fix: set language singleton early so git diff auto-detection works for non-Python languages

[mashraf-222]

Summary

Running codeflash --subagent without --file (auto-detect mode) always found 0 functions for Java/JS/TS projects because the language singleton defaulted to Python when get_git_diff() filtered by file extension.

Fix

• config_parser.py: Always set config["language"] based on config file type (codeflash.toml → java, pyproject.toml → python; package.json already handled)
• cli.py: Call set_current_language() in process_pyproject_config() before the optimizer runs

Testing

• 15/15 tests/test_git_utils.py pass
• E2E verified: cc-plugin stop hook → codeflash --subagent → Java function discovered from git diff

🤖 Generated with Claude Code

[claude]

Claude finished @mashraf-222's task in 9s —— View job

---

PR Review Summary

• Triage: check changed files
• Lint and typecheck
• Resolve stale threads
• Code review
• Duplicate detection
• Test coverage

---

Prek Checks

uv run prek run --from-ref origin/main — passed (ruff check + ruff format both clean). No auto-fixes needed.

Mypy errors in config_parser.py (lines 151, 180, 187, 195, 208, 209) are pre-existing and not introduced by this PR.

---

Code Review

codeflash/code_utils/config_parser.py:163-166

if "language" not in config:
    config["language"] = "java" if config_file_path.name == "codeflash.toml" else "python"

The inference is correct: find_pyproject_toml() (line 138) walks up directories and returns either pyproject.toml or codeflash.toml, so mapping codeflash.toml → java and anything else → python is sound. JS/TS projects return early from parse_config_file via parse_package_json_config and never reach this code. The "language" not in config guard prevents overriding an explicitly set value.

codeflash/cli_cmds/cli.py:115-118

# Set the language singleton early so downstream code (e.g. get_git_diff)
# can use current_language_support() before function discovery.
if pyproject_config.get("language"):
    set_current_language(pyproject_config["language"])

Correct placement — process_pyproject_config runs before the optimizer/discovery, so get_git_diff (which calls current_language_support().file_extensions) will see the right language.

Minor note: since config_parser.py now always sets "language", the if pyproject_config.get("language") guard will always be truthy for valid configs. This is harmless but slightly redundant.

The set_current_language call in optimizer.py:478 (set_current_language(funcs[0].language)) is complementary — it runs later post-discovery and is consistent with the config-based value.

No bugs or security issues found.

---

Duplicate Detection

No duplicates detected. The two set_current_language call sites serve different purposes: the new one in cli.py sets the language from config early (before discovery), while optimizer.py:478 sets it from the discovered function's language after discovery.

---

Test Coverage

Relevant tests pass: 15/15 test_git_utils.py, 49 passed across test_worktree.py, test_formatter.py, test_git_utils.py.

One gap: there is no dedicated unit test for the new language-inference logic in config_parser.py (specifically the codeflash.toml → java path). The process_pyproject_config early set_current_language call also lacks a direct test for non-Python configs. Consider adding a test that parses a minimal codeflash.toml and asserts config["language"] == "java" to lock in this behavior.

---

Last updated: 2026-03-18T00:36Z