What does this PR do?

The C++ extractor emitted no node for an alias declaration (using X = …;) or a typedef. A field, parameter or return typed with one, even in the declaring header, therefore resolved to a sourceless stub, which the corpus rewire can grow into a hub or bind to an unrelated same-named symbol elsewhere.

Changes (graphify/extractors/engine.py, in _cpp_extra_walk):

• A namespace- or file-scope alias_declaration / type_definition becomes a type node contained by its file. Every declarator of a typedef counts: typedef int Handle, *HandlePtr;, typedef int (*Fn)(int);, typedef int Arr[4]; (_cpp_declarator_type_name walks pointer, function, parenthesized and array declarators). Alias templates (template <class T> using Ptr = T*;) count too. The handler returns False, so the walk continues and a typedef'd struct body still gets its own node and fields.
• Not declarations here, on purpose:
  • Member aliases (using value_type = T;, typedef … size_type; in a class). They sit in every container, and C++ member ids aren't class-qualified, so they would fold into one hub node. Scope is checked structurally (parent is translation_unit or a namespace declaration_list, looking through template_declaration and through #ifndef/#if/#ifdef/#else wrappers, which tree-sitter models as preproc_* nodes: an include guard is not a scope, as the graphify-labs bot pointed out). A member alias inside a conditional in a class body is still a member.
  • The struct a C-style typedef struct Foo {…} Foo; names: the struct's own node is it.
  • Same-name re-exports (using Bar = ns::Bar;), which denote the original type and declare no new name. My Go alias PR uses the same rule; there a re-export node measurably regressed uber-go/zap.
• Declarator names are de-duplicated with a set: one typedef can declare thousands of names, and a list in check was O(N²) (a 40,000-name typedef took 6.2 s; now 1.1 s, most of it emitting the 40,000 nodes v8 dropped).
• C++ never populates namespace_stack, so the id is _make_id(stem, name), the one ensure_named_node looks up for a name declared in the file.
• graphify/cache.py: AST cache schema 20 → 21 (the branch includes v8 at 0.9.93).

Two existing tests updated (test_cpp_nested_and_cli.py): typedef enum { A, B } Flag; / typedef union {…} Anon; asserted that Flag / Anon are not nodes. Their stated concern is a nameless node and the anonymous specifier's members. They now expect Flag / Anon as alias nodes owning none of the members, and still forbid a nameless node and the members (A, a, b).

Type of change

• Bug fix
• New feature
• Documentation
• Tests or CI
• Refactor
• Security fix

Verification & Invariants

Invariant: a source-backed declaration wins over a sourceless stub, without making a previously unique, correct resolution ambiguous.

• User-visible: a name declared in two files is now honestly ambiguous. On fmt, graphify explain test_string used to pick compile-test.cc's struct silently. It now refuses with "Ambiguous: matches 2 nodes" (exit 3), because util.h declares using test_string = … too. That's the existing fail-closed behaviour, now fed by the real declarations.

• Persisted state invalidated: AST cache schema bump (first build after upgrading re-extracts code files once).

• Conflict note: my other open PRs also bump the schema 20 → 21. Whichever merges later needs a one-line rebase.

• Read the CONTRIBUTING.md guide.

• Reproduced the issue and identified the invariant.

• Made the smallest fix necessary.

• Added a regression test (if bug fix) or isolated boundary test.

• Kept the PR description synchronized with the final implementation.

• Documented any limitations / unsupported cases explicitly.

Limitations: the anonymous specifier named by a typedef still contributes no members (pre-existing). C (.c with the C config) models no types at all, so C typedefs are a separate feature.

How was this tested?

• New tests/test_cpp_type_alias.py (9 tests): sourced alias nodes on their lines for every form, contains edges, field references binding to the aliases, member aliases not emitted, the C-style typedef struct Foo {…} Foo; resolving to the struct, the re-export boundary, a cross-file reference rewired onto the alias, aliases inside an include guard and #if/#elif/#else/#ifdef blocks, and a guarded member alias staying a member. 4 fail on v8 (0.9.93) and 5 are guards; the conditional test also fails on this branch's earlier revision.

• Real repos, full-graph diff v8 (0.9.93) → this branch (every edge keyed by its site):

  • google/leveldb (132 files): sourceless nodes 644 → 630, edges into sourceless 1,440 → 1,393, 47 stub → real, 0 real → stub, 0 moved, 0 lost.
  • fmtlib/fmt (73 files): edges 12,251 → 12,373, sourceless nodes 1,112 → 987, edges into sourceless 2,271 → 1,958, 311 stub → real, 0 real → stub, 0 moved. fmt keeps many aliases inside #if blocks (enable_if_t, conditional_t, basic_format_context, …). I checked all 12 edge sites that exist only on v8:
    • 4 are #include <variant> / <any> hitting the same external stub, whose label had been taken over by a case-folded type stub Variant/Any;
    • 1 was gtest.h's Print(const Optional<T>&) wrongly bound to gmock.h's function Optional(), now bound to the using Optional = … alias in gtest.h itself;
    • 1 was test_string(...) in ostream-test.cc wrongly bound to a struct in the unrelated compile-test.cc. It's now unresolved: the included util.h declares using test_string = …, and two declarations make the name ambiguous (fail-closed);
    • 5 are native_int128 / native_uint128 / ullong, declared twice in mutually exclusive #if branches (an alias, and an enum class fallback): the node now sits at the first declaration (core.h L412 instead of L421), the duplicate map() reference collapses into one, and a sourceless native_int128 stub that a forward reference used to mint is gone;
    • 1 was days(42) in chrono-test.cc bound to an unrelated days() method in chrono.h; it now lands on the test's own using days = … inside #if.

• Scenarios (fmt, leveldb): identical graph under a different PYTHONHASHSEED, parallel extraction, a warm cache, and a cache filled by v8 first. The guarded-alias fixture gives the same graph as LF, CRLF, BOM+CRLF and tab-indented. 1,680 mutated fmt/leveldb sources with injected guards, #if/#else, aliases and braces: 0 crashes or hangs on either side, same total time (22.6 s vs 22.7 s).

• Performance: full extract(), cold cache, 3 interleaved rounds, v8 vs this branch: fmt 4.10/4.22/4.23 s vs 4.00/4.18/4.01 s, leveldb 1.81/1.89/1.81 s vs 1.88/1.75/2.01 s; peak RSS identical. Per file, no file is slower beyond noise. The handler runs only on alias and typedef nodes, and the scope walk only climbs a node's ancestors.

uv run pytest tests/test_cpp*.py tests/test_languages.py tests/test_multilang.py -q   # 664 passed
# Full suite: per Python, `uv sync --all-extras --frozen`, then `uv run pytest tests/ -q --tb=short`.
#   this branch on v8 0.9.93 — 3.10: 8472 | 3.13: 8470 passed (3.13 also hit a watch-lifecycle timing flake that fails on v8 under the same load)
#   with my other open branches merged on top — 3.10: 8536 | 3.12: 8535 | 3.13: 8535 | 3.14: 8535 passed
#   only the 7 DNS-dependent test_security tests fail (no network here; they fail the same way on v8)
uv run python -m tools.skillgen --check / --audit-coverage / --schema-singleton / --monolith-roundtrip / --always-on-roundtrip   # all OK
uv run python -m tools.skillgen --bless   # 134 artifacts, no changes
uv run ruff check .                       # v0.15.14, as pinned in .pre-commit-config.yaml
uv run pyright                            # 643 errors on v8 (0.9.93) and on this branch; none new
uv run bandit -r graphify -ll             # 12 findings, identical to v8
GRAPHIFY_TEST_CLI=.venv/bin/graphify bun test tests/pi.test.ts   # 20/20
uv run graphify --help && uv run graphify install   # OK, in a throwaway HOME

Graphify-specific checklist

• I updated generated skill artifacts (uv run python -m tools.skillgen --bless) when changing their source fragments. (Ran --bless: all 134 artifacts regenerated with no changes, and --check passes. No source fragments are touched.)
• I confirmed that AST/structural extraction remains deterministic (no ambient state dependencies like ENV variables).
• I reviewed changes for security implications (no unsafe interpolation into shell/Python).
• I confirmed no API keys or local-only graph data are included.
• (If applicable) I disclosed AI authorship in my commit messages.