Bug fixes from full codebase review:
- clear() with DELETE_FILE_ON_TRASHCAN now also deletes chapter_files
and subtitle_files, not just the primary filename
- _PERSISTED_DOWNLOAD_FIELDS now includes 'subtitle_files' so subtitle
file info survives app restart
Tests:
- Added test_subtitle_lang_list_with_region_variant_input (zh-Hans)
- Added test_subtitle_lang_list_dedup_region_variant (pt-BR)
P1 fixes:
- _subtitle_lang_list now extracts base language from qualified tags
(e.g. zh-Hans → zh base) so regional variants are still added as
fallbacks. Original input is preserved as first priority.
- DownloadInfo.__init__ now explicitly initializes filename=None and
chapter_files=[] to avoid AttributeError on dynamic attribute access.
- __setstate__ adds filename=None migration for old persistent data.
- Frontend Download interface gains subtitle_files field; completed
downloads show each subtitle file with download link and CC icon.
P2 improvements:
- _LANGUAGE_REGION_VARIANTS expanded from 10 to 30 languages covering
it/ja/ko/hi/th/vi/id/pl/uk/ru/cs/sv/da/no/fi/tr/el/he/hu/bn/ta.
- Languages with no regional variants use empty tuple (they still get
-orig suffix when include_orig=True).
Update test expectations to match the new _subtitle_lang_list() output
which includes regional variants (e.g. fr-FR, en-GB) for better
subtitle matching.
DownloadInfo doesn't have a 'filename' attribute by default — it's only
set dynamically in update_status when a 'downloaded' status is received.
When captions are silently skipped by yt-dlp (no matching language), the
'downloaded' status is never emitted, so 'filename' doesn't exist on the
object at all. Using getattr(download.info, 'filename', None) instead of
direct attribute access prevents AttributeError.
Two fixes for captions downloads:
1. Language variant fallback: When user requests 'en' but the video only
has 'en-GB' (or other regional variants), yt-dlp silently skips
subtitle extraction because it uses exact language matching. Now the
subtitleslangs list includes common regional variants (e.g. en-US,
en-GB, en-AU, etc.) so yt-dlp can match the first available variant.
2. Error on no files: When yt-dlp completes successfully (ret=0) but
produces no subtitle files (e.g. no matching language at all), the
download was incorrectly shown as 'completed'. Now it's marked as
'error' with a descriptive message like 'No subtitles found for
language "en"'. Same treatment for thumbnail downloads.
- Reset download_phase when transitioning to paused (pause(), update_status(),
_post_download_cleanup()) so UI shows 'Paused' instead of stale phase label
- Reset download_phase when starting/resuming download (start() preparing state)
- Count 'postprocessing' as active download in updateMetrics() so dashboard
shows accurate counts during ffmpeg post-processing
- Add missing completedDownload Subject to DownloadsServiceStub in app.spec.ts
- Add Download type import to app.spec.ts
Fixes: TypeError Cannot read properties of undefined (reading 'pipe')
at app.ts:300 (completedDownload not in test stub)
- _download_phase_from_status now falls back to checking requested_formats
when top-level vcodec/acodec are both 'none' (common when yt-dlp uses
separate downloaders for video+audio streams)
- Clear download_phase on download completion/error (was persisting stale
phase value in finished state)
- Add debug logging in put_status for phase detection
- Update docker-compose.local.yml to match docker-compose.yml structure
with local image build (adds TARGETARCH build arg)
- Removed shadow from URL entry wrap in app.html for a cleaner look.
- Updated various icon files to improve visual consistency across the application.
- Add suffix to identify different files of same video
- Fixed invisible download for short link
Co-authored-by: Copilot <copilot@github.com>
The on_prepare handler unconditionally reflected the Origin request
header into Access-Control-Allow-Origin, and Socket.IO was configured
with cors_allowed_origins='*'. This allowed any website to make
authenticated cross-origin requests to all API endpoints, enabling
cross-origin download initiation, cookie overwrite, and data deletion.
Replace the blanket origin reflection with an explicit allowlist via
the CORS_ALLOWED_ORIGINS environment variable. When unset, cross-origin
requests are denied by default. Users who need cross-origin access can
set CORS_ALLOWED_ORIGINS to a comma-separated list of trusted origins.
The playlist/channel processing loop now sets playlist_count,
playlist_autonumber, n_entries, and __last_playlist_index on each
video entry so that templates like %(playlist_autonumber)s,
%(playlist_count)s, and %(playlist_index&{} - |)s resolve correctly
instead of showing NA.
Also updates _compact_persisted_entry to preserve n_entries and
__last_playlist_index across restarts.
Fixes#692
Agent-Logs-Url: https://github.com/alexta69/metube/sessions/b5aeb55a-3197-4a14-b8b4-96c9a67796e8
Co-authored-by: alexta69 <7450369+alexta69@users.noreply.github.com>
Replace the hand-rolled _outtmpl_substitute_field() / _compile_outtmpl_pattern()
with a new _resolve_outtmpl_fields() that delegates to yt-dlp's
YoutubeDL.evaluate_outtmpl(). This gives playlist/channel output templates
access to yt-dlp's full template syntax: defaults (%(field|fallback)s),
conditional formatting (%(field&prefix {})s), math (%(field+N)d),
datetime formatting (%(field>%Y-%m-%d)s), and more.
Only field references whose root name matches the targeted prefix (e.g.
"playlist" or "channel") are resolved; all other references remain as
template placeholders for yt-dlp to fill during the actual download.
Agent-Logs-Url: https://github.com/alexta69/metube/sessions/0ae5ff34-540f-4fc8-a81c-358fb92b7c15
Co-authored-by: alexta69 <7450369+alexta69@users.noreply.github.com>
Allow users to prefer a specific video codec (H.264, H.265, AV1, VP9)
when adding downloads. The selector filters available formats via
yt-dlp format strings, falling back to best available if the preferred
codec is not found. The completed downloads table now shows Quality
and Codec columns.
When adding a playlist, deleting a video from the queue causes it to be
re-added on the next loop iteration because queue.exists() returns False
for the deleted key. Track canceled URLs in a set so __add_entry skips
them.
Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>
