github docling-project/docling v2.150.1

38 minutes ago

Fix

  • cli: Omit from_formats in convert-remote unless --from is set (#5346) (ebc50e3)
  • docx: Detect headings from paragraph-level w:outlineLvl (#3312) (#4129) (31d14a7)
  • docx: Keep hyperlink targets in footnotes and endnotes (#5322) (0c9a1cd)
  • docx: Keep a single-cell table nested in a table cell inside the cell (#5258) (b9bf356)
  • docx: Close an open list before a table that follows it (#5309) (fec4dc3)
  • docx: Use default properties when an OMML element has none (#5325) (8e210fb)
  • Scale 16-bit images to 8 bits before converting them to RGB (#5248) (293f620)
  • ods: Compute sheet page size as bounding-box extent across all content layers (#5300) (ca9a4e0)
  • jats: Keep the section title next to its label in headings (#5244) (b4cdb91)
  • docx: Keep the fragment of an external hyperlink (#5249) (eb97111)
  • md: Parent the image caption with its picture (#5272) (9b3ebca)
  • jats: Keep ext-link hyperlinks in fn-group footnotes (#5290) (da1eda8)
  • jats: Keep the text of and blocks (#5283) (328b69d)
  • asciidoc: Read the table cols attribute (#5259) (e2f556e)
  • epub: Drop query and fragment from image src (#4636) (1ae470a)
  • epub: Rewrite links and image sources in start tags only, not in text (#5227) (a7738e1)
  • docx: Keep w:sym characters and map Symbol font text to Unicode (#4622) (c5226aa)
  • docx: Keep the text of wrapped runs in plain table cells (#4455) (c59ef9d)
  • jats: Preserve author footnotes (#5216) (fe35625)
  • html: Make unclosed

    fix-up independent of hash seed (#5048) (f14c72b)

  • docx: Read table rows and cells wrapped in customXml or sdt (#4393) (fa83e59)
  • csv: Read Excel's SEP= hint in any letter case (#5261) (bc1a812)
  • images: Complete JATS file URI figure loading (#5017) (d642657)
  • uspto: Map italic h to the Planck constant sign (#5233) (c60fa22)

Documentation

Performance

  • docx: Skip run formatting when reading the text of plain cells (#5279) (336e9dd)
  • md: Build the link TypeAdapter once (#5241) (a7760ab)
  • pptx: Compute each text shape's provenance box once, not per paragraph (2x to 4x on text-heavy decks) (#5262) (d79a727)

Don't miss a new docling release

NewReleases is sending notifications on new releases.