls: 'on'};
+    if (t.article) return {label: 'Artikel', run: () => { if (epScreen && epScreen.box && epScreen.box.trUpdate) epScreen.box.trUpdate('article'); }, cls: 'on'};
+    if (!(window.DioraAI && DioraAI.available())) return null;
+    return {label: t.has ? 'Artikel erstellen' : 'Transkribieren und Artikel erstellen', run: () => articleAction(id)};
+  }
+
+  // ---- settings of the transcription instance (GET/POST /podcasts/transcription/settings/) ----
+
+  const TR_LANGUAGES = [['', 'Automatisch'], ['de', 'Deutsch'], ['en', 'Englisch'], ['fr', 'Französisch'], ['es', 'Spanisch'],
+    ['it', 'Italienisch'], ['nl', 'Niederländisch'], ['pt', 'Portugiesisch'], ['pl', 'Polnisch'], ['ru', 'Russisch'],
+    ['tr', 'Türkisch'], ['uk', 'Ukrainisch'], ['ja', 'Japanisch'], ['zh', 'Chinesisch']];
+  let trSheet = null;
+
+  function closeTrSettings() {
+    if (!trSheet) return false;
+    trSheet.remove();
+    trSheet = null;
+    return true;
+  }
+
+  async function transcriptionSettings() {
+    closeTrSettings();
+    const {sheet: made, panel, top} = podSheet('nui-pod-notes-sheet nui-pod-trset', 'Schließen', closeTrSettings);
+    trSheet = made;
+    const body = el('div', 'nui-pod-trform');
+    panel.append(top, el('div', 'nui-pod-player-title', 'Transkription'), body);
+    document.body.append(made);
+    body.append(el('div', 'nui-pod-meta', 'Einstellungen werden geladen …'));
+    let s;
+    try {
+      const r = await trFetch('GET', '/podcasts/transcription/settings/');
+      if (!r.ok) throw new Error(r.status);
+      s = r.data;
+    } catch (e) {
+      body.textContent = '';
+      body.append(el('div', 'nui-pod-meta', navigator.onLine ? 'Die Einstellungen ließen sich nicht laden.' : 'Die Einstellungen brauchen eine Verbindung.'));
+      return;
+    }
+    if (trSheet !== made) return;
+    drawTranscriptionSettings(body, s, made);
+  }
+
+  function drawTranscriptionSettings(body, s, made) {
+    body.textContent = '';
+    const field = (label, input) => {
+      const l = el('label', 'nui-pod-trfield');
+      l.append(el('span', null, label), input);
+      return l;
+    };
+    const text = (type, value, placeholder) => {
+      const i = el('input', 'search-input');
+      i.type = type;
+      i.value = value == null ? '' : value;
+      if (placeholder) i.placeholder = placeholder;
+      i.setAttribute('autocomplete', 'off');
+      i.setAttribute('autocapitalize', 'off');
+      i.spellcheck = false;
+      return i;
+    };
+    body.append(el('div', 'nui-pod-meta', 'Der Server schneidet die Folge in Stücke und schickt sie an eine Whisper-Instanz ' +
+      '(OpenAI-kompatibel). Das passiert nur, wenn du es in einer Folge auslöst.'));
+    let engine = s.engine === 'builtin' && s.builtin_available ? 'builtin' : 'custom';
+    const engines = el('div', 'nui-pod-subtabs');
+    const options = [s.builtin_available && ['builtin', 'Eingebaute Instanz'], ['custom', 'Eigene Instanz']].filter(Boolean);
+    const url = text('url', s.url, 'https://whisper.example.org/v1');
+    const token = text('password', '', s.has_token ? 'gespeichert (leer lassen = behalten)' : 'optional');
+    token.setAttribute('autocomplete', 'new-password');
+    const urlField = field('Adresse', url);
+    const tokenField = field('Token', token);
+    const sync = () => {
+      engines.querySelectorAll('button').forEach(b => b.classList.toggle('active', b.dataset.engine === engine));
+      urlField.hidden = tokenField.hidden = engine !== 'custom';
+    };
+    for (const [value, label] of options) {
+      const b = el('button', 'nui-pod-subtab', label);
+      b.type = 'button';
+      b.dataset.engine = value;
+      b.addEventListener('click', () => { engine = value; sync(); });
+      engines.append(b);
+    }
+    body.append(engines, urlField, tokenField);
+    const model = text('text', s.model, 'z. B. large-v3-turbo');
+    const lang = el('select', 'search-input');
+    const langs = TR_LANGUAGES.some(l => l[0] === (s.language || '')) ? TR_LANGUAGES : TR_LANGUAGES.concat([[s.language, s.language]]);
+    for (const [v, l] of langs) {
+      const o = el('option', null, l);
+      o.value = v;
+      lang.append(o);
+    }
+    lang.value = s.language || '';
+    const par = text('number', s.parallel || 3);
+    par.min = '1';
+    par.max = '8';
+    par.inputMode = 'numeric';
+    body.append(field('Modell', model), field('Sprache', lang), field('Parallel (1 bis 8 Stücke gleichzeitig)', par));
+    const status = el('div', 'nui-pod-tr-status');
+    status.setAttribute('aria-live', 'polite');
+    const say = (msg, bad) => { status.textContent = msg; status.classList.toggle('nui-pod-tr-error', !!bad); };
+
+    // Save; the token is sent only when typed (empty keeps it) or removed on purpose.
+    const save = async removeToken => {
+      if (engine === 'custom' && !url.value.trim()) { say('Bitte eine Adresse eintragen.', true); return null; }
+      const post = {
+        engine, url: url.value.trim(), model: model.value.trim(), language: lang.value,
+        parallel: Math.max(1, Math.min(8, parseInt(par.value, 10) || 3)),
+      };
+      if (removeToken) post.token = '';
+      else if (token.value) post.token = token.value;
+      const r = await trFetch('POST', '/podcasts/transcription/settings/', post);
+      if (!r.ok) { say((r.data && r.data.error) || `Speichern ging nicht (${r.status}).`, true); return null; }
+      let next = r.data && 'engine' in r.data ? r.data : null;
+      if (!next) {
+        const g = await trFetch('GET', '/podcasts/transcription/settings/');
+        next = g.ok ? g.data : Object.assign({}, s, post, {has_token: !!(s.has_token && !removeToken) || !!post.token});
+      }
+      return next;
+    };
+    const run = fn => async () => {
+      try { await fn(); } catch (e) { say(navigator.onLine ? 'Das ging nicht.' : 'Keine Verbindung.', true); }
+    };
+    const buttons = el('div', 'nui-pod-trbar');
+    buttons.append(
+      trButton('Speichern', run(async () => {
+        const next = await save(false);
+        if (next && trSheet === made) { toast('Gespeichert'); drawTranscriptionSettings(body, next, made); }
+      }), true),
+      trButton('Verbindung testen', run(async () => {
+        say('Speichere und teste …');
+        const next = await save(false);
+        if (!next) return;
+        const r = await trFetch('POST', '/podcasts/transcription/test/', {});
+        if (trSheet !== made) return;
+        drawTranscriptionSettings(body, next, made);
+        const out = body.querySelector('.nui-pod-tr-status');
+        const ok = r.ok && r.data && r.data.ok;
+        out.textContent = ok ? `Verbindung steht${r.data.model ? ' (Modell ' + r.data.model + ')' : ''}.`
+          : 'Verbindung fehlgeschlagen: ' + ((r.data && r.data.error) || `Fehler ${r.status}`);
+        out.classList.toggle('nui-pod-tr-error', !ok);
+      })));
+    if (s.has_token) {
+      buttons.append(trButton('Token entfernen', run(async () => {
+        const next = await save(true);
+        if (next && trSheet === made) { toast('Token entfernt'); drawTranscriptionSettings(body, next, made); }
+      })));
+    }
+    body.append(buttons, status);
+    if (s.has_token) body.append(el('div', 'nui-pod-meta', 'Token: gespeichert'));
+    sync();
+  }
+
   // ---- the episode screen: a tap on a row opens it ----------------------------------------
   // Cover, title, podcast (→ its page), date, length, where it stands; one big button to play;
   // the actions with the swipes' names and icons; chapters (a tap plays from there) and notes.
@@ -4552,7 +5138,7 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
       }).catch(() => {});
     }
     document.body.append(notesSheet);
-    epScreen = {e, ctx, actions, playBtn, facts: info.querySelector('.nui-pod-epfacts')};
+    epScreen = {e, ctx, actions, playBtn, box, facts: info.querySelector('.nui-pod-epfacts')};
     episodeScreenParts();
     showDetails(box, e.id, false, seconds => playAt(e, seconds));
   }
@@ -4613,6 +5199,9 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
     if (N.podcastDiscard) {
       act('trash', '', 'Folge löschen', () => { closeNotes(); discardOne(e, ctx.list || '', ctx.filters || []); }, 'nui-pod-epact-danger');
     }
+    act(null, '⇪', 'Teilen', () => shareEpisode(e));
+    const art = trArticleButton(e);
+    if (art) act(null, '✦', art.label, art.run, art.cls);
     const fav = e.raw.favorite;
     act(null, fav ? '★' : '☆', 'Favorit', () => toggleFavorite(e, star => { e.raw.favorite = star; episodeScreenParts(); }),
       fav ? 'on' : '');
@@ -5828,6 +6417,7 @@ neuesten Folgen) und die vorhandenen Ordner. Schlage nur Änderungen vor, die wi
     window.__dioraBack = function () {
       if (closeStats()) return true;
       if (closeDownloads()) return true;
+      if (closeTrSettings()) return true;
       if (closeNotes()) return true;
       if (closePlayer()) return true;
       const onPodcasts = typeof _currentTopTab !== 'undefined' && _currentTopTab === 'podcasts';

static/podcasts/podcast-ui.css

@@ -1,4 +1,4 @@
-/* Copied from diora-android 7f37d4e 2026-10-25: the podcast parts of web-native/static/native.css
+/* Copied from diora-android 6a0a362 2026-11-09: the podcast parts of web-native/static/native.css
    (bottom sheet, podcasts) by tools/sync_podcast_ui.py. Do not edit here.
    The web's own additions are in podcasts-web.css. */
 
@@ -500,7 +500,9 @@ body.eink .nui-pod-chip.on strong,
 body.eink .nui-pod-sleep.on { color: #000 !important; }
 
 /* podcast chapters and shownotes */
-.nui-pod-notesbox { margin-top: 6px; border-top: 1px solid var(--border); padding-top: 6px; min-height: 40px; }
+/* flex: none – in the sheet's scrolling column the box would otherwise shrink to its 40 px and
+   the notes spill over what comes after (the similar episodes). */
+.nui-pod-notesbox { flex: none; margin-top: 6px; border-top: 1px solid var(--border); padding-top: 6px; min-height: 40px; }
 .nui-pod-subtabs { display: flex; gap: 4px; margin-bottom: 6px; }
 .nui-pod-subtab {
   min-height: 36px;
@@ -548,6 +550,21 @@ body.eink .nui-pod-sleep.on { color: #000 !important; }
 .nui-pod-epact-icon { font-size: 20px; line-height: 22px; }
 .nui-pod-epact.on .nui-pod-epact-icon { color: var(--accent); }
 .nui-pod-epact-danger { color: var(--red, #e5484d); }
+/* Transcript and article of an episode (the Transkript / Artikel tabs, the transcription settings). */
+.nui-pod-tr { display: flex; flex-direction: column; align-items: flex-start; gap: 10px; padding: 6px 2px; }
+.nui-pod-tr-status { font-size: 14.5px; font-weight: 600; }
+.nui-pod-tr-error { font-size: 14px; color: var(--red, #e5484d); }
+.nui-pod-tr-origin { margin-bottom: 6px; font-style: italic; }
+.nui-pod-trbar { display: flex; flex-wrap: wrap; align-items: center; gap: 8px; margin: 4px 0 10px; }
+.nui-pod-trbtn { min-height: 40px; }
+.nui-pod-article h2 { font-size: 17px; margin: 1.2em 0 0.2em; }
+.nui-pod-article p { margin: 0.5em 0; }
+.nui-pod-article blockquote { margin: 0.6em 0; padding-left: 12px; border-left: 3px solid var(--border); }
+.nui-pod-trform { display: flex; flex-direction: column; gap: 10px; }
+.nui-pod-trfield { display: flex; flex-direction: column; gap: 4px; font-size: 13px; color: var(--nui-muted); }
+.nui-pod-trfield input, .nui-pod-trfield select { width: 100%; box-sizing: border-box; font-size: 15px; color: inherit; }
+.nui-pod-trfield[hidden] { display: none; }
+body.eink .nui-pod-tr-error { color: #000; }
 body.eink .nui-pod-epact { border-color: #000; color: #000; }
 body.eink .nui-pod-epact-danger { color: #000; }
 body.eink .nui-pod-chapter.current,

templates/accounts/login.html

@@ -28,6 +28,12 @@
     {% endif %}
     <button type="submit" class="btn btn-primary btn-full">Anmelden</button>
   </form>
+  {% if oidc_enabled %}
+  <div class="auth-alt">
+    <div class="auth-or"><span>oder</span></div>
+    <a class="btn btn-full" href="{% url 'oidc_start' %}{% if request.GET.next %}?next={{ request.GET.next|urlencode }}{% endif %}">Mit {{ oidc_provider }} anmelden</a>
+  </div>
+  {% endif %}
   <p class="auth-switch">
     Noch kein Konto? <a href="{% url 'register' %}">Registrieren</a>
   </p>

templates/accounts/oidc_key.html

@@ -0,0 +1,120 @@
+<!DOCTYPE html>
+<html lang="de">
+<head>
+<meta charset="utf-8">
+<meta name="viewport" content="width=device-width, initial-scale=1">
+<title>Anmelden — diora</title>
+<style>
+  body { margin: 0; min-height: 100vh; display: flex; align-items: center; justify-content: center;
+         font: 15px/1.5 system-ui, sans-serif; background: #111; color: #ccc; }
+  @media (prefers-color-scheme: light) { body { background: #fafafa; color: #444; } }
+</style>
+</head>
+<body>
+<p id="status">Bücher werden entsperrt …</p>
+{{ config|json_script:"oidc-key-config" }}
+<script>
+// The book key with Authelia's help (accounts/oidc_views.py oidc_key, docs/accounts.md).
+// Every way out ends on `next`; whatever fails here, the app asks for the key itself
+// (static/js/app/33-unlock.js).
+(async function () {
+  const C = JSON.parse(document.getElementById('oidc-key-config').textContent);
+  const SS = 'diora_oidc_key_flow';
+  const KEY_NAME = 'diora_enc_key_' + C.user_id;
+  const status = document.getElementById('status');
+  const b64 = (buf) => btoa(String.fromCharCode(...new Uint8Array(buf)));
+  const unb64 = (s) => Uint8Array.from(atob(s), c => c.charCodeAt(0));
+  const b64url = (buf) => b64(buf).replace(/\+/g, '-').replace(/\//g, '_').replace(/=+$/, '');
+  const hexToBytes = (h) => Uint8Array.from(h.match(/../g) || [], x => parseInt(x, 16));
+  const csrf = () => (document.cookie.split('; ').find(r => r.startsWith('csrftoken=')) || '').split('=')[1] || '';
+  const done = (to) => location.replace(to || C.next || '/');
+
+  async function wrapKey(secret, sub) {
+    const hk = await crypto.subtle.importKey('raw', unb64(secret), 'HKDF', false, ['deriveKey']);
+    return crypto.subtle.deriveKey(
+      {name: 'HKDF', hash: 'SHA-256', salt: new TextEncoder().encode(sub), info: new TextEncoder().encode('diora authelia key v1')},
+      hk, {name: 'AES-GCM', length: 256}, false, ['encrypt', 'decrypt']);
+  }
+
+  // Is this the key of the books? Tried on one book's metadata (none: nothing to check).
+  async function fits(raw) {
+    try {
+      const books = await (await fetch('/books/', {credentials: 'same-origin'})).json();
+      const b = Array.isArray(books) ? books.find(x => x && x.meta_ct && x.meta_iv) : null;
+      if (!b) return true;
+      const k = await crypto.subtle.importKey('raw', raw, {name: 'AES-GCM'}, false, ['decrypt']);
+      await crypto.subtle.decrypt({name: 'AES-GCM', iv: hexToBytes(b.meta_iv)}, k, unb64(b.meta_ct));
+      return true;
+    } catch (e) { return false; }
+  }
+
+  const params = new URLSearchParams(location.search);
+  const isCallback = location.pathname.endsWith('/callback/');
+
+  if (!isCallback) {
+    // Also a device that has its key goes once round: it leaves the key (again) wrapped with
+    // the current claim, so a changed claim at Authelia heals with the next login.
+    const flow = {
+      state: b64url(crypto.getRandomValues(new Uint8Array(24))),
+      verifier: b64url(crypto.getRandomValues(new Uint8Array(48))),
+      next: C.next,
+    };
+    sessionStorage.setItem(SS, JSON.stringify(flow));
+    const challenge = b64url(await crypto.subtle.digest('SHA-256', new TextEncoder().encode(flow.verifier)));
+    const q = new URLSearchParams({
+      response_type: 'code', client_id: C.client_id, redirect_uri: C.redirect_uri,
+      scope: 'openid profile diora_key', state: flow.state, prompt: 'none',
+      code_challenge: challenge, code_challenge_method: 'S256',
+    });
+    location.replace(C.authorize + '?' + q);
+    return;
+  }
+
+  let flow = null;
+  try { flow = JSON.parse(sessionStorage.getItem(SS) || 'null'); } catch (e) {}
+  sessionStorage.removeItem(SS);
+  const next = flow && flow.next;
+  try {
+    if (!flow || params.get('state') !== flow.state || !params.get('code')) return done(next);
+    const tok = await (await fetch(C.token, {
+      method: 'POST',
+      headers: {'Content-Type': 'application/x-www-form-urlencoded'},
+      body: new URLSearchParams({
+        grant_type: 'authorization_code', code: params.get('code'), redirect_uri: C.redirect_uri,
+        client_id: C.client_id, code_verifier: flow.verifier,
+      }),
+    })).json();
+    if (!tok.access_token) return done(next);
+    const info = await (await fetch(C.userinfo, {headers: {Authorization: 'Bearer ' + tok.access_token}})).json();
+    // Only the Authelia account linked to this diora account counts.
+    if (!info.sub || !C.subs.includes(info.sub) || !info.diora_key) return done(next);
+    const wrap = await wrapKey(info.diora_key, info.sub);
+
+    let have = null;
+    try { have = localStorage.getItem(KEY_NAME); } catch (e) {}
+    if (have) {
+      // This device has the key: leave it with the server, wrapped — only the right one (an
+      // old browser may hold a stray key of its own).
+      if (!(await fits(unb64(have)))) return done(next);
+      const iv = crypto.getRandomValues(new Uint8Array(12));
+      const ct = await crypto.subtle.encrypt({name: 'AES-GCM', iv}, wrap, unb64(have));
+      await fetch('/accounts/vault/authelia-key/', {
+        method: 'POST', credentials: 'same-origin',
+        headers: {'Content-Type': 'application/json', 'X-CSRFToken': csrf()},
+        body: JSON.stringify({ct: b64(ct), iv: b64(iv)}),
+      });
+      return done(next);
+    }
+    const v = await (await fetch('/accounts/vault/authelia-key/', {credentials: 'same-origin'})).json();
+    if (!v.ct) return done(next);
+    const raw = new Uint8Array(await crypto.subtle.decrypt({name: 'AES-GCM', iv: unb64(v.iv)}, wrap, unb64(v.ct)));
+    if (raw.length === 32 && await fits(raw)) localStorage.setItem(KEY_NAME, b64(raw));
+  } catch (e) {
+    status.textContent = 'Entsperren über Authelia ging nicht – du kannst es gleich mit Code oder Passwort tun.';
+    await new Promise(r => setTimeout(r, 1500));
+  }
+  done(next);
+})();
+</script>
+</body>
+</html>

templates/accounts/settings.html

@@ -94,6 +94,25 @@
       </form>
     </details>
 
+    {% if oidc_enabled or oidc_identities %}
+    <div class="oidc-links" style="margin-top:1rem;">
+      <h3 style="font-size:1rem; margin:0 0 0.4rem;">Anmelden über {{ oidc_provider }}</h3>
+      {% for ident in oidc_identities %}
+        <form method="post" action="{% url 'oidc_unlink' ident.pk %}" class="inline-form" style="display:flex; gap:0.6rem; align-items:center; margin-bottom:0.4rem;">
+          {% csrf_token %}
+          <span>Verknüpft mit <strong>{{ ident.username|default:ident.sub }}</strong>{% if ident.email %} ({{ ident.email }}){% endif %}</span>
+          <button type="submit" class="btn">Lösen</button>
+        </form>
+      {% empty %}
+        <p class="muted" style="margin:0 0 0.4rem;">Verknüpfe dein Konto, dann kannst du dich auch über {{ oidc_provider }} anmelden. Die Anmeldung mit Passwort bleibt.</p>
+      {% endfor %}
+      {% if oidc_enabled and not oidc_identities %}
+        <a class="btn" href="{% url 'oidc_start' %}?link=1">Mit {{ oidc_provider }} verknüpfen</a>
+      {% endif %}
+      <p class="muted" style="font-size:0.85rem; margin-top:0.4rem;">Deine Bücher bleiben mit deinem diora-Passwort verschlüsselt. Ein Gerät, das du über {{ oidc_provider }} anmeldest, bekommt den Schlüssel von einem schon entsperrten Gerät, oder du gibst dort einmal dein Passwort ein.</p>
+    </div>
+    {% endif %}
+
     <form method="post" action="{% url 'logout' %}" class="inline-form" style="margin-top:1rem;">
       {% csrf_token %}
       <button type="submit" class="btn">Abmelden</button>
@@ -268,10 +287,8 @@ async function _getOrCreateEncKey() {
       return crypto.subtle.importKey('raw', raw, {name: 'AES-GCM'}, false, ['encrypt', 'decrypt']);
     } catch (e) {}
   }
-  const key = await crypto.subtle.generateKey({name: 'AES-GCM', length: 256}, true, ['encrypt', 'decrypt']);
+  // Never a new key here: it would lock the existing books away (static/js/app/33-unlock.js).
-  const raw = await crypto.subtle.exportKey('raw', key);
+  throw new Error('Bücher sind auf diesem Gerät noch nicht entsperrt – öffne zuerst die Bibliothek');
-  localStorage.setItem(storageKey, _bytesToBase64(raw));
-  return key;
 }
 
 document.getElementById('pw-change-form')?.addEventListener('submit', function(e) {

templates/radio/player.html

@@ -371,6 +371,7 @@
 <script src="/static/js/app/29-m3u.js"></script>
 <script src="/static/js/app/30-contrast.js"></script>
 <script src="/static/js/app/32-crypto.js"></script>
+<script src="/static/js/app/33-unlock.js"></script>
 <script src="/static/js/app/34-wallpaper.js"></script>
 <script src="/static/js/app/40-epub.js"></script>
 <script src="/static/js/app/44-books.js"></script>
@@ -386,7 +387,9 @@
 <script src="/static/js/app/58-pdf.js"></script>
 <script src="/static/js/app/60-tts.js"></script>
 <script src="/static/js/app/62-reader.js"></script>
+<script src="/static/js/app/63-pages.js"></script>
 <script src="/static/js/app/64-reader-settings.js"></script>
+<script src="/static/js/app/65-article.js"></script>
 <script src="/static/js/app/66-pdf-paginated.js"></script>
 <script src="/static/js/app/68-bookmarks.js"></script>
 <script src="/static/js/app/70-search.js"></script>
@@ -397,7 +400,8 @@
 <script src="/static/js/app/80-radio-sidebar.js"></script>
 <script src="/static/js/app/99-init.js"></script>
 {% if user.is_authenticated %}{# The KI features (docs/ai.md): the user's own key, calls from the device. #}
-<script type="module" src="/static/js/ai.js"></script>{% endif %}
+<script type="module" src="/static/js/ai.js"></script>
+<script type="module" src="/static/js/podcast-article.js"></script>{% endif %}
 <script src="/static/js/podcasts-web.js"></script>
 <script src="/static/js/home.js"></script>
 <script src="/static/js/feeds.js"></script>

tools/tests/podcast-article.test.mjs

@@ -0,0 +1,172 @@
+// Tests of static/js/podcast-article.js under Node: node --test tools/tests/
+// No network, no LLM: fetch and DioraAI are stubbed.
+import test from 'node:test';
+import assert from 'node:assert/strict';
+import * as pa from '../../static/js/podcast-article.js';
+
+test('transcriptToText groups 30 seconds and splits on speaker change', () => {
+  const segs = [
+    {start: 0, end: 5, speaker: '', text: 'Hallo  zusammen.'},
+    {start: 20, end: 29, speaker: '', text: 'Heute geht es um X.'},
+    {start: 31, end: 40, speaker: '', text: 'Zweite Gruppe.'},
+    {start: 3725, end: 3730, speaker: '', text: 'Spät.'},
+    {start: 3726, end: 3730, speaker: '', text: '   '},
+  ];
+  assert.equal(pa.transcriptToText(segs),
+    '[00:00] Hallo zusammen. Heute geht es um X.\n[00:31] Zweite Gruppe.\n[62:05] Spät.');
+  const sp = [
+    {start: 0, speaker: 'Anna', text: 'Eins.'},
+    {start: 4, speaker: 'Anna', text: 'Zwei.'},
+    {start: 8, speaker: 'Bernd', text: 'Drei.'},
+  ];
+  assert.equal(pa.transcriptToText(sp), '[00:00] Anna: Eins. Zwei.\n[00:08] Bernd: Drei.');
+  assert.equal(pa.transcriptToText([]), '');
+});
+
+test('clock and mmss', () => {
+  assert.equal(pa.mmss(754), '12:34');
+  assert.equal(pa.clock(754), '12:34');
+  assert.equal(pa.clock(3725), '1:02:05');
+});
+
+test('token estimate and the limit message', () => {
+  assert.equal(pa.estimateTokens('a'.repeat(400)), 100);
+  const big = {segments: Array.from({length: 20000}, (_, i) => ({start: i * 31, speaker: '', text: 'x'.repeat(30)}))};
+  assert.throws(() => pa.buildPrompt({title: 'T'}, big), e => e.diora && /zu lang/.test(e.message));
+  const ok = pa.buildPrompt({title: 'T', feed: {title: 'P'}, description: '<p>Notes &amp; mehr</p>', duration_seconds: 3725,
+    chapters: [{start: 60, title: 'Kap 1'}]}, {segments: [{start: 0, text: 'Hallo'}]});
+  assert.match(ok, /Podcast: P\nFolge: T\nDauer: 1:02:05\nShownotes:\nNotes & mehr\nKapitel:\n\[01:00\] Kap 1\n\nTranskript:\n\[00:00\] Hallo$/);
+  assert.throws(() => pa.buildPrompt({title: 'T'}, {segments: []}), /leer/);
+});
+
+test('system prompt is the one of the contract', async () => {
+  const fs = await import('node:fs');
+  const doc = fs.readFileSync(new URL('../../docs/transcription.md', import.meta.url), 'utf8').split('\r\n').join('\n');
+  assert.ok(doc.includes(pa.SYSTEM));
+  assert.deepEqual(pa.SCHEMA.required, ['title', 'summary', 'kind', 'speakers', 'sections']);
+});
+
+test('markdownToHtml: paragraphs, emphasis, lists, quotes, escaping', () => {
+  assert.equal(pa.markdownToHtml('Eins *kursiv* und **fett**.\nWeiter.\n\nZwei.'),
+    '<p>Eins <em>kursiv</em> und <strong>fett</strong>. Weiter.</p>\n<p>Zwei.</p>');
+  assert.equal(pa.markdownToHtml('- a\n- b'), '<ul><li>a</li><li>b</li></ul>');
+  assert.equal(pa.markdownToHtml('1. a\n2. b'), '<ol><li>a</li><li>b</li></ol>');
+  assert.equal(pa.markdownToHtml('> „Zitat“\n> mehr'), '<blockquote><p>„Zitat“ mehr</p></blockquote>');
+  const evil = pa.markdownToHtml('<script>alert(1)</script> **<img src=x onerror=y>** [l](javascript:x) "q"');
+  assert.ok(!/<script|<img/.test(evil));
+  assert.ok(evil.includes('&lt;script&gt;') && evil.includes('<strong>&lt;img'));
+  assert.equal(pa.markdownToHtml('2 * 3 * 4'), '<p>2 * 3 * 4</p>');
+});
+
+test('articleHtml: summary, heading, time link, body', () => {
+  const h = pa.articleHtml({summary: 'Kurz & gut.', sections: [
+    {heading: 'Erste <b>Idee</b>', start: 754.9, body_markdown: 'Text.'}]});
+  assert.equal(h, '<p>Kurz &amp; gut.</p>\n<h2>Erste &lt;b&gt;Idee&lt;/b&gt;</h2>\n<p><a href="#t=754">▶ ab 12:34</a></p>\n<p>Text.</p>');
+});
+
+// ---- stubs ---------------------------------------------------------------------------------
+
+function stubWorld({pending, claims = {}, hasKey = true, askImpl, saveStatus = 200}) {
+  const calls = [];
+  const notices = [];
+  globalThis.DioraNative = undefined;
+  globalThis.DioraAI = {
+    available: () => true,
+    hasKey: async () => hasKey,
+    withKey: fn => fn(),
+    modelName: () => 'test-model',
+    toast: t => notices.push(t),
+    ask: askImpl || (async opts => {
+      opts.onProgress && opts.onProgress(1200);
+      return {title: 'Titel', summary: 'Sum.', kind: 'Gespräch', speakers: [], sections: [{heading: 'H', start: 61, body_markdown: 'B'}]};
+    }),
+  };
+  globalThis.fetch = async (url, init = {}) => {
+    const method = init.method || 'GET';
+    calls.push(method + ' ' + url + (init.body ? ' ' + init.body : ''));
+    const json = (status, data) => ({status, ok: status < 400, json: async () => data});
+    if (url === '/podcasts/transcription/pending/') return json(200, pending);
+    let m = /^\/podcasts\/episodes\/(\d+)\/article\/claim\/$/.exec(url);
+    if (m) return method === 'POST' ? json(claims[m[1]] || 200, {}) : json(200, {});
+    m = /^\/podcasts\/episodes\/(\d+)\/transcript\/$/.exec(url);
+    if (m) return m[1] === '3' ? json(404, {error: 'no transcript'}) : json(200, {segments: [{start: 0, speaker: '', text: 'Hallo'}]});
+    m = /^\/podcasts\/episodes\/(\d+)\/article\/$/.exec(url);
+    if (m) return json(saveStatus, {item_id: 900 + Number(m[1])});
+    m = /^\/podcasts\/episodes\/(\d+)\/$/.exec(url);
+    if (m) return json(200, {id: Number(m[1]), title: 'Folge ' + m[1], description: '', chapters: [], feed: {title: 'P'}});
+    return json(404, {});
+  };
+  return {calls, notices};
+}
+
+test('write: loads, asks with the contract options, saves the HTML', async () => {
+  const w = stubWorld({pending: []});
+  let opts;
+  globalThis.DioraAI.ask = async o => { opts = o; o.onProgress(5); return {title: 'T', summary: 'S', kind: 'Monolog', speakers: [], sections: [{heading: 'H', start: 61, body_markdown: 'B'}]}; };
+  const states = [];
+  const r = await pa.write(7, {onState: s => states.push(s.phase)});
+  assert.deepEqual(r, {item_id: 907, title: 'T'});
+  assert.equal(opts.effort, 'medium');
+  assert.equal(opts.maxTokens, 32000);
+  assert.equal(opts.system, pa.SYSTEM);
+  assert.match(opts.prompt, /Transkript:\n\[00:00\] Hallo$/);
+  assert.deepEqual(states, ['loading', 'writing', 'saving']);
+  const post = w.calls.find(c => c.startsWith('POST /podcasts/episodes/7/article/ '));
+  const body = JSON.parse(post.slice(post.indexOf('{')));
+  assert.equal(body.model, 'test-model');
+  assert.equal(body.kind, 'Monolog');
+  assert.match(body.body_html, /<a href="#t=61">▶ ab 01:01<\/a>/);
+});
+
+test('write: no transcript gives a German message', async () => {
+  stubWorld({pending: []});
+  await assert.rejects(pa.write(3), /noch kein Transkript/);
+});
+
+test('pickUp: claim, 409 skipped, failure releases the claim, notifier per success', async () => {
+  pa._reset();
+  const w = stubWorld({
+    pending: [{episode_id: 1, title: 'Eins'}, {episode_id: 2, title: 'Zwei'}, {episode_id: 4, title: 'Vier'}],
+    claims: {2: 409},
+    askImpl: async o => {
+      if (o.prompt.includes('Folge 4')) throw new Error('kaputt');
+      return {title: 'A', summary: 'S', kind: 'Gespräch', speakers: [], sections: [{heading: 'H', start: 0, body_markdown: 'B'}]};
+    },
+  });
+  const r = await pa.pickUp();
+  assert.deepEqual(r, {done: 1, failed: 1, skipped: 1});
+  assert.ok(w.calls.some(c => /^POST \/podcasts\/episodes\/2\/article\/claim\/ \{"device":"d-/.test(c)));
+  assert.ok(!w.calls.some(c => c.startsWith('GET /podcasts/episodes/2/')));      // 409: not written
+  assert.ok(w.calls.some(c => c.startsWith('DELETE /podcasts/episodes/4/article/claim/')));     // failed: released
+  assert.ok(!w.calls.some(c => c.startsWith('DELETE /podcasts/episodes/1/article/claim/')));
+  assert.deepEqual(w.notices, ['Artikel fertig: A']);
+  // The failed one is not tried again in this visit.
+  const w2 = stubWorld({pending: [{episode_id: 4, title: 'Vier'}]});
+  assert.deepEqual(await pa.pickUp(), {done: 0, failed: 0, skipped: 1});
+  assert.ok(!w2.calls.some(c => c.includes('claim')));
+});
+
+test('pickUp: android bridge notifies instead of the toast', async () => {
+  pa._reset();
+  const w = stubWorld({pending: [{episode_id: 1, title: 'Eins'}]});
+  const got = [];
+  globalThis.DioraNative = {notify: (...a) => got.push(a)};
+  await pa.pickUp();
+  assert.deepEqual(got, [['Artikel fertig', 'Titel', 'episode:1']]);
+  assert.deepEqual(w.notices, []);
+  globalThis.DioraNative = undefined;
+});
+
+test('pickUp: no key does nothing, not even a request; no second run at the same time', async () => {
+  pa._reset();
+  const w = stubWorld({pending: [{episode_id: 1, title: 'x'}], hasKey: false});
+  assert.deepEqual(await pa.pickUp(), {done: 0, failed: 0, skipped: 0});
+  assert.deepEqual(w.calls, []);
+
+  const w2 = stubWorld({pending: [{episode_id: 1, title: 'x'}]});
+  const a = pa.pickUp();
+  const b = await pa.pickUp();
+  assert.deepEqual(b, {busy: true});
+  assert.equal((await a).done, 1);
+  assert.equal(w2.calls.filter(c => c.startsWith('POST') && c.includes('/claim/ {"device":')).length, 1);
+});

tui/README.md

@@ -0,0 +1,126 @@
+# diora-tui
+
+EPUB-Reader im Terminal — die erste Stufe einer TUI-Version von diora.
+
+Zeigt ein Buch als **eine durchgehende Ansicht** über alle Kapitel hinweg (wie diora's
+Web-Reader), nicht Kapitel für Kapitel. Lesefortschritt wird als dieselbe
+`"blockIndex:innerFraction"`-Positionsangabe geführt wie der Web-Reader (`books/models.py`,
+`EBookProgress`) — `blockIndex` zählt dabei exakt wie `static/js/app.js`s
+`EPUB_BLOCK_SELECTOR` (`p, h1-h6, li, blockquote, dt, dd, figcaption` + textbasierte
+`div`s ohne Element-Kinder), fortlaufend über das ganze Buch. Dadurch ist die Position
+zwischen TUI und Web-Reader direkt vergleichbar, und `diora-tui sync` kann Fortschritt in
+beide Richtungen synchronisieren (siehe unten).
+
+Fortschritt wird — genau wie im Web-Reader (`save_progress`) — nur vorwärts überschrieben
+("furthest wins"), lokal in `progress.json` (`~/.local/share/diora-tui/`, via
+`platformdirs`), damit ein älterer/gestaffelter Lauf nie eine bereits weiter gelesene
+Position zurücksetzt.
+
+## Setup
+
+```bash
+cd tui
+pip install -e .
+```
+
+## Nutzung
+
+```bash
+diora-tui --library ~/Books   # Standard: ~/Books
+```
+
+Tastenkürzel:
+
+- `↑`/`k`, `↓`/`j` — zeilenweise scrollen
+- `n` — nächstes Kapitel, `p` — vorheriges Kapitel
+- `f` — Fußnote in der Nähe der aktuellen Position anzeigen (Peek-Overlay, `Escape`/`f`/`q`
+  zum Schließen); Erkennung folgt derselben Heuristik wie `app.js`s
+  `_looksLikeFootnoteLink` (Link in/um `<sup>`, Klassenname mit note/footnote/fn, oder
+  `epub:type="noteref"`)
+- `Enter` — markiertes Buch aus der Bibliothek öffnen
+- `r` (in der Bibliothek) — gelesene Bücher ein-/ausblenden (siehe unten)
+- `Escape` / `q` — zurück zur Bibliothek (im Reader) bzw. beenden (in der Bibliothek)
+
+Der Fließtext ist auf 120 Zeichen Breite begrenzt und horizontal zentriert (lesbarer als
+volle Terminalbreite bei breiten Fenstern).
+
+Unten rechts zeigt eine Statusleiste Akkustand (falls vorhanden, via `psutil`) und Uhrzeit,
+sekündlich aktualisiert, auf Bibliotheks- und Reader-Ansicht.
+
+Die Bibliotheksansicht ist nach zuletzt geöffnetem Buch sortiert (neueste zuerst; anhand
+des Zeitstempels der zuletzt gespeicherten Position), und blendet gelesene Bücher
+standardmäßig aus (`EBook.is_read` aus dem Sync-Snapshot, lokal in `library.json`
+gespiegelt) — `r` zeigt sie wieder an, für diese Sitzung.
+
+## Bücher + Fortschritt vom Server holen (`diora-tui sync`)
+
+Sobald einmal Zugangsdaten gespeichert sind (`~/.config/diora-tui/config.json`, siehe
+unten), synct `diora-tui` **automatisch** — einmal leise im Hintergrund beim Start (neue
+Bücher + Fortschritt werden nachgeladen, die Bibliotheksliste aktualisiert sich von
+selbst) und einmal beim Beenden über `q` (kurzer Moment Verzögerung, bevor die App
+tatsächlich schließt). Der explizite Befehl ist für's Ersteinrichten und für
+Nicht-interaktive Nutzung (Cron o.ä.):
+
+```bash
+diora-tui sync                              # nutzt gespeicherte Zugangsdaten, sonst interaktive Abfrage
+diora-tui sync --server https://diora.creamfresh.xyz --save   # einmalig einrichten + speichern
+```
+
+Lädt alle EPUBs des Accounts über `GET /api/sync/` + `GET /books/<id>/data/` herunter,
+entschlüsselt sie lokal (AES-256-GCM, kompatibel zu `static/js/app.js`) und legt sie als
+normale `.epub`-Dateien in `--library` ab (Dateiname `<id> - <Titel>.epub`) — von da an
+funktionieren sie wie jedes andere lokale Buch. Bereits heruntergeladene Bücher werden
+beim nächsten Lauf übersprungen (kein erneuter Download). Der Fortschritt aus dem Snapshot
+wird dabei ebenfalls übernommen (nur vorwärts, wie lokal auch).
+
+Während ein so heruntergeladenes Buch geöffnet ist (erkennbar am `<id> - `-Dateinamens-
+Präfix), schickt der Reader Fortschritts-Updates zusätzlich zurück an den Server
+(`POST /books/<id>/progress/`, `force: false` — überschreibt also nie eine weiter
+gelesene Position, egal ob die vom Web-Reader oder einem anderen Gerät stammt). Rein
+lokale Bücher (ohne dieses Präfix) bleiben unangetastet, kein Netzwerkzugriff.
+
+Dafür nötig, beim ersten Lauf abgefragt (danach optional lokal gespeichert unter
+`~/.config/diora-tui/config.json`, `chmod 600`):
+
+- **Server-URL** — z.B. `https://diora.creamfresh.xyz`.
+- **API-Token** — diora → Einstellungen (`/accounts/settings/`) → "Personal Access Token".
+- **Verschlüsselungs-Key** — der AES-256-Schlüssel, mit dem deine Bücher clientseitig
+  verschlüsselt wurden. Der `sync`-Prompt bietet zwei Wege:
+  1. **Aus Benutzername + Passwort ableiten** (Standard) — reproduziert exakt, was diora's
+     "Unlock with password"-Formular im Browser tut (PBKDF2-HMAC-SHA256, 200.000
+     Iterationen, Salt `"diora:" + username`; siehe `static/js/app.js:deriveAndStoreKey`).
+     Funktioniert nur, wenn der Account diesen Weg im Browser mindestens einmal benutzt
+     hat — sonst wurde der Key ursprünglich zufällig im Browser erzeugt, und diese
+     Ableitung trifft ihn nicht. `sync` meldet einen falschen Key als
+     Entschlüsselungsfehler (harmlos, kein Datenverlust), nie als falsches Ergebnis.
+  2. **Base64-Key direkt einfügen** — für den Fall, dass Weg 1 nicht passt. Der
+     Export-Button dafür ist im Browser-UI aktuell nicht verdrahtet (`exportEncKey()` in
+     `app.js` existiert, hat aber keinen sichtbaren Button); bis das nachgezogen ist, in
+     der Browser-Devtools-Konsole auf der diora-Seite (nicht `/accounts/settings/` — die
+     lädt `app.js` nicht) ausführen: `await exportEncKey()` — kopiert den Key ins
+     Clipboard, von dort ins `sync`-Prompt einfügen. Schlägt das mit einem
+     `NotAllowedError`/`InvalidAccessError` fehl, stattdessen direkt aus `localStorage`
+     lesen: `localStorage.getItem('diora_enc_key_' + window.USER_ID)`.
+
+Der Key/Token wird genauso vertrauensvoll behandelt wie im Web-Client (dort liegt der
+Schlüssel unverschlüsselt in `localStorage`): lokal als Klartext in einer 0600-Datei.
+
+## Performance bei großen Büchern
+
+Sehr große Bücher (mehrstellige Tausend Absätze — z.B. Sammelbände) können beim
+*ersten* Öffnen mehrere Sekunden bis niedrige zweistellige Sekunden brauchen (Parsing +
+Zeilenumbruch-Berechnung für die durchgehende Ansicht). Ein zweites Öffnen desselben
+Buchs bei gleicher Terminalbreite ist dank Cache (`~/.cache/diora-tui/layout_cache/`)
+deutlich schneller. Das Rendering selbst skaliert nicht mit der Buchgröße — nur die
+tatsächlich sichtbaren Zeilen werden gezeichnet (Textual Line API), nicht das ganze Buch
+auf einmal.
+
+## Grenzen der aktuellen Version
+
+- Nur EPUB, kein PDF — `sync` lädt PDFs im Account gar nicht erst herunter (übersprungen,
+  wird gemeldet), da der Reader sie ohnehin nicht darstellen kann.
+- Text wird als Fließtext ohne Bild-/Layout-Rendering dargestellt.
+- `innerFraction` (die Position *innerhalb* eines Blocks) ist eine Terminal-Näherung
+  (zeilenbasiert statt pixelbasiert wie im Browser) — für die Fortschritts-Sortierung
+  zählt primär `blockIndex`, der exakt mit dem Web-Reader übereinstimmt; `innerFraction`
+  ist nur ein Tiebreaker innerhalb desselben Blocks.

tui/diora_tui/__init__.py

@@ -0,0 +1 @@
+"""diora-tui: lokaler EPUB-Reader, Vorstufe eines diora-TUI-Clients."""

tui/diora_tui/__main__.py

@@ -0,0 +1,4 @@
+from diora_tui.app import main
+
+if __name__ == "__main__":
+    main()

tui/diora_tui/api.py

@@ -0,0 +1,48 @@
+"""HTTP client for diora's sync API (see the repo's CLAUDE.md, "Sync API for
+local clients"). Status: that API is provisional — expect endpoint/field
+changes as the server-side implementation settles.
+"""
+
+from __future__ import annotations
+
+import requests
+
+
+class ApiError(Exception):
+    pass
+
+
+def _headers(token: str) -> dict[str, str]:
+    return {"Authorization": f"Bearer {token}"}
+
+
+def fetch_sync_snapshot(server_url: str, token: str) -> dict:
+    resp = requests.get(f"{server_url}/api/sync/", headers=_headers(token), timeout=30)
+    if resp.status_code == 401:
+        raise ApiError("Authentifizierung fehlgeschlagen — Token falsch oder abgelaufen?")
+    resp.raise_for_status()
+    return resp.json()
+
+
+def fetch_book_data(server_url: str, token: str, book_id: int) -> dict:
+    resp = requests.get(f"{server_url}/books/{book_id}/data/", headers=_headers(token), timeout=120)
+    if resp.status_code == 401:
+        raise ApiError("Authentifizierung fehlgeschlagen — Token falsch oder abgelaufen?")
+    if resp.status_code == 404:
+        raise ApiError(f"Buch {book_id} nicht gefunden (falscher Owner oder gelöscht?)")
+    resp.raise_for_status()
+    return resp.json()
+
+
+def post_progress(
+    server_url: str, token: str, book_id: int, *, scroll_fraction: float, position_anchor: str, force: bool = False
+) -> None:
+    body = {"scroll_fraction": scroll_fraction, "position_anchor": position_anchor, "force": force}
+    resp = requests.post(
+        f"{server_url}/books/{book_id}/progress/", headers=_headers(token), json=body, timeout=30
+    )
+    if resp.status_code == 401:
+        raise ApiError("Authentifizierung fehlgeschlagen — Token falsch oder abgelaufen?")
+    if resp.status_code == 404:
+        raise ApiError(f"Buch {book_id} nicht gefunden (falscher Owner oder gelöscht?)")
+    resp.raise_for_status()

tui/diora_tui/app.py

@@ -0,0 +1,231 @@
+"""diora-tui: terminal EPUB reader for diora, syncing books + reading progress
+against a diora server (see project README)."""
+
+from __future__ import annotations
+
+import argparse
+import asyncio
+import getpass
+from pathlib import Path
+
+from textual import work
+from textual.app import App, ComposeResult
+from textual.binding import Binding
+from textual.screen import Screen
+from textual.widgets import Footer, Header, Label, ListItem, ListView
+
+from . import config as config_mod
+from . import crypto, epub, library_meta, progress, remote
+from .reader_screen import ReaderScreen
+from .statusbar import StatusBar
+
+DEFAULT_LIBRARY = Path.home() / "Books"
+
+
+class LibraryScreen(Screen):
+    BINDINGS = [
+        Binding("enter", "open_selected", "Öffnen"),
+        Binding("r", "toggle_read_filter", "Gelesene ein-/ausblenden"),
+        Binding("q", "quit", "Beenden"),
+    ]
+
+    def __init__(self, library_dir: Path) -> None:
+        super().__init__()
+        self.library_dir = library_dir
+        self.paths: list[Path] = []
+        self.visible_paths: list[Path] = []
+        self.show_read = False
+
+    def compose(self) -> ComposeResult:
+        yield Header()
+        yield ListView(id="library-list")
+        yield StatusBar()
+        yield Footer()
+
+    def on_mount(self) -> None:
+        self.sub_title = str(self.library_dir)
+        self._refresh_list()
+        cfg = config_mod.load()
+        if cfg is not None:
+            self._auto_sync(cfg)
+
+    @work(thread=True)
+    def _auto_sync(self, cfg: config_mod.RemoteConfig) -> None:
+        try:
+            result = remote.sync_library(self.library_dir, cfg)
+        except remote.SyncError:
+            return  # best-effort — e.g. offline; the local library still works
+        self.app.call_from_thread(self._on_auto_sync_done, result)
+
+    def _on_auto_sync_done(self, result: remote.SyncResult) -> None:
+        if result.downloaded or result.progress_pulled:
+            self._refresh_list()
+        if result.downloaded:
+            self.notify(f"{len(result.downloaded)} neue(s) Buch/Bücher synchronisiert.", timeout=3)
+
+    def _refresh_list(self) -> None:
+        self.paths = epub.scan_library(self.library_dir)
+
+        entries = []
+        for path in self.paths:
+            book_id = epub._book_id(path)
+            read = library_meta.is_read(book_id)
+            if read and not self.show_read:
+                continue
+            saved = progress.load(book_id)
+            last_opened = saved.updated_at if saved else 0.0
+            entries.append((last_opened, path, read))
+        entries.sort(key=lambda e: e[0], reverse=True)
+        self.visible_paths = [path for _, path, _ in entries]
+
+        list_view = self.query_one("#library-list", ListView)
+        list_view.clear()
+        if not self.paths:
+            list_view.append(ListItem(Label(f"Keine EPUBs gefunden in {self.library_dir}")))
+            return
+        if not entries:
+            list_view.append(ListItem(Label("Alle Bücher als gelesen markiert — 'r' zum Anzeigen")))
+            return
+        for _, path, read in entries:
+            label = f"✓ {path.stem}" if read else path.stem
+            list_view.append(ListItem(Label(label)))
+        list_view.index = 0
+        list_view.focus()
+
+    def action_toggle_read_filter(self) -> None:
+        self.show_read = not self.show_read
+        self._refresh_list()
+        state = "eingeblendet" if self.show_read else "ausgeblendet"
+        self.notify(f"Gelesene Bücher {state}.", timeout=2)
+
+    def action_open_selected(self) -> None:
+        list_view = self.query_one("#library-list", ListView)
+        if not self.visible_paths or list_view.index is None:
+            return
+        self.app.push_screen(ReaderScreen(self.visible_paths[list_view.index]))
+
+    def on_list_view_selected(self, event: ListView.Selected) -> None:
+        self.action_open_selected()
+
+
+class DioraTuiApp(App):
+    CSS_PATH = "app.tcss"
+    TITLE = "diora-tui"
+
+    def __init__(self, library_dir: Path) -> None:
+        super().__init__()
+        self.library_dir = library_dir
+
+    def on_mount(self) -> None:
+        self.push_screen(LibraryScreen(self.library_dir))
+
+    async def action_quit(self) -> None:
+        cfg = config_mod.load()
+        if cfg is not None:
+            try:
+                await asyncio.to_thread(remote.sync_library, self.library_dir, cfg)
+            except remote.SyncError:
+                pass  # best-effort — don't block quitting on a sync failure
+        self.exit()
+
+
+def _prompt(label: str, *, secret: bool = False) -> str:
+    value = (getpass.getpass(f"{label}: ") if secret else input(f"{label}: ")).strip()
+    if not value:
+        raise SystemExit(f"Abgebrochen: {label} darf nicht leer sein.")
+    return value
+
+
+def _ask_yes_no(question: str) -> bool:
+    return input(f"{question} [y/N]: ").strip().lower() in ("y", "yes", "j", "ja")
+
+
+def _prompt_enc_key() -> str:
+    print()
+    print("Verschlüsselungs-Key — zwei Wege:")
+    print("  [1] Aus Benutzername + Passwort ableiten (wie diora's 'Unlock'-Formular im Browser)")
+    print("  [2] Base64-Key direkt einfügen (z.B. per Browser-Konsole exportiert)")
+    choice = input("Wahl [1/2, Standard 1]: ").strip() or "1"
+    if choice == "2":
+        return _prompt("Verschlüsselungs-Key (Base64)", secret=True)
+
+    print(
+        "Hinweis: das liefert nur dann den richtigen Key, wenn dieser Account je über "
+        "diora's Passwort-Ableitung ('Unlock with password' im Browser) entsperrt wurde. "
+        "War der Key dort nur automatisch zufällig erzeugt, kommt hier ein anderer "
+        "(falscher) Key raus — 'sync' meldet das dann als Entschlüsselungsfehler, ohne "
+        "etwas kaputtzumachen; in dem Fall stattdessen [2] mit dem exportierten Key nutzen."
+    )
+    username = _prompt("diora-Benutzername")
+    password = _prompt("diora-Passwort", secret=True)
+    return crypto.derive_key_b64(username, password)
+
+
+def _resolve_remote_config(server_override: str | None, *, save: bool) -> config_mod.RemoteConfig:
+    cfg = None if server_override else config_mod.load()
+    if cfg is not None:
+        return cfg
+
+    print("Keine gespeicherten Zugangsdaten gefunden — bitte einmalig eingeben.")
+    server_url = (server_override or _prompt("Server-URL (z.B. https://diora.creamfresh.xyz)")).rstrip("/")
+    token = _prompt("API-Token (diora → Einstellungen → /accounts/settings/)", secret=True)
+    enc_key = _prompt_enc_key()
+    cfg = config_mod.RemoteConfig(server_url=server_url, api_token=token, enc_key_b64=enc_key)
+    if save or _ask_yes_no("Zugangsdaten lokal speichern, damit du sie nicht erneut eingeben musst?"):
+        config_mod.save(cfg)
+        print("Gespeichert.")
+    return cfg
+
+
+def run_sync(library_dir: Path, server_override: str | None, *, save: bool) -> None:
+    cfg = _resolve_remote_config(server_override, save=save)
+    print(f"Verbinde zu {cfg.server_url} …")
+    try:
+        result = remote.sync_library(library_dir, cfg)
+    except remote.SyncError as e:
+        raise SystemExit(f"Fehler: {e}")
+
+    print(f"{len(result.downloaded)} Buch/Bücher neu heruntergeladen nach {library_dir}.")
+    for name in result.downloaded:
+        print(f"  + {name}")
+    if result.unchanged:
+        print(f"{result.unchanged} Buch/Bücher bereits lokal vorhanden, übersprungen.")
+    if result.skipped_pdf:
+        titles = ", ".join(result.skipped_pdf)
+        print(f"{len(result.skipped_pdf)} PDF(s) übersprungen (TUI liest aktuell nur EPUB): {titles}")
+    if result.failed:
+        titles = ", ".join(result.failed)
+        print(f"{len(result.failed)} Buch/Bücher konnten nicht entschlüsselt werden (falscher Key?): {titles}")
+    if result.progress_pulled:
+        print(f"Lesefortschritt für {result.progress_pulled} Buch/Bücher vom Server übernommen.")
+
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        description="diora-tui — lokaler EPUB-Reader (Vorstufe des diora-Sync-Clients)"
+    )
+    parser.add_argument(
+        "--library",
+        type=Path,
+        default=DEFAULT_LIBRARY,
+        help=f"Verzeichnis mit EPUB-Dateien (Standard: {DEFAULT_LIBRARY})",
+    )
+    subparsers = parser.add_subparsers(dest="command")
+    sync_parser = subparsers.add_parser("sync", help="EPUBs vom diora-Server holen und entschlüsseln")
+    sync_parser.add_argument("--server", help="Server-URL, überschreibt gespeicherte Zugangsdaten für diesen Lauf")
+    sync_parser.add_argument(
+        "--save", action="store_true", help="Eingegebene Zugangsdaten lokal speichern, ohne zu fragen"
+    )
+
+    args = parser.parse_args()
+    library_dir = args.library.expanduser()
+
+    if args.command == "sync":
+        run_sync(library_dir, args.server, save=args.save)
+        return
+
+    DioraTuiApp(library_dir).run()
+
+
+if __name__ == "__main__":
+    main()

tui/diora_tui/app.tcss

@@ -0,0 +1,49 @@
+#library-list {
+    height: 1fr;
+}
+
+StatusBar {
+    dock: bottom;
+    height: 1;
+    padding: 0 1;
+    background: $panel;
+    color: $text-muted;
+    text-align: right;
+}
+
+#reader-scroll {
+    height: 1fr;
+    padding: 1 2;
+    align-horizontal: center;
+}
+
+#reader-loading {
+    height: 1fr;
+}
+
+FootnoteScreen {
+    align: center middle;
+}
+
+#footnote-body {
+    width: 80%;
+    max-width: 100;
+    height: auto;
+    max-height: 80%;
+    background: $panel;
+    border: round $primary;
+    padding: 1 2;
+}
+
+.footnote-note {
+    margin-bottom: 1;
+}
+
+#footnote-hint {
+    dock: bottom;
+    width: 80%;
+    max-width: 100;
+    background: $panel;
+    color: $text-muted;
+    padding: 0 2;
+}

tui/diora_tui/blocks.py

@@ -0,0 +1,214 @@
+"""Whole-book flat block extraction, matching static/js/app.js's EPUB_BLOCK_SELECTOR
+and getPositionAnchor()/restoreFromAnchor() exactly in *structure* (blockIndex is
+purely DOM order, no rendering needed) so position anchors are comparable across
+the web reader and this TUI. See CLAUDE.md's `tui/` section for the full picture.
+
+innerFraction on the web is defined via getBoundingClientRect() pixel geometry,
+which has no TUI equivalent — a terminal has no font metrics/reflow the way a
+browser does. We approximate it with row-based geometry from Rich's own text
+wrapping (see reader.py), which is close enough because the server's "furthest
+wins" comparison (_progress_is_further) only falls back to comparing fractions
+when two anchors share the exact same block index — block index is the primary,
+exactly-reproducible signal.
+
+Footnote references are detected with the same heuristic as app.js's
+_looksLikeFootnoteLink: wrapped in/wrapping a <sup>, a class name containing
+note/footnote/fn, or an epub:type="noteref" attribute.
+"""
+
+from __future__ import annotations
+
+import re
+import warnings
+from dataclasses import dataclass, field
+from pathlib import Path
+
+import ebooklib
+from bs4 import BeautifulSoup, NavigableString, Tag, XMLParsedAsHTMLWarning
+from ebooklib import epub
+
+# EPUB content documents are XHTML; treating them as HTML (matching app.js's
+# `new DOMParser().parseFromString(html, 'text/html')`, static/js/app.js:2638)
+# is intentional, not a mistake — silence bs4's XML-vs-HTML nudge for it.
+# lxml's HTML parser is ~3x faster than bs4's built-in html.parser, which
+# matters here: some real-world EPUBs run to tens of thousands of blocks.
+warnings.filterwarnings("ignore", category=XMLParsedAsHTMLWarning)
+_PARSER = "lxml"
+
+# Matches app.js's EPUB_BLOCK_SELECTOR = 'p, h1, h2, h3, h4, h5, h6, li,
+# blockquote, dt, dd, figcaption, div:not(:has(*))'
+_BLOCK_TAGS = {"p", "h1", "h2", "h3", "h4", "h5", "h6", "li", "blockquote", "dt", "dd", "figcaption"}
+# Matches app.js's sanitizeEpubHtml() strip list (script/style are also stripped
+# via regex before DOMParser even runs there; decomposing here is equivalent).
+_STRIP_TAGS = ["script", "style", "iframe", "object", "embed", "head", "meta", "link"]
+# Matches app.js's _looksLikeFootnoteLink's class-name check.
+_FOOTNOTE_CLASS_RE = re.compile(r"\bnote|\bfootnote|\bfn\b")
+
+# Anchor format the server accepts (books/views.py:save_progress); anything else
+# is silently discarded back to ''.
+_ANCHOR_RE = re.compile(r"\d{1,7}:\d(\.\d{1,6})?")
+
+
+@dataclass
+class FootnoteRef:
+    marker: str  # visible text of the reference link, e.g. "1"
+    offset: int  # character offset into the block's text where the marker sits
+    target_id: str  # fragment id to resolve against block ids
+
+
+@dataclass
+class Block:
+    text: str
+    tag: str
+    chapter_index: int
+    id: str | None = None
+    footnotes: list[FootnoteRef] = field(default_factory=list)
+    # Every id found anywhere in this block's subtree, not just on the block
+    # element itself — footnote *targets* are frequently an <a id="..."> or
+    # similar nested a level or two inside the actual containing paragraph
+    # (see showFootnotePopover's `.closest('.footnote') || .parentElement`
+    # walk-up in app.js), so a target id resolves to "the block containing
+    # it" rather than requiring the id to sit on the block tag itself.
+    ids: list[str] = field(default_factory=list)
+
+
+@dataclass
+class FlatBook:
+    id: str
+    title: str
+    author: str
+    path: Path
+    blocks: list[Block]
+    chapter_titles: list[str]
+    chapter_start_block: list[int]  # blocks[chapter_start_block[i]] is chapter i's first block
+    footnote_targets: dict[str, int]  # fragment id -> block index of its content
+
+
+def _is_leaf_div(tag: Tag) -> bool:
+    return tag.name == "div" and tag.find(True) is None
+
+
+def _looks_like_footnote_link(el: Tag, href: str) -> bool:
+    if "#" not in href:
+        return False
+    cls = " ".join(el.get("class") or []).lower()
+    if _FOOTNOTE_CLASS_RE.search(cls):
+        return True
+    epub_type = (el.get("epub:type") or "").lower()
+    if "noteref" in epub_type:
+        return True
+    return el.find_parent("sup") is not None or el.find("sup") is not None
+
+
+def _walk_text(el: Tag, footnotes: list[FootnoteRef], out: list[str]) -> None:
+    for child in el.children:
+        if isinstance(child, NavigableString):
+            out.append(str(child))
+        elif isinstance(child, Tag):
+            if child.name == "a":
+                href = child.get("href") or ""
+                if _looks_like_footnote_link(child, href):
+                    marker = child.get_text(" ", strip=True)
+                    offset = len("".join(out))
+                    target_id = href.split("#", 1)[1] if "#" in href else ""
+                    footnotes.append(FootnoteRef(marker=marker, offset=offset, target_id=target_id))
+                    out.append(marker)
+                    continue
+            _walk_text(child, footnotes, out)
+
+
+def _extract_blocks(html: bytes, chapter_index: int) -> list[Block]:
+    soup = BeautifulSoup(html, _PARSER)
+    for name in _STRIP_TAGS:
+        for el in soup.find_all(name):
+            el.decompose()
+
+    blocks: list[Block] = []
+    for el in soup.find_all(True):
+        if el.name in _BLOCK_TAGS or _is_leaf_div(el):
+            footnotes: list[FootnoteRef] = []
+            parts: list[str] = []
+            _walk_text(el, footnotes, parts)
+            text = re.sub(r"\s+", " ", "".join(parts)).strip()
+            ids = [tag_id for tag_id in (el.get("id"), *(d.get("id") for d in el.find_all(True))) if tag_id]
+            blocks.append(
+                Block(
+                    text=text,
+                    tag=el.name,
+                    chapter_index=chapter_index,
+                    id=el.get("id"),
+                    footnotes=footnotes,
+                    ids=ids,
+                )
+            )
+    return blocks
+
+
+def load_flat_book(path: Path) -> FlatBook:
+    from .epub import _book_id  # reuse the same path-hash id scheme
+
+    raw = epub.read_epub(str(path), options={"ignore_ncx": True})
+
+    title_meta = raw.get_metadata("DC", "title")
+    title = title_meta[0][0] if title_meta else path.stem
+    author_meta = raw.get_metadata("DC", "creator")
+    author = author_meta[0][0] if author_meta else "Unbekannt"
+
+    blocks: list[Block] = []
+    chapter_titles: list[str] = []
+    chapter_start_block: list[int] = []
+
+    # app.js's parseEpub() includes every spine itemref unconditionally — no
+    # `linear` filtering (static/js/app.js:2623-2625) — since footnote/endnote
+    # targets are commonly parked in a linear="no" document. Skipping it here
+    # would both break footnote-target resolution and shift blockIndex
+    # numbering out of sync with the web reader for every block after it.
+    for idref, _linear in raw.spine:
+        item = raw.get_item_with_id(idref)
+        if item is None or item.get_type() != ebooklib.ITEM_DOCUMENT:
+            continue
+        chapter_index = len(chapter_titles)
+        chapter_blocks = _extract_blocks(item.get_content(), chapter_index)
+        chapter_start_block.append(len(blocks))
+        first_text = next((b.text for b in chapter_blocks if b.text), None)
+        chapter_titles.append((first_text or item.get_name())[:60])
+        blocks.extend(chapter_blocks)
+
+    footnote_targets: dict[str, int] = {}
+    for idx, b in enumerate(blocks):
+        for tag_id in b.ids:
+            footnote_targets.setdefault(tag_id, idx)
+
+    return FlatBook(
+        id=_book_id(path),
+        title=title,
+        author=author,
+        path=path,
+        blocks=blocks,
+        chapter_titles=chapter_titles,
+        chapter_start_block=chapter_start_block,
+        footnote_targets=footnote_targets,
+    )
+
+
+def format_anchor(block_index: int, inner_fraction: float) -> str:
+    inner_fraction = max(0.0, min(1.0, inner_fraction))
+    return f"{block_index}:{inner_fraction:.6f}"
+
+
+def parse_anchor(anchor: str) -> tuple[int, float] | None:
+    if not anchor or not _ANCHOR_RE.fullmatch(anchor):
+        return None
+    block_str, _, frac_str = anchor.partition(":")
+    return int(block_str), float(frac_str)
+
+
+def anchor_is_further(new_anchor: str, old_anchor: str) -> bool:
+    """Mirrors _progress_is_further / _cmpProgress (books/views.py, app.js)."""
+    new_parts = parse_anchor(new_anchor)
+    old_parts = parse_anchor(old_anchor)
+    if new_parts is None or old_parts is None:
+        return bool(new_anchor) and not old_anchor
+    nb, ni = new_parts
+    ob, oi = old_parts
+    return ni >= oi if nb == ob else nb >= ob

tui/diora_tui/book_view.py

@@ -0,0 +1,42 @@
+"""Line-API scroll view for the continuous reader — renders only the rows
+actually visible on screen (via Widget.render_line), reading from the flat
+Strip list diora_tui.layout.build_layout() precomputed once. This is what
+keeps very large books responsive: nothing here scales with book size at
+paint time, only with viewport height.
+"""
+
+from __future__ import annotations
+
+from textual.geometry import Size
+from textual.scroll_view import ScrollView
+from textual.strip import Strip
+
+from .layout import BookLayout
+
+
+class ContinuousBookView(ScrollView):
+    # We always wrap text to fit exactly, so a horizontal scrollbar should
+    # never be needed — and "scroll" (not "auto") for the vertical one keeps
+    # its gutter reserved from the very first (still-empty) layout pass, so
+    # the width we wrap text at later never has to guess whether a scrollbar
+    # will appear and steal columns out from under already-wrapped lines.
+    DEFAULT_CSS = """
+    ContinuousBookView {
+        overflow-x: hidden;
+        overflow-y: scroll;
+    }
+    """
+
+    def __init__(self, book_layout: BookLayout) -> None:
+        super().__init__()
+        self.book_layout = book_layout
+        self.virtual_size = Size(book_layout.width, book_layout.total_rows)
+
+    def render_line(self, y: int) -> Strip:
+        _scroll_x, scroll_y = self.scroll_offset
+        row = scroll_y + y
+        strips = self.book_layout.row_strips
+        width = self.scrollable_content_region.width
+        if row < 0 or row >= len(strips):
+            return Strip.blank(width, self.rich_style)
+        return strips[row].crop_extend(0, width, self.rich_style)

tui/diora_tui/cache.py

@@ -0,0 +1,52 @@
+"""On-disk cache for the (FlatBook, BookLayout) pair a book open computes —
+extraction + row-layout for very large books (tens of thousands of blocks)
+can take several seconds each; reopening the same book at the same terminal
+width should be near-instant instead of paying that cost again every time.
+
+Not a correctness-critical cache: any miss (new book, different width, edited
+file) just falls back to recomputing from scratch, so a stale/corrupt cache
+entry is handled by overwriting it, never by crashing the reader.
+"""
+
+from __future__ import annotations
+
+import hashlib
+import pickle
+from pathlib import Path
+
+from platformdirs import user_cache_dir
+
+from .blocks import FlatBook
+from .layout import BookLayout
+
+_CACHE_DIR = Path(user_cache_dir("diora-tui", "diora")) / "layout_cache"
+
+
+def _cache_key(path: Path, width: int) -> str:
+    stat = path.stat()
+    raw = f"{path.resolve()}|{stat.st_size}|{stat.st_mtime_ns}|{width}"
+    return hashlib.sha256(raw.encode()).hexdigest()[:32]
+
+
+def load(path: Path, width: int) -> tuple[FlatBook, BookLayout] | None:
+    cache_file = _CACHE_DIR / f"{_cache_key(path, width)}.pickle"
+    if not cache_file.exists():
+        return None
+    try:
+        with cache_file.open("rb") as f:
+            book, book_layout = pickle.load(f)
+        if not isinstance(book, FlatBook) or not isinstance(book_layout, BookLayout):
+            return None
+        return book, book_layout
+    except Exception:
+        return None
+
+
+def save(path: Path, width: int, book: FlatBook, book_layout: BookLayout) -> None:
+    _CACHE_DIR.mkdir(parents=True, exist_ok=True)
+    cache_file = _CACHE_DIR / f"{_cache_key(path, width)}.pickle"
+    try:
+        with cache_file.open("wb") as f:
+            pickle.dump((book, book_layout), f, protocol=pickle.HIGHEST_PROTOCOL)
+    except Exception:
+        pass  # best-effort — a failed cache write shouldn't break reading

tui/diora_tui/config.py

@@ -0,0 +1,42 @@
+"""Local storage for diora server credentials used by `diora-tui sync`.
+
+Mirrors the trust model of the web client's key storage (static/js/app.js,
+getOrCreateEncKey/exportEncKey): the raw AES-256 key and API token are kept
+in plaintext on disk, scoped to this machine/user (0600), with no additional
+at-rest encryption — same exposure as the browser's localStorage already has.
+"""
+
+from __future__ import annotations
+
+import json
+import stat
+from dataclasses import asdict, dataclass
+from pathlib import Path
+
+from platformdirs import user_config_dir
+
+_CONFIG_DIR = Path(user_config_dir("diora-tui", "diora"))
+_CONFIG_FILE = _CONFIG_DIR / "config.json"
+
+
+@dataclass
+class RemoteConfig:
+    server_url: str
+    api_token: str
+    enc_key_b64: str
+
+
+def load() -> RemoteConfig | None:
+    if not _CONFIG_FILE.exists():
+        return None
+    try:
+        data = json.loads(_CONFIG_FILE.read_text())
+        return RemoteConfig(**data)
+    except (json.JSONDecodeError, OSError, TypeError):
+        return None
+
+
+def save(cfg: RemoteConfig) -> None:
+    _CONFIG_DIR.mkdir(parents=True, exist_ok=True)
+    _CONFIG_FILE.write_text(json.dumps(asdict(cfg), indent=2))
+    _CONFIG_FILE.chmod(stat.S_IRUSR | stat.S_IWUSR)

tui/diora_tui/crypto.py

@@ -0,0 +1,51 @@
+"""AES-256-GCM helpers matching static/js/app.js's encryptBytes/decryptBytes
+(Web Crypto AES-GCM, 12-byte IV hex-encoded, ciphertext base64, key handled
+as raw bytes) — diora's books are end-to-end encrypted client-side, so the
+server (and this module) only ever sees ciphertext plus the key the user
+supplies out-of-band.
+"""
+
+from __future__ import annotations
+
+import base64
+
+from cryptography.exceptions import InvalidTag
+from cryptography.hazmat.primitives import hashes
+from cryptography.hazmat.primitives.ciphers.aead import AESGCM
+from cryptography.hazmat.primitives.kdf.pbkdf2 import PBKDF2HMAC
+
+_PBKDF2_ITERATIONS = 200_000
+_KEY_LENGTH_BYTES = 32
+
+
+class DecryptError(Exception):
+    """Ciphertext could not be decrypted with the given key (wrong key, or corrupt data)."""
+
+
+def decrypt(key_b64: str, iv_hex: str, ciphertext_b64: str) -> bytes:
+    try:
+        key = base64.b64decode(key_b64)
+        iv = bytes.fromhex(iv_hex)
+        ct = base64.b64decode(ciphertext_b64)
+        return AESGCM(key).decrypt(iv, ct, None)
+    except (InvalidTag, ValueError) as e:
+        raise DecryptError(str(e)) from e
+
+
+def derive_key_b64(username: str, password: str) -> str:
+    """Re-derive the AES-256 key the same way app.js's deriveAndStoreKey() does:
+    PBKDF2-HMAC-SHA256, 200000 iterations, salt = "diora:" + username. Only
+    yields the key that actually decrypts a user's books if that account's
+    key was ever set up via diora's "unlock with password" flow — a browser
+    that only ever auto-generated a random key (the default) has a key this
+    can't reproduce.
+    """
+    salt = f"diora:{username}".encode("utf-8")
+    kdf = PBKDF2HMAC(
+        algorithm=hashes.SHA256(),
+        length=_KEY_LENGTH_BYTES,
+        salt=salt,
+        iterations=_PBKDF2_ITERATIONS,
+    )
+    raw = kdf.derive(password.encode("utf-8"))
+    return base64.b64encode(raw).decode()

tui/diora_tui/epub.py

@@ -0,0 +1,19 @@
+"""Library scanning and the stable per-book id shared across progress.py,
+blocks.py, and remote.py."""
+
+from __future__ import annotations
+
+import hashlib
+from pathlib import Path
+
+
+def _book_id(path: Path) -> str:
+    # Identifies a book by its resolved path for now. Once the sync API exists,
+    # this should switch to whatever stable id diora assigns server-side.
+    return hashlib.sha256(str(path.resolve()).encode()).hexdigest()[:16]
+
+
+def scan_library(library_dir: Path) -> list[Path]:
+    if not library_dir.exists():
+        return []
+    return sorted(library_dir.rglob("*.epub"))

tui/diora_tui/layout.py

@@ -0,0 +1,90 @@
+"""Row-based layout for the continuous reader view — the TUI's terminal-grid
+analogue of the browser's pixel-based getBoundingClientRect() geometry (see
+blocks.py's module docstring for why blockIndex is exact but innerFraction is
+only an approximation here).
+
+Each block is wrapped independently at a known width, once, into a flat list
+of pre-rendered Strip objects (one per terminal row) that book_view.py's
+Line-API widget indexes directly in render_line() — this is what makes very
+large books (tens of thousands of blocks) open in seconds rather than
+minutes: Textual only ever renders the rows actually on screen, instead of
+laying out the whole book up front the way a single giant Static would.
+"""
+
+from __future__ import annotations
+
+from dataclasses import dataclass
+
+from rich.console import Console
+from rich.text import Text
+from textual.strip import Strip
+
+_SEPARATOR_ROWS = 1  # one blank row between blocks, matching the old "\n\n".join style
+
+
+@dataclass
+class BookLayout:
+    starts: list[int]  # row where block i begins
+    heights: list[int]  # rendered row count for block i
+    row_strips: list[Strip]  # one Strip per absolute row, len == total_rows
+    total_rows: int
+    width: int
+
+
+def build_layout(block_texts: list[str], width: int, console: Console) -> BookLayout:
+    width = max(1, width)
+    starts: list[int] = []
+    heights: list[int] = []
+    row_strips: list[Strip] = []
+    blank = Strip.blank(width)
+    row = 0
+    for text in block_texts:
+        starts.append(row)
+        wrapped_lines = list(Text(text).wrap(console, width)) if text else []
+        if not wrapped_lines:
+            wrapped_lines = [Text("")]
+        heights.append(len(wrapped_lines))
+        for line in wrapped_lines:
+            strip = Strip(line.render(console), None).adjust_cell_length(width)
+            row_strips.append(strip)
+        row_strips.append(blank)
+        row += len(wrapped_lines) + _SEPARATOR_ROWS
+
+    return BookLayout(starts=starts, heights=heights, row_strips=row_strips, total_rows=row, width=width)
+
+
+def get_position_anchor(layout: BookLayout, scroll_y: int) -> tuple[int, float]:
+    """Mirrors app.js's getPositionAnchor(): the last block whose top row is at
+    or above scroll_y, or the first block if none has scrolled that far yet."""
+    n = len(layout.heights)
+    if n == 0:
+        return 0, 0.0
+
+    best_index = 0
+    found = False
+    for i in range(n):
+        if layout.heights[i] < 1:
+            continue
+        if layout.starts[i] > scroll_y:
+            break
+        best_index = i
+        found = True
+
+    if not found:
+        best_index = next((i for i in range(n) if layout.heights[i] >= 1), 0)
+
+    top = layout.starts[best_index]
+    height = max(1, layout.heights[best_index])
+    inner_fraction = max(0.0, min(1.0, (scroll_y - top) / height))
+    return best_index, inner_fraction
+
+
+def scroll_y_for_anchor(layout: BookLayout, block_index: int, inner_fraction: float) -> int:
+    """Mirrors app.js's restoreFromAnchor()."""
+    n = len(layout.heights)
+    if n == 0:
+        return 0
+    idx = max(0, min(block_index, n - 1))
+    top = layout.starts[idx]
+    height = layout.heights[idx]
+    return top + round(inner_fraction * height)

tui/diora_tui/library_meta.py

@@ -0,0 +1,43 @@
+"""Local per-book metadata pulled from the server — currently just "read"
+status (EBook.is_read from the /api/sync/ snapshot). Kept separate from
+progress.py (reading position): different concern, different cadence — this
+is only ever set by `diora-tui sync`, never by the reader itself.
+"""
+
+from __future__ import annotations
+
+import json
+from pathlib import Path
+
+from platformdirs import user_data_dir
+
+_DATA_DIR = Path(user_data_dir("diora-tui", "diora"))
+_META_FILE = _DATA_DIR / "library.json"
+
+
+def _load_all() -> dict[str, dict]:
+    if not _META_FILE.exists():
+        return {}
+    try:
+        return json.loads(_META_FILE.read_text())
+    except (json.JSONDecodeError, OSError):
+        return {}
+
+
+def _save_all(data: dict[str, dict]) -> None:
+    _DATA_DIR.mkdir(parents=True, exist_ok=True)
+    _META_FILE.write_text(json.dumps(data, indent=2))
+
+
+def is_read(book_id: str) -> bool:
+    entry = _load_all().get(book_id)
+    return bool(entry and entry.get("is_read"))
+
+
+def set_many(read_status: dict[str, bool]) -> None:
+    if not read_status:
+        return
+    data = _load_all()
+    for book_id, read in read_status.items():
+        data.setdefault(book_id, {})["is_read"] = read
+    _save_all(data)

tui/diora_tui/progress.py

@@ -0,0 +1,68 @@
+"""Local reading-progress storage.
+
+Positions are stored as the same "blockIndex:innerFraction" anchor string the
+server uses (books/models.py, EBookProgress.save_progress) — see
+diora_tui/blocks.py for how blockIndex is computed to match static/js/app.js
+exactly, and anchor_is_further() for the identical "furthest wins" comparison
+used server-side (_progress_is_further in books/views.py). A saved position
+only ever advances (unless forced), so an older/offline run can't regress a
+further-along read position — matching that same rule locally.
+"""
+
+from __future__ import annotations
+
+import json
+import time
+from dataclasses import dataclass
+from pathlib import Path
+
+from platformdirs import user_data_dir
+
+from .blocks import anchor_is_further
+
+_DATA_DIR = Path(user_data_dir("diora-tui", "diora"))
+_PROGRESS_FILE = _DATA_DIR / "progress.json"
+
+
+@dataclass
+class Position:
+    anchor: str  # "blockIndex:innerFraction", e.g. "42:0.500000"
+    updated_at: float = 0.0
+
+
+def _load_all() -> dict[str, dict]:
+    if not _PROGRESS_FILE.exists():
+        return {}
+    try:
+        return json.loads(_PROGRESS_FILE.read_text())
+    except (json.JSONDecodeError, OSError):
+        return {}
+
+
+def _save_all(data: dict[str, dict]) -> None:
+    _DATA_DIR.mkdir(parents=True, exist_ok=True)
+    _PROGRESS_FILE.write_text(json.dumps(data, indent=2))
+
+
+def load(book_id: str) -> Position | None:
+    entry = _load_all().get(book_id)
+    if not isinstance(entry, dict) or not entry.get("anchor"):
+        return None
+    return Position(anchor=entry["anchor"], updated_at=entry.get("updated_at", 0.0))
+
+
+def save(book_id: str, position: Position, *, force: bool = False) -> bool:
+    """Returns True if the position was actually written (i.e. it was further
+    along, or forced) — callers that push to the server only need to do so
+    when this returns True."""
+    data = _load_all()
+    existing = data.get(book_id)
+    if existing is not None and not force:
+        old_anchor = existing.get("anchor", "")
+        if not anchor_is_further(position.anchor, old_anchor):
+            return False
+    if not position.updated_at:
+        position.updated_at = time.time()
+    data[book_id] = {"anchor": position.anchor, "updated_at": position.updated_at}
+    _save_all(data)
+    return True

tui/diora_tui/reader_screen.py

@@ -0,0 +1,286 @@
+"""Continuous whole-book reader screen — mirrors the web reader's single
+scrollable view (see blocks.py's module docstring) instead of the old
+per-chapter pagination, so reading position is expressed in the same
+"blockIndex:innerFraction" anchor space as the server and the web client.
+
+Uses book_view.ContinuousBookView (a Line-API widget) rather than dumping the
+whole book into one Static: for large books (tens of thousands of blocks) a
+single giant renderable made Textual's layout/paint pass take upwards of a
+minute, whereas the Line API only ever renders the rows on screen.
+"""
+
+from __future__ import annotations
+
+import bisect
+import re
+from pathlib import Path
+
+from rich.console import Console
+from textual import work
+from textual.binding import Binding
+from textual.containers import Container, VerticalScroll
+from textual.geometry import Size
+from textual.screen import ModalScreen, Screen
+from textual.widgets import Footer, Header, Label, LoadingIndicator, Static
+
+from . import api, blocks, cache, config as config_mod, layout, progress
+from .book_view import ContinuousBookView
+from .statusbar import StatusBar
+
+MAX_LINE_WIDTH = 120
+AUTOSAVE_INTERVAL = 5.0
+# ContinuousBookView forces its vertical scrollbar always-on (see book_view.py)
+# so this stays constant — measuring the container's width and subtracting
+# this fixed amount avoids a measure-after-constrain race against book_view's
+# own (possibly already-constrained-from-a-previous-layout) width.
+SCROLLBAR_GUTTER = 2
+
+_EMPTY_LAYOUT = layout.BookLayout(starts=[], heights=[], row_strips=[], total_rows=0, width=1)
+
+# Downloaded-via-sync files are named "<server-id> - <title>.epub" (see
+# remote.py) — reused here to find the server book id for progress push.
+_SERVER_ID_RE = re.compile(r"^(\d{4,7}) - ")
+
+
+def _server_book_id(path: Path) -> int | None:
+    m = _SERVER_ID_RE.match(path.name)
+    return int(m.group(1)) if m else None
+
+
+class FootnoteScreen(ModalScreen[None]):
+    BINDINGS = [Binding("escape,f,q", "dismiss_self", "Schließen")]
+
+    def __init__(self, notes: list[str]) -> None:
+        super().__init__()
+        self._notes = notes
+
+    def compose(self):
+        with VerticalScroll(id="footnote-body"):
+            for note in self._notes:
+                yield Static(note, classes="footnote-note")
+        yield Label("Escape/f zum Schließen", id="footnote-hint")
+
+    def action_dismiss_self(self) -> None:
+        self.dismiss(None)
+
+
+class ReaderScreen(Screen):
+    BINDINGS = [
+        Binding("j,down", "scroll_down_line", "Runter", show=False),
+        Binding("k,up", "scroll_up_line", "Hoch", show=False),
+        Binding("n", "next_chapter", "Nächstes Kapitel"),
+        Binding("p", "prev_chapter", "Vorheriges Kapitel"),
+        Binding("f", "peek_footnote", "Fußnote"),
+        Binding("q,escape", "back", "Zurück zur Bibliothek"),
+    ]
+
+    def __init__(self, path: Path) -> None:
+        super().__init__()
+        self.path = path
+        self.book: blocks.FlatBook | None = None
+        self.book_layout: layout.BookLayout = _EMPTY_LAYOUT
+        self._server_id = _server_book_id(path)
+        self._remote_cfg = config_mod.load() if self._server_id is not None else None
+        self._last_pushed_anchor = ""
+
+    def compose(self):
+        yield Header()
+        yield LoadingIndicator(id="reader-loading")
+        with Container(id="reader-scroll") as scroll_container:
+            self._scroll_container = scroll_container
+            # Stays visible (empty) from the start rather than toggling display
+            # on once loaded — a widget that's just been switched from hidden
+            # to visible hasn't been through a layout pass yet, so its `.size`
+            # is still (0, 0) and an immediate scroll_to() right after has
+            # nothing to clamp against and silently resets to 0.
+            self._book_view = ContinuousBookView(_EMPTY_LAYOUT)
+            yield self._book_view
+        yield StatusBar()
+        yield Footer()
+
+    def on_mount(self) -> None:
+        # Deferred rather than read synchronously here: right after mount the
+        # container hasn't been through a layout pass yet, so its size isn't
+        # reliable — same lesson as the scroll-restore race below.
+        self.call_after_refresh(self._start_loading)
+
+    def _start_loading(self) -> None:
+        if self._scroll_container.size.width == 0:
+            # Not sized yet after all — keep deferring instead of guessing a
+            # fixed number of refresh cycles (mirrors _restore_position below).
+            self.call_after_refresh(self._start_loading)
+            return
+        self._load_book(self._content_width())
+
+    @work(thread=True)
+    def _load_book(self, width: int) -> None:
+        cached = cache.load(self.path, width)
+        if cached is not None:
+            book, book_layout = cached
+        else:
+            book = blocks.load_flat_book(self.path)
+            console = Console(width=width)
+            book_layout = layout.build_layout([b.text for b in book.blocks], width, console)
+            cache.save(self.path, width, book, book_layout)
+        self.app.call_from_thread(self._on_book_loaded, book, book_layout, width)
+
+    def _on_book_loaded(self, book: blocks.FlatBook, book_layout: layout.BookLayout, width: int) -> None:
+        self.book = book
+        self.sub_title = book.title
+        self._apply_layout(book_layout, width)
+        self.query_one("#reader-loading", LoadingIndicator).display = False
+        self.call_after_refresh(self._restore_position)
+        self.set_interval(AUTOSAVE_INTERVAL, self._autosave)
+
+    def _content_width(self) -> int:
+        # Measuring the *container* rather than book_view's own size avoids a
+        # measure-after-constrain race: once a layout has been applied,
+        # book_view's width is pinned to a previous value via styles.width
+        # (below), so re-measuring book_view itself on a later resize would
+        # just read that stale pinned width back instead of the new
+        # available space. The container's width is unaffected by that.
+        available = self._scroll_container.size.width - SCROLLBAR_GUTTER
+        return max(20, min(MAX_LINE_WIDTH, available))
+
+    def _apply_layout(self, book_layout: layout.BookLayout, width: int) -> None:
+        self.book_layout = book_layout
+        # Pin book_view's outer width to content-width-plus-scrollbar so its
+        # *inner* content region (what render_line actually draws into) ends
+        # up exactly `width` — matching what block texts were wrapped at.
+        self._book_view.styles.width = width + SCROLLBAR_GUTTER
+        self._book_view.book_layout = book_layout
+        self._book_view.virtual_size = Size(width, book_layout.total_rows)
+        self._book_view.refresh()
+
+    def _build_layout(self) -> None:
+        assert self.book is not None
+        width = self._content_width()
+        console = Console(width=width)
+        block_texts = [b.text for b in self.book.blocks]
+        new_layout = layout.build_layout(block_texts, width, console)
+        cache.save(self.path, width, self.book, new_layout)
+        self._apply_layout(new_layout, width)
+
+    def on_resize(self) -> None:
+        if self.book is None:
+            return
+        old_y = self._book_view.scroll_y
+        anchor_block, anchor_frac = layout.get_position_anchor(self.book_layout, old_y)
+        self._build_layout()
+        new_y = layout.scroll_y_for_anchor(self.book_layout, anchor_block, anchor_frac)
+        self._book_view.scroll_to(y=new_y, animate=False, immediate=True)
+
+    def _restore_position(self, attempt: int = 0) -> None:
+        if self.book is None:
+            return
+        saved = progress.load(self.book.id)
+        if saved is None:
+            return
+        parsed = blocks.parse_anchor(saved.anchor)
+        if parsed is None:
+            return
+        block_index, inner_fraction = parsed
+        y = layout.scroll_y_for_anchor(self.book_layout, block_index, inner_fraction)
+        if y <= 0:
+            return
+        self._book_view.scroll_to(y=y, animate=False, immediate=True)
+        # A widget that's only just become part of the layout doesn't always
+        # honor an immediate scroll on the first attempt (its own size/scroll
+        # bounds can still be mid-update) — verify it actually landed and
+        # retry a bounded number of times rather than guessing a fixed delay.
+        if attempt < 20 and abs(self._book_view.scroll_y - y) > 1:
+            self.call_after_refresh(lambda: self._restore_position(attempt + 1))
+
+    def _current_anchor_str(self) -> str:
+        block_index, inner_fraction = layout.get_position_anchor(self.book_layout, self._book_view.scroll_y)
+        return blocks.format_anchor(block_index, inner_fraction)
+
+    def _autosave(self) -> None:
+        self._save_progress()
+
+    def _save_progress(self) -> None:
+        if self.book is None or not self.book_layout.heights:
+            return
+        anchor = self._current_anchor_str()
+        advanced = progress.save(self.book.id, progress.Position(anchor=anchor))
+        if advanced and self._remote_cfg is not None and self._server_id is not None:
+            self._push_remote_progress(anchor)
+
+    @work(thread=True, exclusive=True, group="progress-push")
+    def _push_remote_progress(self, anchor: str) -> None:
+        if anchor == self._last_pushed_anchor or self._remote_cfg is None or self._server_id is None:
+            return
+        try:
+            api.post_progress(
+                self._remote_cfg.server_url,
+                self._remote_cfg.api_token,
+                self._server_id,
+                scroll_fraction=0.0,
+                position_anchor=anchor,
+                force=False,
+            )
+            self._last_pushed_anchor = anchor
+        except Exception:
+            pass  # best-effort — local progress is already saved regardless
+
+    def on_unmount(self) -> None:
+        self._save_progress()
+
+    def action_scroll_down_line(self) -> None:
+        self._book_view.scroll_relative(y=1, animate=False)
+
+    def action_scroll_up_line(self) -> None:
+        self._book_view.scroll_relative(y=-1, animate=False)
+
+    def _current_chapter_index(self) -> int:
+        if self.book is None:
+            return 0
+        block_index, _ = layout.get_position_anchor(self.book_layout, self._book_view.scroll_y)
+        return bisect.bisect_right(self.book.chapter_start_block, block_index) - 1
+
+    def _jump_to_block(self, block_index: int) -> None:
+        y = layout.scroll_y_for_anchor(self.book_layout, block_index, 0.0)
+        self._book_view.scroll_to(y=y, animate=False, immediate=True)
+
+    def action_next_chapter(self) -> None:
+        if self.book is None:
+            return
+        chapter = self._current_chapter_index()
+        if chapter + 1 < len(self.book.chapter_start_block):
+            self._jump_to_block(self.book.chapter_start_block[chapter + 1])
+            self._save_progress()
+
+    def action_prev_chapter(self) -> None:
+        if self.book is None:
+            return
+        chapter = self._current_chapter_index()
+        if chapter > 0:
+            self._jump_to_block(self.book.chapter_start_block[chapter - 1])
+            self._save_progress()
+
+    def action_peek_footnote(self) -> None:
+        if self.book is None or not self.book_layout.heights:
+            return
+        current_block, _ = layout.get_position_anchor(self.book_layout, self._book_view.scroll_y)
+
+        forward = [(i, b) for i, b in enumerate(self.book.blocks) if i >= current_block and b.footnotes]
+        backward = [(i, b) for i, b in enumerate(self.book.blocks) if i < current_block and b.footnotes]
+        candidate = forward[0] if forward else (backward[-1] if backward else None)
+        if candidate is None:
+            self.notify("Keine Fußnote in diesem Buch gefunden.", timeout=3)
+            return
+
+        _, block = candidate
+        notes: list[str] = []
+        for ref in block.footnotes:
+            target_idx = self.book.footnote_targets.get(ref.target_id)
+            if target_idx is None:
+                continue
+            notes.append(self.book.blocks[target_idx].text)
+        if not notes:
+            self.notify("Fußnote konnte nicht aufgelöst werden.", timeout=3)
+            return
+        self.app.push_screen(FootnoteScreen(notes))
+
+    def action_back(self) -> None:
+        self.app.pop_screen()

tui/diora_tui/remote.py

@@ -0,0 +1,110 @@
+"""Fetch + decrypt books from a diora server into the local TUI library, and
+pull reading progress for them into the local progress store.
+
+Progress is pull-only here (server -> local); the reader pushes local ->
+server itself while a book is open (see reader_screen.py). Both directions
+use the same "blockIndex:innerFraction" anchor format as the web reader
+(diora_tui/blocks.py) and the same furthest-wins merge rule
+(diora_tui/progress.py, books/views.py's _progress_is_further) — a pull can
+only ever advance local progress, never regress it, unless forced.
+"""
+
+from __future__ import annotations
+
+import json
+import re
+from dataclasses import dataclass, field
+from pathlib import Path
+
+import requests
+
+from . import api, crypto, library_meta
+from .config import RemoteConfig
+from .epub import _book_id
+from .progress import Position, save as save_progress
+
+
+class SyncError(Exception):
+    pass
+
+
+@dataclass
+class SyncResult:
+    downloaded: list[str] = field(default_factory=list)
+    unchanged: int = 0
+    skipped_pdf: list[str] = field(default_factory=list)
+    failed: list[str] = field(default_factory=list)
+    progress_pulled: int = 0
+
+
+def _sanitize_filename(name: str) -> str:
+    name = re.sub(r"[^\w\s.-]", "_", name).strip()
+    return (name or "buch")[:80]
+
+
+def _parse_meta(raw: bytes) -> dict:
+    return json.loads(raw.decode("utf-8"))
+
+
+def sync_library(library_dir: Path, cfg: RemoteConfig) -> SyncResult:
+    try:
+        snapshot = api.fetch_sync_snapshot(cfg.server_url, cfg.api_token)
+    except api.ApiError as e:
+        raise SyncError(str(e)) from e
+    except requests.RequestException as e:
+        raise SyncError(f"Verbindung zu {cfg.server_url} fehlgeschlagen: {e}") from e
+
+    library_dir.mkdir(parents=True, exist_ok=True)
+    result = SyncResult()
+    local_path_by_server_id: dict[int, Path] = {}
+
+    for book in snapshot.get("books", []):
+        book_id = book["id"]
+        try:
+            meta = _parse_meta(crypto.decrypt(cfg.enc_key_b64, book["meta_iv"], book["meta_ct"]))
+        except crypto.DecryptError:
+            result.failed.append(f"#{book_id}")
+            continue
+
+        if meta.get("type") == "pdf":
+            result.skipped_pdf.append(meta.get("title") or f"#{book_id}")
+            continue
+
+        existing = next(library_dir.glob(f"{book_id:04d} - *.epub"), None)
+        if existing is not None:
+            result.unchanged += 1
+            local_path_by_server_id[book_id] = existing
+            continue
+
+        try:
+            data = api.fetch_book_data(cfg.server_url, cfg.api_token, book_id)
+            raw = crypto.decrypt(cfg.enc_key_b64, data["data_iv"], data["data_ct"])
+        except (api.ApiError, crypto.DecryptError, requests.RequestException):
+            result.failed.append(meta.get("title") or f"#{book_id}")
+            continue
+
+        title = meta.get("title") or f"book-{book_id}"
+        dest = library_dir / f"{book_id:04d} - {_sanitize_filename(title)}.epub"
+        dest.write_bytes(raw)
+        result.downloaded.append(dest.name)
+        local_path_by_server_id[book_id] = dest
+
+    for entry in snapshot.get("book_progress", []):
+        anchor = entry.get("position_anchor") or ""
+        if not anchor:
+            continue  # PDF-only scroll_fraction progress — this TUI is EPUB-only
+        dest = local_path_by_server_id.get(entry.get("book_id"))
+        if dest is None:
+            continue
+        local_id = _book_id(dest)
+        if save_progress(local_id, Position(anchor=anchor)):
+            result.progress_pulled += 1
+
+    read_status = {
+        _book_id(path): bool(book.get("is_read"))
+        for book in snapshot.get("books", [])
+        if (path := local_path_by_server_id.get(book["id"])) is not None
+    }
+    library_meta.set_many(read_status)
+
+    return result

tui/diora_tui/statusbar.py

@@ -0,0 +1,44 @@
+"""Bottom status bar: battery level + current time, refreshed every second.
+Battery is read via psutil (cross-platform); on a desktop machine with no
+battery, psutil.sensors_battery() returns None and the bar just shows time.
+"""
+
+from __future__ import annotations
+
+from datetime import datetime
+
+from textual.widgets import Static
+
+try:
+    import psutil
+except ImportError:
+    psutil = None  # battery display degrades gracefully to time-only
+
+
+class StatusBar(Static):
+    def on_mount(self) -> None:
+        self._refresh()
+        self.set_interval(1.0, self._refresh)
+
+    def _refresh(self) -> None:
+        self.update(self._render_text())
+
+    def _render_text(self) -> str:
+        parts = []
+        battery = self._battery_text()
+        if battery:
+            parts.append(battery)
+        parts.append(datetime.now().strftime("%H:%M:%S"))
+        return "  ".join(parts)
+
+    def _battery_text(self) -> str | None:
+        if psutil is None:
+            return None
+        try:
+            battery = psutil.sensors_battery()
+        except Exception:
+            return None
+        if battery is None:
+            return None
+        state = "lädt" if battery.power_plugged else "Akku"
+        return f"{state} {round(battery.percent)}%"

tui/pyproject.toml

@@ -0,0 +1,25 @@
+[project]
+name = "diora-tui"
+version = "0.1.0"
+description = "Terminal-EPUB-Reader für diora mit Server-Sync für Bücher und Lesefortschritt."
+requires-python = ">=3.9"
+dependencies = [
+    "textual>=0.60",
+    "ebooklib>=0.18",
+    "beautifulsoup4>=4.12",
+    "lxml>=5.0",
+    "platformdirs>=4.0",
+    "requests>=2.31",
+    "cryptography>=42.0",
+    "psutil>=5.9",
+]
+
+[project.scripts]
+diora-tui = "diora_tui.app:main"
+
+[build-system]
+requires = ["hatchling"]
+build-backend = "hatchling.build"
+
+[tool.hatch.build.targets.wheel]
+packages = ["diora_tui"]