Repository navigation

0.9.0: extraction correctness pass — images, page tags, shapes, cells, spreadsheets, PDF quality - #20

Merged

CocoRoF merged 16 commits into

mainfrom

feat/merge-doc2chunk-improvements

Sep 14, 2026

Commits on Sep 12, 2026
- fix: repair the handler↔service contracts that silenced image & tag extraction
  
  CocoRoF
  and
  claude
  committed
  107c9c7
- feat(docx): read shapes, content controls and rich headers instead of skipping them
  
  CocoRoF
  and
  claude
  committed
  c7af3d5
- fix(hwpx): walk the section by containment instead of collecting every paragraph
  
  CocoRoF
  and
  claude
  committed
  c4afcd9
- fix(chunking): keep nested HTML tables in one protected region
  
  CocoRoF
  and
  claude
  committed
  85facdc
- fix(chunking): re-issue merged cells that span a table chunk boundary
  
  CocoRoF
  and
  claude
  committed
  1d141ad
- feat(excel): extract HTML exports served under a spreadsheet filename
  
  CocoRoF
  and
  claude
  committed
  ca3cf5b
- fix(excel): scan the whole sheet, and keep merged columns and small blocks intact
  
  CocoRoF
  and
  claude
  committed
  a4e401b
- perf(chunking): pack spreadsheet segments instead of one chunk per segment
  
  CocoRoF
  and
  claude
  committed
  32683a5
- fix(pdf): judge the text layer by what it contains, not only by whether it exists
  
  CocoRoF
  and
  claude
  committed
  dbe050e
- fix(charts): read the whole chart title, in every format that embeds one
  
  CocoRoF
  and
  claude
  committed
  eeb1de8
- fix(doc,ocr): drop Word field instructions from body text; convert each image once
  
  CocoRoF
  and
  claude
  committed
  69a0906
- feat(tables): read what a cell actually holds — pictures, sub-tables, controls
  
  CocoRoF
  and
  claude
  committed
  b3cb953
- feat: add extract_text_fast() for scanning, alongside extraction for reading
  
  CocoRoF
  and
  claude
  committed
  351d8df
- feat(ocr): add the DeepSeek-OCR engine; stop mistaking valid answers for failures
  
  CocoRoF
  and
  claude
  committed
  11fbcb4
- docs: 0.9.0 — changelog, README, version
  
  CocoRoF
  and
  claude
  committed
  70bd2d1
Commits on Sep 14, 2026
- style: ruff format — CI runs the latest ruff and checks formatting
  
  CocoRoF
  and
  claude
  committed
  14b01d6

Navigation Menu

Repository navigation

0.9.0: extraction correctness pass — images, page tags, shapes, cells, spreadsheets, PDF quality - #20

0.9.0: extraction correctness pass — images, page tags, shapes, cells, spreadsheets, PDF quality#20
CocoRoF merged 16 commits into
mainfrom
feat/merge-doc2chunk-improvements

Commits on Sep 12, 2026

fix: repair the handler↔service contracts that silenced image & tag extraction

feat(docx): read shapes, content controls and rich headers instead of skipping them

fix(hwpx): walk the section by containment instead of collecting every paragraph

fix(chunking): keep nested HTML tables in one protected region

fix(chunking): re-issue merged cells that span a table chunk boundary

feat(excel): extract HTML exports served under a spreadsheet filename

fix(excel): scan the whole sheet, and keep merged columns and small blocks intact

perf(chunking): pack spreadsheet segments instead of one chunk per segment

fix(pdf): judge the text layer by what it contains, not only by whether it exists

fix(charts): read the whole chart title, in every format that embeds one

fix(doc,ocr): drop Word field instructions from body text; convert each image once

feat(tables): read what a cell actually holds — pictures, sub-tables, controls

feat: add extract_text_fast() for scanning, alongside extraction for reading

feat(ocr): add the DeepSeek-OCR engine; stop mistaking valid answers for failures

docs: 0.9.0 — changelog, README, version

Commits on Sep 14, 2026

style: ruff format — CI runs the latest ruff and checks formatting