GeezerVison

What's Changed

Fixes

• fix(lang): keep the languages a caller passes on short ASCII text by @chrikrah in #4601
• fix(pdf): extract vector-heavy PDFs with the fast strategy instead of returning nothing by @CyMule in #4540
• fix(cleaners): stop swapping en and em dashes in replace_unicode_quotes by @chrikrah in #4654

Other changes

• chore: start 0.27.37-dev after 0.27.36 release by @ryannikolaidis in #4634
• fix(pptx): keep visible text in translated, scaled and nested groups by @feiiiiii5 in #4619
• fix(metrics): read evaluation files as UTF-8 on every platform by @tanishqbaweja in #4623
• fix(deps): require filelock>=4.2.1 for GHSA-v4x8-3wr5-3wq8 by @tanishqbaweja in #4628
• fix(html): render each Title at its own heading level in elements_to_html by @tanishqbaweja in #4627
• fix(json): accept non-seekable streams while preserving legacy adapters by @Zhi-Xiang-Guo in #4649
• fix(docx): keep text that python-docx skips inside wrapper elements by @qued in #4661
• fix(documents): count the hash-id sequence number per page, not per contiguous run by @Eapp1e in #4621
• fix(coordinates): default CoordinateSystem.orientation to CARTESIAN by @Eapp1e in #4615
• chore: release 0.27.37 by @qued in #4668

New Contributors

• @Zhi-Xiang-Guo made their first contribution in #4649

Full Changelog: 0.27.36...0.27.37

Contributors

What's Changed

Other changes

• chore: start 0.27.36-dev after 0.27.35 release by @ryannikolaidis in #4607
• fix(xml): keep text that runs through child elements by @L4XB in #4573
• fix(pptx): filter off-slide shapes by how much of the box is visible by @feiiiiii5 in #4604
• fix(html_table): keep a declared rowspan of zero on empty cells by @Eapp1e in #4611
• fix(utils): detect temp files by path containment, not name prefix by @Eapp1e in #4609
• fix(ocr): expand an inferred list element into one element per item by @Eapp1e in #4617
• fix(pdf): keep FAST partitioning independent of ONNX load failures by @fallenmi in #4600
• fix: switch skip_infer_table_types default to None by @TrueFurina in #4464
• fix(filetype): detect the file type when the content type is application/octet-stream by @tanishqbaweja in #4625
• fix(partition): classify text as an address only when all of it is one by @Eapp1e in #4630
• fix: tolerate a dangling DOCX header/footer relationship instead of crashing by @aballman in #4624
• chore: release 0.27.36 by @aballman in #4631

New Contributors

• @Eapp1e made their first contribution in #4611
• @fallenmi made their first contribution in #4600
• @tanishqbaweja made their first contribution in #4625

Full Changelog: 0.27.35...0.27.36

Contributors

What's Changed

Other changes

• chore: replace per-PR changelog edits with generated release notes by @cragwolfe in #4578
• fix(utils): compare bounding boxes for elements with no page number by @simpleqt in #4561
• fix(chunking): measure combine thresholds in tokens when max_tokens is set by @joaquinhuigomez in #4528
• ci: guard release versions without blocking ordinary PRs by @cragwolfe in #4589
• fix(xml): honor explicit encodings when extracting text by @JackyHsu-X in #4572
• fix(msg): read the Cc, Bcc and Message-ID headers in any letter case by @L4XB in #4579
• fix(tsv): honor infer_table_structure when partitioning TSV files by @feiiiiii5 in #4513
• fix(msg): keep a comma in a quoted sender or recipient name by @RaphaelFakhri in #4570
• fix(xlsx): keep cell text like N/A and None instead of emptying it by @wenqiw777 in #4505
• feat: expose the DOCX paragraph style name in element metadata by @Ei-Ayw in #4591
• fix(text): keep a bulleted list that follows its lead-in line as list items by @L4XB in #4580
• fix(image): charge GIF frames at the canvas size Pillow decodes them at by @badGarnet in #4526
• fix(utils): recognize a box nested when the tolerance is zero by @SashaMIT in #4507
• perf(pdf): hi_res read every page with pdfminer just to learn whether the PDF has text by @CyMule in #4588
• fix(csv): read a line break in a quoted CSV or TSV cell as whitespace, not a literal \n by @CyberNativeAI in #4592
• ci: fail check-version when a PR edits the frozen CHANGELOG.md by @cragwolfe in #4594
• fix(metrics): let _mean ignore unmeasured scores like its siblings by @sclfcz in #4510
• fix(filetype): detect a .markdown file as Markdown, not plain text by @sinnercode228 in #4599
• fix(cleaners): translate each chunk in translate_text instead of the whole text by @sinnercode228 in #4568
• fix: render analysis bbox overlays at the pdf_image_dpi used for partitioning by @sinnercode228 in #4602
• fix(chunking): cap combine_text_under_n_chars at new_after_n_chars by @simpleqt in #4605
• fix(email): partition an attached email (message/rfc822) instead of raising AssertionError by @sinnercode228 in #4597
• ci: automate post-release development bump PRs by @cragwolfe in #4590
• fix(partition): recognize addresses in Alabama and Wyoming by @SergeevDmitry in #4581
• fix(md): return no elements for an empty text body by @feiiiiii5 in #4603
• chore: release 0.27.35 by @cragwolfe in #4606

New Contributors

• @simpleqt made their first contribution in #4561
• @joaquinhuigomez made their first contribution in #4528
• @JackyHsu-X made their first contribution in #4572
• @RaphaelFakhri made their first contribution in #4570
• @wenqiw777 made their first contribution in #4505
• @Ei-Ayw made their first contribution in #4591
• @SashaMIT made their first contribution in #4507
• @CyberNativeAI made their first contribution in #4592
• @sclfcz made their first contribution in #4510
• @sinnercode228 made their first contribution in #4599
• @SergeevDmitry made their first contribution in #4581

Full Changelog: 0.27.34...0.27.35

Contributors

What's Changed

• perf: Free decode buffers sooner when restoring original elements by @CyMule in #4524
• perf: Free image buffers sooner during figure extraction by @CyMule in #4523
• fix: honor encoding= in partition_json and partition_ndjson by @linhongyu510 in #4483
• fix(pptx): read a line break as a newline instead of a vertical tab by @L4XB in #4571
• fix(email): drop the empty entry of an address group by @chrikrah in #4576
• fix(csv): read CSV and TSV cells as the file holds them by @L4XB in #4574
• chore: bump Weaviate test image to 1.39.13 by @dudanogueira in #4575
• fix(partition): classify text as an email address only when all of it is one by @chrikrah in #4577
• chore: release 0.27.34 by @CyMule in #4582

New Contributors

• @L4XB made their first contribution in #4571

Full Changelog: 0.27.33...0.27.34

Contributors

What's Changed

• fix(chunking): don't repeat a very large split element in every chunk by @kholdrex in #4565
• fix(deps): bump certifi to 2026.10.19 by @lawrence-u10d in #4564

Full Changelog: 0.27.32...0.27.33

Contributors

What's Changed

• fix: label numbered headings without making them list items, and skip section-break paragraphs by @qued in #4538
• fix(partition): read spooled uploads in place instead of copying them by @CyMule in #4419
• perf(pdf): bound OCR upload and rendered-page memory by @CyMule in #4516
• perf(chunking): bound image payload and original-element serialization memory by @CyMule in #4517
• feat(xlsx): expose a guarded worksheet-at-a-time workbook reader by @CyMule in #4562
• perf(xlsx): reduce dense-worksheet partitioning memory by @CyMule in #4563

Full Changelog: 0.27.30...0.27.32

Contributors

What's Changed

• fix(email): keep non-ASCII sender and recipient names readable by @usmanmateen in #4543

Full Changelog: 0.27.29...0.27.30

Contributors

What's Changed

• fix(xlsx): read .xls files whose Workbook stream ends in padding by @usmanmateen in #4545

New Contributors

• @usmanmateen made their first contribution in #4545

Full Changelog: 0.27.28...0.27.29

Contributors

What's Changed

• feat: opt-in langdetect-rs backend for partition language detection by @kholdrex in #4547
• fix(html): keep text inside unrecognized elements by @huuufu in #4497
• fix(html): keep table captions in Table text and text_as_html by @Sriram-PR in #4553
• perf(docx): resolve each paragraph style once per document (5.3 s -> 1.0 s outside spaCy) by @deadstrobe5 in #4554
• fix(xlsx): collapse empty-row runs instead of rejecting far-away cells by @palaniappa in #4525
• fix(ocr): include underlying error when loading OCR agent fails by @Anurag-M1 in #4491

New Contributors

• @kholdrex made their first contribution in #4547
• @Sriram-PR made their first contribution in #4553
• @deadstrobe5 made their first contribution in #4554
• @palaniappa made their first contribution in #4525
• @Anurag-M1 made their first contribution in #4491

Full Changelog: 0.27.27...0.27.28

Contributors

What's Changed

• fix(json): preserve rehydrated element identity by @rohan-patnaik in #4504
• fix: keep overlap category names from element 100 on by @jipeng6036-del in #4529
• fix: normalize None text to empty string in Text elements by @TrueFurina in #4463
• fix: stop silently returning mojibake from encoding detection by @huuufu in #4550
• fix: partition_email raises KeyError on a multipart/* attachment by @mittalpk in #4424
• fix: preserve MathML content in HTML partitioning by @fireblade2534 in #4478

New Contributors

• @rohan-patnaik made their first contribution in #4504
• @jipeng6036-del made their first contribution in #4529
• @TrueFurina made their first contribution in #4463
• @mittalpk made their first contribution in #4424
• @fireblade2534 made their first contribution in #4478

Full Changelog: 0.27.25...0.27.27

Contributors