Parser fixes from #274, #276, #285 (and Julia impl from #319) were silently reverted by merge-conflict resolutions

[npkriami18]

Summary

Four already-landed fixes on main have been silently reverted (partially or fully) by two chore: resolve merge conflicts with main commits: 3cf31ec and cb9d25a. These merges accepted the feature-branch side of code_review_graph/parser.py for the ReScript PR (#323) and Julia PR (#319), both of which had forked from db2d2df — before the fixes below landed. The result is 44+ pre-existing test failures on main that directly match the behavior the original PRs set out to fix.

Affected PRs (now effectively reverted on main)

PR	Author	Merge commit	What was lost
#274	@azizur100389	aa627fb	CRLF-safe Databricks notebook detection (1 test)
#276	@azizur100389	e6e3144	SHEBANG_INTERPRETER_TO_LANGUAGE, _detect_language_from_shebang(), .ksh extension (10 tests)
#285	@michael-denyer	fe383c7	File-attribution fallback at 5 enclosing_func-gated CALLS emission sites + include_file_sources=False in detect_entry_points (7 tests)
#319	@danvinci	75ce2f1	Julia parser implementation (parser.py Julia references dropped 38 → 6; 21 tests)

Reproduction

git checkout main
uv run pytest tests/ --tb=no -q
# 60 failed, 1015 passed, 1 skipped, 6 errors

The 6 errors are a separate issue (TestApplyToolFilter accesses FastMCP._tool_manager, a private API that does not exist in the pinned fastmcp>=2.14.0). Not addressed here.

Root cause (evidence)

• cb9d25a chore: resolve merge conflicts with main — parent^1 (main) has 6 Julia refs in parser.py; parent^2 (Julia feature branch, forked from db2d2df) has 38 Julia refs; resolution kept main's side, dropping the Julia implementation.
• 3cf31ec chore: resolve merge conflicts with main — same pattern for the ReScript merge, wiping PRs fix: resolve all 8 Windows test failures — UTF-8, CRLF, FastMCP API, stop_at (#239, #241) #274/feat: .ksh extension + shebang-based language detection for extension-less scripts (#235, #237) #276/fix: emit CALLS edges for module-scope code (closes #284) #285.

Neither conflict emitted markers because the feature branches had not touched those regions at all — only the main side had them — so git merge produced a clean auto-resolution with the wrong side accepted.

Proposed fix

Restore the four fixes via cherry-pick into a single PR: #<FILLED_IN>. Branch: fix/restore-reverted-parser-fixes from main. Result: 60 failed → 16 failed on main (remaining 16 are unrelated Java/PHP/GDScript regressions + 6 _tool_manager errors, all out of scope).

To prevent recurrence

When resolving a merge conflict of parser.py with a long-lived feature branch that forked before recent fixes:

1. Prefer git merge --strategy=ort -X theirs only for specific hunks the feature branch truly owns — not for the whole file.
2. Always run uv run pytest tests/test_parser.py tests/test_multilang.py tests/test_notebook.py -q after any merge commit that touches parser.py.
3. Consider gating parser.py merges behind a mandatory CI job on main.

cc @tirth8205 @azizur100389 @michael-denyer @danvinci

[npkriami18]

Second restoration PR opened: #364. Same silent-revert pattern from #361 — restores Java fixes (#275/#278/#280), GDScript support (#316), and a stale stdio-transport test assertion. 60 → 45 failures on main when applied alone; 60 → ≤8 when combined with #362.