kp-forks / grobid / 27904051704

39%

push

github

Improve matching in sentence segmentation re-alignment  (#1457)

- Rewrote `getSentenceOffsets` as a **forward two-pointer alignment**: consumes the original text strictly in order, whitespace-tolerant, cannot drift, and never drops text between sentences (replaces the window/`indexOf` reconstruction that lost runs of short reference-marker sentences on long paragraphs).
- Added **trailing-tail coverage**: text after the last returned chunk (e.g. a paragraph-final footnote/citation the segmenter doesn’t emit as a sentence) is covered instead of being dropped by TEIFormatter.
- Made `detect()` **thread-safe** (`synchronized` around the shared JRuby `put("text")` + segment) — fixes nondeterministic, large body/caption text loss under concurrent requests caused by the shared `ScriptingContainer` global. 
- Added `DiffMatchPatch` utility + diff-based `findInText` (used by the legacy/fallback path).

8637 of 24850 branches covered (34.76%)

Branch coverage included in aggregate %.

18387 of 45177 relevant lines covered (40.7%)

4.94 hits per line