fix(stats): stop counting duplicate typeset subtitle lines (#191)

This commit is contained in:
2026-08-14 00:12:56 -07:00
committed by GitHub
parent 046e74ea91
commit b98d4d65c7
34 changed files with 2644 additions and 35 deletions
@@ -267,3 +267,123 @@ test('flushPlaybackPositionOnMediaPathClear ignores disconnected mpv time-pos re
assert.deepEqual(recorded, [42]);
});
test('media and subtitle-track transitions reset live subtitle-line deduplication', () => {
const recordedStarts: number[] = [];
const handlers = createBuildBindMpvMainEventHandlersMainDepsHandler({
appState: {
initialArgs: null,
overlayRuntimeInitialized: true,
mpvClient: null,
immersionTracker: {
recordSubtitleLine: (_text: string, start: number) => recordedStarts.push(start),
},
subtitleTimingTracker: null,
activeParsedSubtitleCues: null,
currentMediaPath: '/video-a.mkv',
currentSubText: '',
currentSubAssText: '',
playbackPaused: null,
previousSecondarySubVisibility: false,
},
getQuitOnDisconnectArmed: () => false,
scheduleQuitCheck: () => {},
quitApp: () => {},
reportJellyfinRemoteStopped: () => {},
syncOverlayMpvSubtitleSuppression: () => {},
maybeRunAnilistPostWatchUpdate: async () => {},
logSubtitleTimingError: () => {},
broadcastToOverlayWindows: () => {},
onSubtitleChange: () => {},
ensureImmersionTrackerInitialized: () => {},
updateCurrentMediaPath: () => {},
restoreMpvSubVisibility: () => {},
resetSubtitleSidebarEmbeddedLayout: () => {},
getCurrentAnilistMediaKey: () => null,
resetAnilistMediaTracking: () => {},
maybeProbeAnilistDuration: () => {},
ensureAnilistMediaGuess: () => {},
syncImmersionMediaState: () => {},
updateCurrentMediaTitle: () => {},
resetAnilistMediaGuessState: () => {},
reportJellyfinRemoteProgress: () => {},
updateSubtitleRenderMetrics: () => {},
refreshDiscordPresence: () => {},
})();
for (let index = 0; index < 8; index += 1) {
handlers.recordImmersionSubtitleLine('待って', index * 0.04, (index + 1) * 0.04);
}
assert.equal(recordedStarts.length, 4);
handlers.updateCurrentMediaPath('/video-b.mkv');
handlers.recordImmersionSubtitleLine('待って', 0.32, 0.36);
assert.equal(recordedStarts.length, 5);
for (let index = 9; index < 16; index += 1) {
handlers.recordImmersionSubtitleLine('待って', index * 0.04, (index + 1) * 0.04);
}
assert.equal(recordedStarts.length, 8);
assert.equal(typeof handlers.onSubtitleTrackChange, 'function');
handlers.onSubtitleTrackChange?.(2);
handlers.recordImmersionSubtitleLine('待って', 0.64, 0.68);
assert.equal(recordedStarts.length, 9);
});
test('subtitle-track transitions ignore stale parsed cues until replacement cues arrive', () => {
const recordedStarts: number[] = [];
const appState = {
initialArgs: null,
overlayRuntimeInitialized: true,
mpvClient: null,
immersionTracker: {
recordSubtitleLine: (_text: string, start: number) => recordedStarts.push(start),
},
subtitleTimingTracker: null,
activeParsedSubtitleCues: [{ startTime: 10, endTime: 14, text: '飛び上がる' }],
currentMediaPath: '/video-a.mkv',
currentSubText: '',
currentSubAssText: '',
playbackPaused: null,
previousSecondarySubVisibility: false,
};
const handlers = createBuildBindMpvMainEventHandlersMainDepsHandler({
appState,
getQuitOnDisconnectArmed: () => false,
scheduleQuitCheck: () => {},
quitApp: () => {},
reportJellyfinRemoteStopped: () => {},
syncOverlayMpvSubtitleSuppression: () => {},
maybeRunAnilistPostWatchUpdate: async () => {},
logSubtitleTimingError: () => {},
broadcastToOverlayWindows: () => {},
onSubtitleChange: () => {},
ensureImmersionTrackerInitialized: () => {},
updateCurrentMediaPath: () => {},
restoreMpvSubVisibility: () => {},
resetSubtitleSidebarEmbeddedLayout: () => {},
getCurrentAnilistMediaKey: () => null,
resetAnilistMediaTracking: () => {},
maybeProbeAnilistDuration: () => {},
ensureAnilistMediaGuess: () => {},
syncImmersionMediaState: () => {},
updateCurrentMediaTitle: () => {},
resetAnilistMediaGuessState: () => {},
reportJellyfinRemoteProgress: () => {},
updateSubtitleRenderMetrics: () => {},
refreshDiscordPresence: () => {},
})();
handlers.recordImmersionSubtitleLine('飛び上がる', 10, 10.04);
handlers.onSubtitleTrackChange?.(2);
for (let index = 1; index <= 8; index += 1) {
handlers.recordImmersionSubtitleLine('飛び上がる', 10 + index * 0.04, 10 + (index + 1) * 0.04);
}
assert.equal(recordedStarts.length, 5);
appState.activeParsedSubtitleCues = [{ startTime: 20, endTime: 24, text: '飛び上がる' }];
handlers.recordImmersionSubtitleLine('飛び上がる', 20, 20.04);
handlers.recordImmersionSubtitleLine('飛び上がる', 20.04, 20.08);
assert.deepEqual(recordedStarts.slice(-1), [20]);
});
+19 -5
View File
@@ -1,4 +1,5 @@
import type { MergedToken, SubtitleData } from '../../types';
import { createSubtitleLineDedupGate } from '../../core/services/subtitle-line-dedup-gate';
import type { MergedToken, SubtitleCue, SubtitleData } from '../../types';
type AnilistPostWatchRunOptions = {
watchedSeconds?: number;
@@ -34,6 +35,7 @@ export function createBuildBindMpvMainEventHandlersMainDepsHandler(deps: {
subtitleTimingTracker: {
recordSubtitle?: (text: string, start: number, end: number, secondaryText?: string) => void;
} | null;
activeParsedSubtitleCues?: SubtitleCue[] | null;
currentMediaPath?: string | null;
currentSubText: string;
currentSubAssText: string;
@@ -86,6 +88,11 @@ export function createBuildBindMpvMainEventHandlersMainDepsHandler(deps: {
deps.ensureImmersionTrackerInitialized();
deps.appState.immersionTracker?.recordPlaybackPosition?.(normalizedTimeSec);
};
// mpv reports every animation frame of a typeset line as its own subtitle event, so
// stats have to collapse bursts the same way the parsed cue list already does.
const immersionLineDedupGate = createSubtitleLineDedupGate({
getParsedCues: () => deps.appState.activeParsedSubtitleCues,
});
const hasInitialPlaybackQuitOnDisconnectArg = (): boolean =>
Boolean(
deps.appState.initialArgs?.managedPlayback ||
@@ -110,6 +117,9 @@ export function createBuildBindMpvMainEventHandlersMainDepsHandler(deps: {
if (!tracker?.recordSubtitleLine) {
return;
}
if (!immersionLineDedupGate.shouldRecord({ text, startSec: start, endSec: end })) {
return;
}
const secondaryText = deps.appState.mpvClient?.currentSecondarySubText || null;
const cachedTokens =
deps.appState.currentSubtitleData?.text === text
@@ -159,9 +169,10 @@ export function createBuildBindMpvMainEventHandlersMainDepsHandler(deps: {
logSubtitleProcessingDebug: deps.logSubtitleProcessingDebug
? (message: string) => deps.logSubtitleProcessingDebug!(message)
: undefined,
onSubtitleTrackChange: deps.onSubtitleTrackChange
? (sid: number | null) => deps.onSubtitleTrackChange!(sid)
: undefined,
onSubtitleTrackChange: (sid: number | null) => {
immersionLineDedupGate.reset();
deps.onSubtitleTrackChange?.(sid);
},
onSubtitleTrackListChange: deps.onSubtitleTrackListChange
? (trackList: unknown[] | null) => deps.onSubtitleTrackListChange!(trackList)
: undefined,
@@ -173,7 +184,10 @@ export function createBuildBindMpvMainEventHandlersMainDepsHandler(deps: {
deps.broadcastToOverlayWindows('subtitle-ass:set', text),
broadcastSecondarySubtitle: (text: string) =>
deps.broadcastToOverlayWindows('secondary-subtitle:set', text),
updateCurrentMediaPath: (path: string) => deps.updateCurrentMediaPath(path),
updateCurrentMediaPath: (path: string) => {
immersionLineDedupGate.reset();
deps.updateCurrentMediaPath(path);
},
restoreMpvSubVisibility: () => deps.restoreMpvSubVisibility(),
resetSubtitleSidebarEmbeddedLayout: () => deps.resetSubtitleSidebarEmbeddedLayout?.(),
getCurrentAnilistMediaKey: () => deps.getCurrentAnilistMediaKey(),
@@ -200,6 +200,59 @@ test('stats cli command fails when immersion tracking is disabled', async () =>
]);
});
test('stats cli command runs a duplicate-line cleanup preview without touching the dashboard', async () => {
const { handler, calls, responses } = makeHandler({
getImmersionTracker: () => ({
cleanupDuplicateSubtitleLines: async (options: {
dryRun?: boolean;
lookbackDays?: number | null;
}) => ({
dryRun: options.dryRun === true,
lookbackDays: options.lookbackDays ?? null,
scannedLines: 900,
burstGroups: 2,
removedLines: 180,
removedWordOccurrences: 540,
removedKanjiOccurrences: 120,
samples: [
{
videoId: 7,
videoTitle: 'Ep 1',
text: '飛び上がる',
frames: 90,
removedLines: 89,
startMs: 1000,
endMs: 5000,
},
],
}),
}),
});
await handler(
{
statsResponsePath: '/tmp/subminer-stats-response.json',
statsCleanup: true,
statsCleanupDuplicateLines: true,
statsCleanupDryRun: true,
statsCleanupLookbackDays: 30,
},
'initial',
);
assert.deepEqual(calls, [
'ensureImmersionTrackerStarted',
'info:Stats duplicate-line cleanup preview (last 30d): scanned=900 bursts=2 removedLines=180 removedWordCounts=540 removedKanjiCounts=120',
'info: Ep 1: "飛び上がる" x90',
]);
assert.deepEqual(responses, [
{
responsePath: '/tmp/subminer-stats-response.json',
payload: { ok: true },
},
]);
});
test('stats cli command runs vocab cleanup instead of opening dashboard when cleanup mode is requested', async () => {
const { handler, calls, responses } = makeHandler({
getImmersionTracker: () => ({
+30
View File
@@ -1,6 +1,7 @@
import fs from 'node:fs';
import path from 'node:path';
import type { CliArgs, CliCommandSource } from '../../cli/args';
import type { DuplicateSubtitleLineCleanupSummary } from '../../core/services/immersion-tracker/duplicate-line-cleanup';
import type {
LifetimeRebuildSummary,
VocabularyCleanupSummary,
@@ -50,6 +51,10 @@ export function createRunStatsCliCommandHandler(deps: {
ensureVocabularyCleanupTokenizerReady?: () => Promise<void> | void;
getImmersionTracker: () => {
cleanupVocabularyStats?: () => Promise<VocabularyCleanupSummary>;
cleanupDuplicateSubtitleLines?: (options: {
dryRun?: boolean;
lookbackDays?: number | null;
}) => Promise<DuplicateSubtitleLineCleanupSummary>;
rebuildLifetimeSummaries?: () => Promise<LifetimeRebuildSummary>;
} | null;
ensureStatsServerStarted: () => string;
@@ -83,6 +88,9 @@ export function createRunStatsCliCommandHandler(deps: {
| 'statsCleanup'
| 'statsCleanupVocab'
| 'statsCleanupLifetime'
| 'statsCleanupDuplicateLines'
| 'statsCleanupDryRun'
| 'statsCleanupLookbackDays'
>,
source: CliCommandSource,
): Promise<void> => {
@@ -126,6 +134,7 @@ export function createRunStatsCliCommandHandler(deps: {
const cleanupModes = [
args.statsCleanupVocab ? 'vocab' : null,
args.statsCleanupLifetime ? 'lifetime' : null,
args.statsCleanupDuplicateLines ? 'duplicate-lines' : null,
].filter(Boolean);
if (cleanupModes.length !== 1) {
throw new Error('Choose exactly one stats cleanup mode.');
@@ -142,6 +151,27 @@ export function createRunStatsCliCommandHandler(deps: {
writeResponseSafe(args.statsResponsePath, { ok: true });
return;
}
if (args.statsCleanupDuplicateLines && tracker.cleanupDuplicateSubtitleLines) {
const result = await tracker.cleanupDuplicateSubtitleLines({
dryRun: args.statsCleanupDryRun === true,
lookbackDays: args.statsCleanupLookbackDays ?? null,
});
const window =
result.lookbackDays === null ? 'all history' : `last ${result.lookbackDays}d`;
deps.logInfo(
`Stats duplicate-line cleanup ${result.dryRun ? 'preview' : 'complete'} (${window}): ` +
`scanned=${result.scannedLines} bursts=${result.burstGroups} ` +
`removedLines=${result.removedLines} removedWordCounts=${result.removedWordOccurrences} ` +
`removedKanjiCounts=${result.removedKanjiOccurrences}`,
);
for (const sample of result.samples.slice(0, 5)) {
deps.logInfo(
` ${sample.videoTitle ?? `video ${sample.videoId}`}: "${sample.text}" x${sample.frames}`,
);
}
writeResponseSafe(args.statsResponsePath, { ok: true });
return;
}
if (!args.statsCleanupLifetime || !tracker.rebuildLifetimeSummaries) {
throw new Error('Stats cleanup mode is not available.');
}