fix: 帧文件按帧号数值排序，修复超 9999 帧字典序错位

Browse Source

- frame-extract 新增 _sorted_frame_files：ffmpeg %04d 编号超过 9999 帧后扩为
  5 位，sorted() 字典序会把 5 位编号排在 4 位之前，导致 frames.json 时间与
  图像错位（真实发生于 run_339ec7ee437f 的 14236 帧任务）
- subtitle-ocr 提取 _merge_kept 供重组装复用
- 回归测试用真实任务留存数据（testdata/frames_boundary/），并新增真实 OCR
  数据按正确时间轴重组装为 SRT 的集成测试（test_integration_reassemble_ocr）

...

This commit is contained in:

cat-shark

2026-08-17 23:20:11 +08:00

parent f9ba96ef3a

commit 2b3a650612

17 changed files with 160 additions and 14 deletions

Show all changes Ignore whitespace when comparing lines Ignore changes in amount of whitespace Ignore changes in whitespace at EOL

Download Patch File Download Diff File Expand all files Collapse all files

testdata/frames_boundary/frame_0001.png

Vendored

BIN

Binary file not shown.

Side by Side After Width:  |  Height:  |  Size: 28 KiB