feat: replace rglob with os.walk and per-language directory pruning

[KRRT7]

Summary

• Replace Path.rglob in get_files_for_language() with a single os.walk pass that prunes excluded directories in-place, avoiding traversal of .venv, node_modules, .git, etc.
• Add dir_excludes property to LanguageSupport protocol so each language declares its own exclusion patterns (exact names, prefix*, *suffix)
• Python excludes ~22 patterns (caches, build dirs, VCS); JS/TS excludes 9 (bundler output, framework caches)

Test plan

• pytest tests/test_function_discovery.py — 14 passed
• pytest tests/test_languages/test_function_discovery_integration.py — 14 passed
• ruff check clean on all changed files

[KRRT7]

codeflash should optimize this

[claude]

PR Review Summary

Prek Checks

✅ All checks pass (ruff check and ruff format). No fixes needed.

Mypy

53 errors found in the 4 changed files, but all are pre-existing — none are introduced by this PR. The new code (parse_dir_excludes, dir_excludes properties, os.walk refactoring) is type-clean.

Code Review

The refactoring from rglob to os.walk with per-language directory pruning looks correct. One existing issue remains unfixed:

⚠️ Test bug (from previous review, still open): Three test files pass Language as the 2nd positional argument to get_files_for_language(), where it gets interpreted as ignore_paths instead of language:

• tests/test_languages/test_javascript_e2e.py:85 — get_files_for_language(js_project_dir, Language.JAVASCRIPT)
• tests/test_languages/test_typescript_e2e.py:105 — get_files_for_language(ts_project_dir, Language.TYPESCRIPT)
• tests/test_languages/test_vitest_e2e.py:96 — get_files_for_language(vitest_project_dir, Language.TYPESCRIPT)

Since Language(str, Enum) is iterable, this silently iterates over characters and creates nonsense Path objects. Fix: use language= keyword arg.

Test Coverage

File	Stmts (main)	Stmts (PR)	Cover (main)	Cover (PR)	Δ
codeflash/discovery/functions_to_optimize.py	524	547	69%	70%	+1%
codeflash/languages/base.py	109	111	99%	99%	—
codeflash/languages/javascript/support.py	951	954	74%	74%	—
codeflash/languages/python/support.py	276	279	51%	52%	+1%
Total	1860	1891	71%	71%	—

✅ No coverage regression. Slight improvement in functions_to_optimize.py (+1%) and python/support.py (+1%).

Note: 8 pre-existing test failures in tests/test_tracer.py (unrelated to this PR).

---

Last updated: 2026-02-19