fix: prefer codeflash.toml over pyproject.toml for config loading

[aseembits93]

Summary

• parse_config_file() computed closest_toml_path (correctly preferring codeflash.toml at same depth) but never used it for config loading — always fell through to find_pyproject_toml() which hard-prioritizes pyproject.toml
• This caused codeflash.toml settings (e.g. Java module-root) to be silently ignored when a pyproject.toml exists in the same or parent directory
• Now uses closest_toml_path for loading, with explicit tie-breaking: codeflash.toml wins at same depth

Test plan

• Added 6 tests covering: same-dir priority, single-file fallbacks, nested directory resolution, explicit --config-file bypass
• Verified codeflash no longer reports "Functions outside module-root" for Java functions when both config files exist

🤖 Generated with Claude Code

[claude]

Claude finished @aseembits93's task in 3m 41s —— View job

---

PR Review Summary

Prek Checks

✅ Ruff (lint + format) — all passed. No auto-fixable issues.
ℹ️ Mypy errors in both changed files are pre-existing (not introduced by this PR).

Code Review

PR classification: SMALL (17 lines of production code in config_parser.py + 6 lines in git_utils.py)

codeflash/code_utils/config_parser.py — ✅ Correct fix

The core fix is correct. closest_toml_path was computed but then discarded — the function always fell through to find_pyproject_toml() regardless. The new fallback logic properly uses closest_toml_path:

if config_file_path is None and closest_toml_path is not None:
    config_file_path = closest_toml_path
else:
    config_file_path = find_pyproject_toml(config_file_path)

The explicit-path case (config_file_path not None, closest_toml_path None) still correctly delegates to find_pyproject_toml. The 6 tests cover the meaningful edge cases well.

codeflash/code_utils/git_utils.py — ⚠️ Unintended regression

The change on line 30 reverts a previous deliberate fix:

# Before (origin/main — intentional multi-language support):
supported_extensions = set(get_supported_extensions())

# After (this PR — accidentally reverts it):
supported_extensions = current_language_support().file_extensions

The commit message a016a5a7 undo irrelevant change suggests the author tried to undo this, but it remains in the diff. This regression means:

• When current_language is Python, Java/JS diffs won't be detected
• The test was renamed from test_java_diff_found_regardless_of_current_language → test_java_diff_ignored_when_language_is_python and the assertion changed from len(result) == 1 to len(result) == 0, explicitly testing the broken behavior

The original get_supported_extensions() approach was added specifically to handle multi-language monorepos. Reverting it here will silently break Java discovery in mixed-language repos.

Suggestion: Revert the git_utils.py change and its test updates. The config_parser.py fix is independent and should be merged without them. Fix this →

Duplicate Detection

No duplicates detected. The parse_config_file and get_git_diff changes are modifications to existing functions, not new logic introduced elsewhere.

Optimization PRs

PR #1877 (⚡️ Speed up JfrProfile.get_method_ranking) has merge conflicts and is less than 3 days old — leaving open to allow the base branch to stabilize.

---

Last updated: 2026-03-19T19:20:00Z