sudacode/SubMiner
1
0
Fork 0
mirror of https://github.com/ksyasuda/SubMiner.git synced 2026-03-02 06:22:42 -08:00
Code Issues Packages Projects Releases Wiki Activity

chore(tokenizer): align enrichment regression notes and test typing

Browse Source
This commit is contained in:
Kyle Yasuda
2026-03-02 01:45:23 -08:00
parent fa97472bce
commit 629fe97ef7
2 changed files with 11 additions and 10 deletions
Download Patch File Download Diff File Expand all files Collapse all files

8
backlog/tasks/task-81 - Tokenization-performance-disable-yomitan-mecab-and-persistent-local-mecab.md
View File

@@ -4,7 +4,7 @@ title: 'Tokenization performance: disable Yomitan MeCab parser, gate local MeCab
status: Done
assignee: []
created_date: '2026-03-02 07:44'
updated_date: '2026-03-02 20:37'
updated_date: '2026-03-02 20:44'
labels: []
dependencies: []
priority: high
@@ -47,8 +47,10 @@ Implemented tokenizer latency optimizations:
- added regression coverage in `src/main/runtime/composers/mpv-runtime-composer.test.ts` for sequential tokenize calls (`warmup` side effects run once);
- post-review critical fix: treat Yomitan default-profile Anki server sync `no-change` as successful check, so `lastSyncedYomitanAnkiServer` is cached and expensive sync checks do not repeat on every subtitle line;
- added regression assertion in `src/core/services/tokenizer/yomitan-parser-runtime.test.ts` for `updated: false` path returning sync success;
- post-review performance fix: refactored POS enrichment to pre-index MeCab tokens by surface/start position and use sliding overlap window + binary-search cursor fallback, removing repeated full MeCab scans per token (`O(n*m)` hotspot);
- added regression test in `src/core/services/tokenizer/parser-enrichment-stage.test.ts` that fails on repeated distant-token scan access and passes with indexed lookup;
- post-review performance fix: refactored POS enrichment to pre-index MeCab tokens by surface plus character-position overlap index, replacing repeated active-candidate filtering/full-scan behavior with direct overlap candidate lookup per token;
- added regression tests in `src/core/services/tokenizer/parser-enrichment-stage.test.ts` for repeated distant-token scan access and repeated active-candidate filter scans; both fail on scan-based behavior and pass with indexed lookup;
- post-review startup fix: moved JLPT/frequency dictionary initialization from synchronous FS APIs to async `fs/promises` path inspection/read and cooperative chunked entry processing to reduce main-thread stall risk during cold start;
- post-review first-line latency fix: decoupled tokenization warmup gating so first `tokenizeSubtitle` only waits on Yomitan extension readiness, while MeCab check + dictionary prewarm continue in parallel background warmups;
- validated with targeted tests and `tsc --noEmit`.
<!-- SECTION:FINAL_SUMMARY:END -->

13
src/core/services/tokenizer/parser-enrichment-stage.test.ts
View File

@@ -101,11 +101,10 @@ test('enrichTokensWithMecabPos1 avoids repeated active-candidate filter scans',
let sentinelFilterCalls = 0;
const originalFilter = Array.prototype.filter;
Array.prototype.filter = function filterWithSentinelCheck<T>(
this: T[],
predicate: (value: T, index: number, array: T[]) => unknown,
thisArg?: unknown,
) {
Array.prototype.filter = (function filterWithSentinelCheck(
this: unknown[],
...args: any[]
): any[] {
const target = this as Array<{ surface?: string }>;
if (target.some((candidate) => candidate?.surface === 'SENTINEL')) {
sentinelFilterCalls += 1;
@@ -113,8 +112,8 @@ test('enrichTokensWithMecabPos1 avoids repeated active-candidate filter scans',
throw new Error('repeated active candidate filter scan detected');
}
}
return originalFilter.call(this, predicate, thisArg);
};
return (originalFilter as (...params: any[]) => any[]).apply(this, args);
}) as typeof Array.prototype.filter;
try {
const enriched = enrichTokensWithMecabPos1(tokens, mecabTokens);