[mccalpink]

Pull Request

Linked issue

Closes #1214

Implements the fix suggested in the issue (#1214).

What & why

parser._OuterBinding is a str subclass whose __new__ takes a second argument, levels. C# nested classes that use a field of their outer class store it in extra["receiver_type"] of the CALLS edge. The process-pool parser pickles every worker result; pickle rebuilds a str subclass through __new__(cls, str(self)), which drops levels. The parent process then fails with TypeError: _OuterBinding.__new__() missing 1 required positional argument: 'levels', and build (and update with 8 or more files) ends with BrokenProcessPool.

The fix adds _OuterBinding.__reduce__, which returns (_OuterBinding, (str(self), self.levels)). The value and levels now survive the round trip, and nothing else changes. The _OuterBinding already sits in receiver_type, so it has to survive the process-pool queue; keeping the type means the edge is the same on the serial, thread and process paths. (The resolver itself reads the int extra["csharp_receiver_levels"], which the worker sets before pickling, not .levels of the unpickled value.) I checked for other str/int/tuple subclasses in code_review_graph/: this is the only one.

Changes:

• code_review_graph/parser.py: _OuterBinding.__reduce__.
• tests/test_csharp_outer_binding_pickle.py: pickle round trip of _OuterBinding; pickle round trip of the (nodes, edges) parsed from the snippet in the issue; full_build with CRG_PARSE_EXECUTOR=process on 9 C# files with a nested outer-field call (it also asserts that the parallel executor was built, so a serial fallback cannot make it pass).
• CHANGELOG.md: a bullet under ## [Unreleased] / ### Fixed.

Without the fix all three new tests fail (the last one with the BrokenProcessPool from the issue); with it they pass.

How it was tested

uv run pytest tests/test_csharp_outer_binding_pickle.py --tb=short -q
# before the fix: 3 failed (BrokenProcessPool, TypeError ... missing 'levels')
# after the fix:  3 passed

uv run pytest tests/ --tb=short -q
# 11 failed, 8255 passed, 981 skipped, 3 xfailed, 2 xpassed

uv run ruff check code_review_graph/
# All checks passed!

uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
# Success: no issues found in 95 source files

The same 11 tests fail on an unmodified staging in my environment (8252 passed there, 3 fewer than here: my 3 new tests), also with a clean temporary directory (TMPDIR pointed at a fresh directory). They exercise git and non-repository directory handling, not the parser code changed here. Python 3.13.

Checklist

• Tests added for new functionality
• All tests pass: uv run pytest tests/ --tb=short -q (11 tests fail locally, identically on staging without this change; see above; CI is the reference)
• Linting passes: uv run ruff check code_review_graph/
• Type checking passes: uv run mypy code_review_graph/ --ignore-missing-imports --no-strict-optional
• Lines are at most 100 characters
• Docs updated where behaviour changed (README, docs/, docstrings)

This PR was prepared with AI assistance (Claude Code). I reproduced the bug, reviewed the change, and ran the tests myself.