unidiff

Major modernization: Python 2 is dropped, the code is fully type-annotated, packaging moves to pyproject.toml, and several long-standing parser issues are fixed.

⚠️ Breaking

• Minimum supported Python is now 3.9 (Python 2 dropped).
• PatchedFile.path now also strips git mnemonic prefixes (c/ i/ o/ w/ and 1/ 2/), so git diff --cached paths resolve to the plain filename (c/README.md → README.md). Standard a//b/ diffs are unaffected. (#81, #27)

Highlights

• File modes via source_mode/target_mode + an is_symlink property. (#125)
• PatchedFile.diff_line_no to locate hunkless/binary entries. (#122)
• Accept bytes input directly (decoded with the given encoding, default UTF-8). (#43)
• Real PEP 484 type annotations + generics (typed iteration) and a py.typed marker. (#117)
• Parse difflib empty-filename output (#115) and fix a trailing-newline error on hunkless git format-patch files (#73, #74); includes the quoted-filename fix missing from 0.7.5 (#113).
• from_filename/from_string gain metadata_only; python -m unidiff works; packaging on pyproject.toml with GitHub Actions CI.

Thanks to @danwald, @cpina, @tovrstra and @huichen-cs.

Full changelog: v0.7.5...v1.0.0

• Fixed issue with spaces in filename when using custom git prefix.
• Support binary patch format.

• Fixed git diff parsing issues (filename with spaces, only one added/deleted file).

• Fixed RE_BINARY_DIFF regex to make it a raw string.

• Fixed issue when parsing git diff header generated with --no-prefix.

• Improved git added/deleted file detection.
• Added newline optional param when parsing from_filename.

• Fixed issues handling multiple git renames.
• Renamed files return target filename as PatchedFile.path.
• Fixed error when first change is a binary file.
• Added source code type hints.

• Updated PatchSet constructor to accept an optional (default to False)
  metadata_only parameter to only keep diff metadata information without
  the diff text data (better performance).
• Identify and track changed binary files.
• Added support for git rename syntax.

• Updated PatchSet constructor to accept string data.
• Added support to parse extended patch info.

• Added PatchSet.from_string helper.
• Do not install tests as top-level package.

• Re-released as 0.5.3 because of issues with PyPI.

• Added diff line number to Line metadata.
• Optimizations for large hunks.
• Fix for git empty new lines.
• Added (optional) errors parameter to PatchSet.from_filename, to specify how to handle encoding errors.

• Added (optional) encoding parameter to PatchSet.
• Added support to get any iterable as PatchSet diff argument.

• Release on PyPI.