Parse MODULE.bazel with Python's ast in the msan guard

[zzcgumn]

Follow-up to #423, which was merged with a hand-written Starlark lexer in python/tests/ci_msan_test.py. This applies the same rewrite as #424.

Why

The lexer kept drawing review findings. The latest: it joined separate statements into a fictitious call. With

dep_function = bazel_dep
("toolchains_llvm", "9.9.9")
bazel_dep(name = "toolchains_llvm", version = "1.11.2")

_toolchains_llvm_bazel_dep_version() returned 9.9.9. A one-line version bump had also grown the file by ~1400 lines.

What

Starlark's syntax is a subset of Python's, so the guard now parses MODULE.bazel with ast.parse. It walks the ast.Call nodes whose callee is a bare bazel_dep / *_override name. Python's tokenizer handles strings, escapes, raw strings, comments, continuations and statement boundaries exactly:

• attribute calls such as ext.bazel_dep(...) are excluded;
• grouping parens around a callable are transparent;
• any argument that isn't a single string literal still fails closed.

The regression test for the example above was added first and confirmed failing. Tests that only exercised tokenizer details (escapes, raw-string backslash parity, comment placement, continuations) were removed. The fail-closed cases are now subtests of one test. The file goes from 1683 to 614 lines; the non-parser tests are unchanged. Full //python/... suite green (24/24).

🤖 Generated with Claude Code

[zzcgumn]

Mirrored #424's keyword-unpacking fix (6085e98): local_path_override(**{"module_name": "toolchains_llvm", ...}) slipped past the override guard because ast records ** as a keyword with arg=None. _argument() now returns _UNRESOLVED in that case, so the guard fails closed. Regression tests were added red first. Full //python/... suite green (24/24).

[Copilot]

🟢 Approval recommended

The focused rewrite addresses the parsing defect, with regression coverage and no identified blocking issues.

0 open findings

What changed in this PR

Simplifies the MemorySanitizer CI guard by replacing its hand-written Starlark lexer with Python’s ast parser.

Changes:

• Fixes incorrect joining of separate statements into calls.
• Preserves dependency-version and override checks.
• Consolidates parser tests and adds statement-boundary and argument-unpacking coverage.

File	Description
python/​tests/​ci_msan_test.py	Replaces the lexer with AST helpers and updates regression tests.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.