What microsoft/markitdown shipped
Written by FoxPlug from the repository's own releases; not affiliated with Microsoft. Paste your own repository in the box below and FoxPlug writes the same weekly update for your product. This page is an automatic summary of the releases, pull requests and commits in github.com/microsoft/markitdown; Microsoft did not write it and does not use or endorse FoxPlug. Every line links to the change it describes.
Get a weekly update like this for your product, free
Week of September 28, 2026
What shipped
- Consolidated format-specific regressions from 35 test files into 18 files with one suite per format, documented in tests/README.md. Pull request #2579
- Restored JSON regression coverage for late non-ASCII text detection by adding
json_late_non_ascii.jsontest fixture. Pull request #2601 - OCR plugin registration tests now discover and load real installed OCR packages through the core plugin loader instead of mocked entry points. Pull request #2588
- Conversion regressions now exercise real PDF parsing, PowerPoint charts, and HTTP responses instead of local parser mocks. Pull request #2586
- PowerPoint missing-text tests now convert real saved presentations and verify surrounding content survives through library and CLI vectors. Pull request #2594
- CLI file-type parsing and formatting tests now run production code paths with valid, normalized, single-type, and invalid inputs. Pull request #2592
- MCP input error tests replaced mocked failures with real conversions using unsupported gopher URIs and binary fixtures. Pull request #2597
- WikipediaConverter fixed to handle italicised titles by reading from h1#firstHeading when mw-page-title-main span is absent. Pull request #2458
- HTTP response reading performance improved by increasing chunk size from 512 bytes to 64 KB. Pull request #1511
Why it matters
This week focused on strengthening test coverage by replacing mocks with real conversions and fixtures, ensuring regressions catch actual failures. The consolidation of 35 test files into 18 improves maintainability. Performance improvements and bug fixes for Wikipedia and CLI output handling make the project more robust.
Changelog entry
- Consolidated format-specific regression tests from 35 files into 18 organized files with shared CLI/library vector suites Pull request #2579
- Restored JSON charset detection coverage for late non-ASCII text Pull request #2601
- Fixed WikipediaConverter to handle italicized titles on Wikipedia Pull request #2458
- OCR plugin tests now use real installed plugin discovery instead of mocks Pull request #2588
- Conversion regression tests now exercise real PDF, PowerPoint, and HTTP parsing Pull request #2586
- Optimized OMML direct tag lookup using frozenset instead of tuple Pull request #1919
- Increased HTTP chunk read size to 64 KB for better performance Pull request #1511
- WikipediaConverter now accepts mobile Wikipedia URLs (m.wikipedia.org) Pull request #1850
- Fixed CLI output handling when sys.stdout.encoding is None Pull request #1656
- Optimized non-seekable stream buffering using shutil.copyfileobj Pull request #1510
- Added language parameter for audio transcription Pull request #1295
Test week: Consolidated 35 regression files to 18, replaced mocks with real conversions, fixed Wikipedia italicized titles, improved OCR discovery and performance.
This week we strengthened our test suite by consolidating 35 format-specific regression test files into 18 organized files with clear documentation. We replaced mocked dependencies with real conversions across PDF, PowerPoint, OCR, and MCP handling. Fixed Wikipedia title parsing for italicized content, improved OMML math conversion efficiency, and optimized HTTP chunk reading from 512 bytes to 64 KB.