- Cap only blind shrinking-window retries (not backend-guided shrinks); a line that exhausts the cap escalates to one parseText fallback instead of dropping to raw text
- Bound the cross-line termsFind cache by retained dictionary-entry weight, not just key count, and re-check it when a lookup resolves
- Keep halfwidth katakana character names in the greedy pre-pass, and let a generic word beat a name it fully contains
- Share one Han code-point table between the character dictionary and the scanner's name pre-pass; narrow the mob-disambiguator filter to the split letters, not every one-character term
- Release the subtitle prefetch pause on a new onProcessingSettled signal instead of the tokenized emit, so duplicate/suppressed/failed lines no longer pause prefetch indefinitely
- Split the Yomitan scan runtime's injected helper script into its own file
- AniList mob names like 女子A/"Joshi A" romanize their trailing
disambiguator letter into a single kana term (A → ア) that collides
with common interjections (あ〜); filter these out unless kanji
- bump CHARACTER_DICTIONARY_FORMAT_VERSION to 20 to invalidate cached
dictionaries built with the old term set
- add tests covering the dropped disambiguator and a kept single-kanji
name part
{"subject": "test(subtitles): fix tests for provisional emit and payload signature", "body": "- Expect the provisional plain-text emit alongside the refresh's empty emit\n- Match emitSubtitlePayload regex against its updated parameter signature"}
Autoplay priming called onSubtitleChange after a cache miss, which only asks
whether the text is new. When the miss came from an invalidation (mining a
card while the line is on screen) the text was unchanged, so nothing was
scheduled and the line stayed unannotated for as long as it was displayed.
refreshCurrentSubtitle checks the cache generation as well, so it re-tokenizes
for the new generation; the resume fallback is kept for the case where it
genuinely has nothing to do.
refreshCurrentSubtitle also returned false for empty text while a run was in
flight, even though that run goes on to emit the empty subtitle. It now
reports the pending emit so callers do not release the prefetch pause early.
The priming tests now drive the real subtitle processing controller instead of
a stub. The previous stub encoded the wrong assumption about unchanged text
and so could not catch either bug.
The repeated-subtitle pause leak was only fixed for ordinary subtitle
changes. Startup autoplay priming and visible-overlay priming pause the same
way and also ignored whether any tokenization was scheduled, so a cache miss
on text the controller already holds (mining a card while the line is on
screen) left prefetching idle for the rest of the cue.
Pause and release are now one operation via pausePrefetchUntilEmit, and both
controller entry points report whether an emit is expected. A repeat arriving
while a run is already in flight keeps the pause, since that run still emits.
Also invalidate the character dictionary lookups centrally from the sync
completion handler instead of at three manager call sites. Ordinary selection
sync never invalidated them, so a stale non-null name candidate list could
skip a newly added name for up to five seconds; a missing list falls back to
the exhaustive scan, but a stale one does not. Ordering matters: the
invalidation runs before the subtitle refreshes so they re-tokenize against
the new dictionary content.
Docs: subtitle-overlay-priming no longer claims every subtitle change calls
onSeek().
onSubtitleChange paused prefetching unconditionally, but the processing
controller returns early when the text matches what it already has. Nothing
is tokenized, so nothing is emitted, so the resume that rides on the emit
never fires and prefetching idles for the rest of the cue. The reachable
trigger is a repeat arriving after a cache invalidation, such as mining a card
while the same line is still on screen.
The controller now reports whether it scheduled processing and the caller
resumes when it did not, so every pause has a matching resume.
Also harden the character-name candidate prefilter: Yomitan collapses emphatic
sequences before matching, so a stretched spelling still resolves to its entry
(ミナァァト matches ミナト). The candidate match now skips small kana and
prolonged marks, which only widens the probe set and so cannot drop a name.
The greedy name pre-pass asked the Yomitan backend at every Japanese
position, because a character name can begin mid-token. With the character
dictionary enabled that roughly doubled the round trips per line (measured
10 -> 21 on a 23-char line).
SubMiner generates the character dictionary, so the cached snapshots already
list every form a character entry can be matched by (term and reading). Those
forms are installed into the scan runtime once per media and the pre-pass now
probes only positions where one of them starts, compared after kana
normalization so a katakana name still matches a hiragana reading form. The
overhead drops to zero (21 -> 10, the same as with the dictionary disabled).
Fail-safe: with no candidate list (no media id, no cached snapshot, failed
install) the pre-pass keeps its exhaustive behavior, so stale character data
costs speed rather than a missing name. Halfwidth katakana positions bypass
the filter since kana normalization does not fold them.
The candidate lookup is consulted per subtitle line, so it caches its snapshot
directory signature for 5s; dictionary writes still call invalidate().
- extract the injected scan runtime (helpers, install script, call-script
builder) into tokenizer/yomitan-scan-runtime-script.ts; the host module drops
from ~2700 to ~1900 lines
- append the kana run to the reading as well as the surface when an unparsed
run extends the previous token, so the reading keeps covering the surface and
the known-word reading fallback stays enabled (bumps scan runtime version)
- stop annotateMs before character-image resolution so the stage timing
measures the annotation stage only
- drop the duplicate parseText full parse per line; the termsFind scanner walk
is now authoritative and emits its own unparsed filler runs (parseText kept
only as error fallback)
- install scan helpers once per parser window (__subminerYomitanScan) instead
of re-shipping ~500 lines of script per subtitle line
- persist termsFind results across lines in a window-scoped LRU keyed by
substring, invalidated via a cache epoch on dictionary/settings changes
- skip lookups at punctuation/whitespace positions and cap the shrinking-window
retry ladder at 4 lookups per position
- build tokenizer runtime deps once (JLPT lookup cache never hit before; mecab
availability check ran per line)
- stop restarting the prefetch run on every subtitle change; resume prefetch
only after the tokenized payload lands, not on provisional raw emits
- add per-stage debug timings (scanMs/mecabMs/frequencyMs/annotateMs)
- Add overlayBody override to ConfiguredStatusNotificationOptions so overlay/OSD/desktop can diverge
- Extract getSubsyncStatusNotificationOptions() to strip the ASCII spinner frame from the overlay card (OSD keeps it since it renders the raw spinner)
- Add tests for spinner stripping and subsync result notifications
The query test suite opened each temp database with SQLite's defaults
(rollback journal, synchronous=FULL) while the runtime opens them via
applyPragmas (WAL, synchronous=NORMAL). Every statement therefore ran as
its own fsync-ing transaction.
That is unnoticeable on a fast local disk but dominates on CI: the file
took 40.6s there versus 0.5s locally, and the 106-insert case
'getVocabularyStats pages past hidden rows' crossed the 5s per-test
timeout and failed the quality gate.
Open test databases through a helper that applies the same pragmas as
production, so the tests exercise the runtime's actual SQLite
configuration instead of a slower one.
- Add TsukiHime and AniSkip to feature table; note xz dependency
- Clarify Anki+AnkiConnect is required to mine, not to run
- Rework Windows setup with winget/scoop commands and PATH steps
- Note ffmpeg has no path-override setting, unlike mpv
- Fix screenshot alt text and installation.md anchor link
- Generate config.example.jsonc with a Linux socket path instead of Windows, so the example stays reproducible across platforms
- Document TsukiHime config (`tsukihime.*`), maturity-based known-word highlighting keys, and other config surfaces missing from the reference
- Correct secondary-subtitle auto-load defaults and Anki field-matching (case-insensitive) claims
- Expand Windows installation guide with winget/Scoop package tables and manual PATH setup
- Add docs-site/links.test.ts to catch broken internal links and heading-anchor mismatches
- Fix stale cross-references in jimaku-integration.md and demos.md
- Add formatRofiPrompt to trim trailing whitespace and append a single space, keeping the prompt from running into the input field
- Apply it across all rofi -p usages in picker.ts and history-command.ts
- Add tests for formatRofiPrompt edge cases (empty/whitespace-only prompts)
- Extract buildHistoryEntryActions to build the series action menu, now including a previous-episode option alongside replay/next/browse/quit
- Update docs and changelog entry to describe the new option
- Add tests covering previous/replay/next ordering and omission when last watched file is missing
- AnkiConnect proxy now detects EADDRINUSE, logs a warning, and surfaces an overlay notification instead of crashing video startup
- Background AppImage bootstrap runs Chromium with in-process-gpu so no GPU child survives app.exit() and outlives the FUSE mount, fixing the remaining DrKonqi "Service Crash" case
- Add changelog fragment for the proxy fix; update the AppImage quit fragment with the GPU child root cause
- Merge the five separate sync change fragments (launcher command, UI window, close/quit fixes, active-session double-count, monthly rollup timezone) into a single changes/sync.md
- Fold kana-pos-content-classification.md into annotation-noise-rule-table.md
- Drop the now-stale Unreleased sections from CHANGELOG.md and docs-site/changelog.md
- Tweak tsukihime-english-subtitles.md wording
- Autoplay-ready gate now ignores untokenized subtitle payloads while tokenization warmup is pending, so playback doesn't resume before the first tokenized cue
- Wire isTokenizationReady dep through main.ts and update gate/wiring tests accordingly
- Drop stale release/release-notes.md, add changelog entry under changes/
Background AppImage launches executed directly from a FUSE mount owned by a
short-lived helper's AppImage runtime. On quit, the runtime unmounted the
squashfs while utility children (network service) were still mid-shutdown,
SIGBUSing them and triggering KDE Service Crash notifications on every video
close. Detach a POSIX-sh supervisor instead: mount via --appimage-mount, run
AppRun from the mount, and release the holder only once no process still
executes from the mount. Falls back to the old direct launch on any failure
or with SUBMINER_NO_APPIMAGE_MOUNT_KEEPALIVE=1.
- Merge related fix entries into single files (name-match-span-merge into character-name-split-and-scan-recovery, kanji-non-independent-noun-annotations + lexicalized-expression-frequency into content-adverb-annotation-stop-list, greedy-name-parsing-priority into character-name-split-and-scan-recovery, kiku-field-grouping-cancel-double-toast into kiku-field-grouping-modal-recovery, history-rofi-cover-art into launcher-history-command)
- Rewrite remaining entries for brevity and reclassify stats-trend-title-limits as "changed"
- Add allowReadingOnlyMatch:false to kanji token known-word lookups so 渓谷/けいこく no longer matches a mined 警告/けいこく card via reading-only index
- Greedy name pre-pass yields when a strictly longer generic word starts at the same position (空 no longer splits 空気; ties still go to the name)
- 確かに and やはり no longer excluded from frequency/JLPT highlighting and vocab stats
- Stop-word list now covers only interjections, pronouns, and grammar fragments
- Drop changes/kiku-field-grouping-modal-wiring.md (landed in #138)
- Drop changes/known-word-partial-reading-highlight.md (landed in #142)
- Remove assets/minecard.mkv demo video
sudacode