[xgbtxy]

Fixes #159717.

Thanks to @digit50 for the original report and reproducer.

glob.translate() currently discards the star positions returned by fnmatch._translate(). As a result, short patterns such as '*a' * 8 + '*b' repeatedly backtrack in PurePath.match(), PurePath.full_match() and Path.glob().

Reuse the atomic-star joining logic for each ordinary glob segment, with a separator-excluding star. Keep the existing fast path for segments with fewer than two stars, and leave recursive ** and hidden-name handling in glob. fnmatch.translate() retains its existing public regex output.

A fixed part that can consume a separator must remain backtrackable: for example, *[!x]*x currently matches a/x. The separator check also covers case-sensitive and case-insensitive uses with custom separators. This patch preserves that behavior instead of changing character-class semantics as a side effect.

Validation on Windows x64, a Debug build of main at 4c73f50ca9885e5850dae64463e87cf5e80992a1:

• test_fnmatch, test_glob, test_pathlib, test_shutil, test_tarfile, test_zipfile, test_unittest: all seven modules passed; regrtest reported run=4,202 skipped=740.
• After restoring the simple-segment fast path, the three directly affected modules passed again (run=1,471 skipped=533). Skips include platform/feature-dependent tests; this is not a Linux validation run.
• A deterministic comparison with the original translator, including hidden paths, recursion, character classes, custom separators and case-insensitive matching, found no matching differences across the sampled cases.
• Tools/patchcheck/patchcheck.py and git diff --check pass; a NEWS entry is included.

For the issue's pattern against 36 a characters, the same Debug interpreter measured full_match at about 631 ms before and 0.096 ms after, and a single-file Path.glob at about 1706 ms before and 0.342 ms after. These are warmed-up medians of three calls for this reproducer, not a general performance claim.

This does not promise to remove all backtracking involving recursive patterns or character classes that consume separators. Cold translation of multi-star patterns has some additional cost; the change targets excessive matching cost and retains the common zero/one-star translation path.

Prepared with OpenAI Codex assistance for implementation, test generation and review. Builds, tests and measurements were executed locally.

• Issue: pathlib match, full_match and glob take exponential time on short patterns because glob.translate skips fnmatch's atomic-group rewrite #159717

[python-cla-bot]

All commit authors signed the Contributor License Agreement.

[Copilot]

Copilot was unable to review this pull request because the user who requested the review has reached their quota limit.

[StanFromIreland]

Sorry but I'm closing, the issue author asked to open the PR, please read the issue before opening.