fix: match exclude_files against relative paths - #463

Actionable comments posted: 1

Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@cpplint.py`:
- Around line 270-273: Document in the exclude_files path-matching guidance in
cpplint.py (lines 270-273) that anchored directory patterns must be changed from
^bar$ to ^bar/. Add ("bar/baz.cc", "^bar$", True) to the relevant compatibility
test cases in cpplint_unittest.py (lines 7295-7302) to lock the behavior that
^bar$ does not exclude the bar subtree.

Re-reviewed current head db3dfc9. The implementation consistently matches exclude_files against the path relative to the defining CPPLINT.cfg, normalizes separators, preserves parent traversal and set noparent, and documents/tests the anchored-directory compatibility boundary (^bar$ vs ^bar/). The prior review finding about that boundary is addressed on this head. The PR reports 7 focused configuration tests and 238 full-suite tests passing, together with Ruff, Mypy, Pylint, Codespell, and git diff --check. LGTM.

Reviewed current head db3dfc9. The relative-path accumulation itself looks correct, including nested configs, parent traversal, set noparent, and the documented ^bar$ / ^bar/ behavior. The focused and full suites also pass.

I did find one case in the newly supported full-path matching that crashes instead of excluding the file. If a matched path contains %, the non-quiet exclusion message is later passed through % string formatting, so percent sequences that came from the filename can be interpreted as format directives.

For example, with a file at percent%2f/baz.cc and:

exclude_files=^percent%2f/baz[.]cc$

this PR matches the new relative path as intended, but then exits with TypeError: must be real number, not str while printing the exclusion notice. With --quiet it excludes correctly.

The formatting weakness already exists in the old message path, but this PR makes it reachable through the newly supported nested-path matching and also changes that diagnostic. I think it would be worth making the message formatting safe and adding a non-quiet %-path regression before merge.

Other than that, the relative-path behavior held up in my testing.

Re-reviewed current head d736a62. My previous %-path blocker is fixed.

I reproduced the original percent%2f/baz.cc failure on the previous head and confirmed it now exits cleanly while preserving the literal diagnostic. The new regression coverage exercises percent characters in the target path, config directory, and pattern in both quiet and non-quiet modes.

I also checked the focused config tests, full suite, pylint, mypy, pre-commit, and git diff --check; all passed locally. No remaining blocker from me on this head. LGTM.