Cover the subtitle path in arr-e2e #205
No Assignees
Notifications
Due Date
No due date set.
Depends on
Reference: yolo/arr#205
Reference in New Issue
Block a user
Cover the subtitle path end to end in
arr-e2e, cross-process like the rest ofthat crate.
At least: a media file with no subtitles gets one fetched, synced and written
with the right name; a file whose embedded English text track is the only
source gets it extracted and translated into a
.mtsidecar; a file whosesubtitle sync is implausible keeps the unsynced original and shows as
sync-rejected; a language once satisfied is never re-searched.
Providers and translators are stubbed at the HTTP boundary and the
remote-command backend by a stub script — the rule that tests never touch a
live tracker (
CLAUDE.md) applies to subtitle providers exactly the same way.ffmpegandalassare real, astmuxandgitare real in this repo'sother integration tests.
Depends on: #195, #196
Implemented three cross-process
arr-e2etests covering the subtitle path (DESIGN.md §15), against the realarrbinary, realffmpegand realalass, with Podnapisi and the remote-command translation backend stubbed at the HTTP/process boundary:a_missing_subtitle_is_fetched_synced_and_named— a media file with no subtitles gets a pt-PT candidate fetched from Podnapisi, synced by realalass, written as<name>.pt-PT.srt, and confirms the provider is never re-searched once satisfied (waits past a second reconcile tick).an_embedded_track_is_extracted_and_translated— a file whose only English source is an embeddedsubriptrack gets it extracted by realffmpegand translated (via a stubtranslate-commandscript) into<name>.pt-PT.mt.srt.an_implausible_sync_keeps_the_original_and_is_flagged— a fetched subtitle timed five minutes off a three-second clip is rejected by realalass(300s shift), the unsynced original is kept verbatim, and it surfaces in/api/queues/subtitlesassync_rejected.There is no HTTP route to adopt an already-imported file (grab/download/import is a separate seam), so each test seeds a movie, an imported grab and a
media_filesrow directly against the daemon's own SQLite file — the harness now exposesDaemon::database_path()andDaemon::media_root()for this, plusDaemon::spawn_with_envfor the extra child env vars the scenarios need.Two small seams were added because there was otherwise no way to write this test at all:
ARR_PODNAPISI_URL, an env-only seam mirroring the existingtmdb_urlone, so Podnapisi can point at awiremockfake.arr-e2eharness now builds the daemon with--features translate-commandso a stub script can stand in for a translation backend.Branch:
subtitles/205-subs-e2e.just cipassed locally (686 tests). The newarr-e2etests are outside that gate per the repo's own convention (.gitea/workflows/e2e.yml) and were run manually against a locally installedalass-cli(symlinked asalass) — all three pass.Findings
Syncer::runincrates/arr-subs/src/sync.rspassedalassits arguments as[subtitle, video, output]. The realalassbinary's signature is<reference-file> <incorrect-sub-file> <output-file-path>, where the reference must be the video. With the old order, every realalassinvocation failed immediately (unknown subtitle format for file <video>), andSyncer::settledegraded this toSyncState::NotRunsilently — meaning every subtitle sync in production was silently a no-op since #194 landed. Fixed the argument order and the one unit test whose fake script depended on it; allarr-substests still pass..gitea/workflows/e2e.ymlnever installedalass(onlyffmpeg), and never built the daemon with anytranslate-*feature, so the e2e suite could never have exercised the sync or translation paths for real. Added an install step (cargo install alass-cli, symlinked toalass) and--features translate-commandto the daemon build step. Also addedcrates/arr-subs/**to the workflow's trigger paths — it wasn't there even though the daemon's subtitle behavior depends on it directly.media_files/grabs/releasesrows by hand. Worth a dedicated test-support seam if more of this kind of test gets written.