fix(cpp): don't mint sourceless god nodes for Metal/CUDA GPU qualifiers

[rajatnagda45]

Closes #4550.

Problem

Metal (.metal) and CUDA (.cu/.cuh) are C++-derived GPU languages that route through the C++ extractor, but their qualifiers are not part of the C++ grammar:

• Metal address-space / function qualifiers: device, constant, threadgroup, thread, kernel, vertex, fragment, ...
• CUDA execution-space / memory qualifiers: __global__, __device__, __host__, __shared__, __constant__, __forceinline__, ...

tree-sitter-cpp parses a leading qualifier as the declaration's type_identifier (the real type is pushed into an ERROR node), and _cpp_collect_type_refs emitted it as a type reference. Each qualifier then became a sourceless, corpus-shared god node that every shader/kernel references (e.g. kernel/device/constant on sample.metal, __global__/__device__/__host__ on sample.cu). That is a provenance violation (a node with no source_file), and the hubs pollute god-node rankings in analyze.

Implementation

• Add a fixed, per-extension set of GPU-dialect qualifier keywords (_CPP_DIALECT_TYPE_QUALIFIERS, keyed by .metal / .cu / .cuh).
• Thread a dialect_quals skip set into _cpp_collect_type_refs (the single choke point for every type_identifier emission) and compute it once per file from the extension.
• Plain .cpp/.cc/.h are untouched — there these words are legal identifiers and still resolve as real user types.

Genuine user types in the shader/kernel (the Vec3 struct) are preserved.

Verification

• New regressions in tests/test_languages.py: test_metal_qualifiers_do_not_mint_sourceless_god_nodes, test_cuda_execution_space_qualifiers_do_not_mint_sourceless_god_nodes (both fail on pre-fix v8, pass after).
• uv/venv commands run:
  • python -m pytest tests/ -q --ignore=tests/test_security.py → 8331 passed, 15 skipped (deselecting the live-DNS security tests and the built-wheel test that need a network/packaged wheel).
  • ruff check . → clean.
  • pyright graphify/extractors/engine.py → unchanged from baseline (0 added).

Limitations

constant Foo* p buries the real Foo type inside the ERROR node, so Foo is not recovered; this change only removes the bogus qualifier node, it does not add the lost type reference. Metal scalar vector types (uint, float4) are left as-is (they are types, not qualifiers).

[github-actions]

Thanks for the pull request, @rajatnagda45. A maintainer will review it soon.

Want to talk it through while it is in review? Come join us on our Discord server. For longer-form discussion there is also GitHub Discussions.

A couple of things that speed up review: make sure the test suite passes on Python 3.10 and 3.13, and that the change keeps extraction deterministic.

[graphify-labs]

Graphify reviewed this change.

No issues found by static checks — no coupling regressions or blocking issues in the code graph. That is not the same as safe to merge: see what was not checked.

Not checked: tests were not run; formal verification proved 0 of 2 changed function(s) (2 not verified).

Formal verification. PR-changed functions: 0/2 verified (0 proven, 0 may-equivalent, 0 distinguished) · 2 not verified (1 vacuous, 1 unsupported).

Not verified on this run: \_cpp\_collect\_type\_refs (vacuous: never exercised), \_extract\_generic (unsupported).

---

Graphify review — findings

Stops Metal and CUDA qualifiers (kernel, device, __global__, __host__, etc.) from being emitted as type references, so they no longer become sourceless god nodes that every shader or kernel points at. tree-sitter-cpp misparses a leading qualifier as the declaration's type_identifier, so _cpp_collect_type_refs now takes a dialect_quals set and skips those names. The filter is chosen by file extension (.metal, .cu, .cuh) in _extract_generic, because these words are legal identifiers in plain C++. Plain .cpp, .cc and .h files are unaffected, so a real type named kernel or device there still resolves. Adds tests for both dialects that check no qualifier leaks as a sourceless node or references target, while the real Vec3 struct reference is preserved.

No blocking issues surfaced. 1 lower-confidence candidate did not survive cross-model review.

Analysis details — impact, health, verification

Impact & health

Graphify review

Impact — 1376 functions depend on the 1017 functions this change touches.

Health — this change adds coupling hotspots:

• worse: extract_cpp() — 39 callers, 5 callees
• worse: walk() — 1 callers, 76 callees

Verification — 1376 functions in the blast radius were not formally verified this run (proofs are advisory here).

Gate & verification

graphify gate

PASS — objectively clean (no health regressions, tests not run — proofs not run this pass (advisory)). Grounded, not self-assessed.

Advisory (not blocking):

• verification_scope: 1285 function(s) in the blast radius were not formally verified this run

Test selection

Test selection

39 of 432 test file(s) selected (9%) via static blast radius.

• tests/test_astro_extraction.py — impact
• tests/test_build.py — impact
• tests/test_cjs_module_extension.py — impact
• tests/test_cpp_nested_and_cli.py — impact
• tests/test_csharp_partial_classes.py — impact
• tests/test_dotnet.py — impact
• tests/test_ejs_extractor.py — impact
• tests/test_ext_tagged_id_clash.py — impact
• tests/test_extract.py — impact
• tests/test_extract_php_closures.py — impact
• tests/test_import_extension_resolution.py — impact
• tests/test_indirect_call_block_scoped_shadow.py — impact
• tests/test_indirect_dispatch.py — impact
• tests/test_indirect_dispatch_assign_return.py — impact
• tests/test_indirect_dispatch_getattr.py — impact
• tests/test_java_member_calls.py — impact
• tests/test_js_exported_scalar_bindings.py — impact
• tests/test_js_member_collisions.py — impact
• tests/test_languages.py — impact, changed-test
• tests/test_multilang.py — impact
• tests/test_php_unparsed_casts.py — impact
• tests/test_python_inherited_method_calls.py — impact
• tests/test_python_property_reads.py — impact
• tests/test_python_pytest_fixture_receivers.py — impact
• tests/test_python_underscore_resolution.py — impact
• tests/test_rationale.py — impact
• tests/test_ruby_resolution.py — impact
• tests/test_scala_context_bounds.py — impact
• tests/test_scala_given_instances.py — impact
• tests/test_scala_self_type.py — impact
• tests/test_scala_top_level_binding.py — impact
• tests/test_scala_type_definition.py — impact
• tests/test_svelte_extraction.py — impact
• tests/test_swift_computed_properties.py — impact
• tests/test_swift_protocol_requirements.py — impact
• tests/test_trailing_newline_not_a_syntax_error.py — impact
• tests/test_ts_new_expression_calls.py — impact
• tests/test_typescript_module_extensions.py — impact
• tests/test_vue_extraction.py — impact

Selection is safe under the controlled-regression assumption; always-run tests + a periodic full run are the backstops. Advisory — it never changes the check verdict.

Docs that may be stale (advisory)

• graphify/skills/agents/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/amp/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/claude/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/claw/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/codex/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/copilot/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/droid/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/kilo/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/kiro/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect
• graphify/skills/opencode/references/query.md § Step 1 — Traversal (lines 61-191): references changed symbols collect

…and 10 more.

Formal verification

Could not verify: Could not verify \_cpp\_collect\_type\_refs.

The verifier did not have enough to check \_cpp\_collect\_type\_refs, so it is saying so rather than guessing. No false assurance is the whole point.

Guarantee: No guarantee either way, this is an honest abstention, not a pass.

Note: Reason: no capturable inputs from the test suite; property tier: not verifiable: all 235 sampled inputs raised on both versions — the function never executed, so 'no divergence' would be vacuous (mostly AttributeError — names the real obstacle, not a sampling gap)

Could not verify: Could not verify \_extract\_generic.

The verifier did not have enough to check \_extract\_generic, so it is saying so rather than guessing. No false assurance is the whole point.

Guarantee: No guarantee either way, this is an honest abstention, not a pass.

Note: Reason: no capturable inputs from the test suite; property tier: parameter `config` is annotated `LanguageConfig` — outside the synthesizable primitive/collection set

· 2 more finding(s) on lines outside this diff (see the check run).