GitoxideLabs/gitoxide gix-pathspec-v0.20.0

gix-pathspec v0.20.0

on GitHub

Chore

• Stabilize fixtures on Windows

  • Normalize the precomputed diff fixture assets to LF before writing blobs,
    populating the index, and creating commits. Git for Windows may check these
    assets out with CRLF, which changes their object IDs and adds carriage returns
    to index paths, causing fixture setup to fail at git mv cli c.

  • Prevent Git Bash from rewriting revision arguments before Git sees them, and
    normalize the two pathspec baseline cases where Git for Windows applies native
    path validation to repository-format paths. Exclude a glob baseline whose
    backslash behavior is specific to Git for Windows rather than Git paths.

  • Generate pathological .gitmodules entries as configuration data instead
    of trying to create module directories whose names cannot be represented
    on Windows. Use Git Bash bundled Perl for binary fixture construction so
    regeneration does not depend on a separately installed Python interpreter.

  • Also pass a literal carriage return to sed through Bash ANSI-C quoting when
    normalizing the jj diff assets. Unlike GNU sed, BSD sed does not interpret
    backslash-r in a single-quoted expression, so the previous spelling could remove
    a trailing letter r on macOS instead of stripping CRLF endings.

Bug Fixes

• match Git's pathspec attribute syntax

  gix-pathspec reused the attribute-file tokenizer even though Git's pathspec attr
  grammar separates requirements only on spaces and rejects values attached to !/-
  requirements. This made tabs and carriage returns into unintended separators and
  accepted !a=b and -a=b.

  Parse pathspec attribute requirements according to their own grammar while still
  allowing the builtin_* namespace for pathspec matching.

  Git baseline: git 2.50.1 and /Users/byron/dev/github.com/git/git at cf5497b14c,
  notably pathspec.c parse_pathspec_attr_match().

• ignore empty keywords in pathspec magic, like Git does
  Git's parse_long_magic() in pathspec.c skips zero-length keywords
  via if (!len) continue;, so forms like :(top,), :(,top),
  :(top,,icase), :(,), :(,,), :(icase,) and :(attr:someAttr,)
  are all valid pathspecs to Git. gitoxide's parser rejected them with
  InvalidKeyword instead.

  Two changes fix the divergence:

  • parse_long_keywords's closure now returns early with Ok(()) when
    the keyword is empty, mirroring Git's skip.
  • split_on_non_escaped_char is rewritten to scan byte-by-byte like
    Git's strcspn_escaped(): a backslash consumes the following byte,
    and a separator at index 0 is now visible. The previous windows(2)
    loop could never see a separator at index 0, which is why :(,top)
    was reported as a single invalid keyword instead of an empty one
    followed by top.

  Verified against real git 2.52.0 with a 32-input differential battery
  (32/32 agree), plus new coverage in tests/parse/valid.rs and
  tests/fixtures/parse_baseline.sh.

Commit Statistics

• 11 commits contributed to the release over the course of 30 calendar days.
• 30 days passed between releases.
• 3 commits were understood as conventional.
• 0 issues like '(#ID)' were seen in commit messages

Commit Details