fix: exclude PHP scalar type-hint/cast keywords from ast --kind string

[carlos-alm]

Summary

• codegraph ast --kind string matched PHP AST nodes by node.kind() string alone. tree-sitter-php's primitive_type production (scalar type-hints like string $x and : string return types) and cast_type production ((string) $x) both lex the string keyword as an anonymous grammar token whose kind() string is identical to the named string literal node type — the same collision as TypeScript's predefined_type production fixed in codegraph ast --kind string misclassifies TypeScript primitive-type annotations as string literals #1729/fix: exclude primitive type keywords from ast --kind string matching #1822.
• Fixed in both engines:
  • Native: added requires_named_node: bool to LangAstConfig (crates/codegraph-core/src/extractors/helpers.rs), checked via node.is_named() in classify_ast_node. Set true only for PHP_AST_CONFIG; all other 29 language configs get false explicitly (no behavior change for them).
  • WASM/TS: extended astRequiresNamedNode() in src/ast-analysis/rules/index.ts — already scoped to javascript/typescript/tsx from codegraph ast --kind string misclassifies TypeScript primitive-type annotations as string literals #1729 — to also cover php. This reuses the existing resolveAstKind() guard in ast-store-visitor.ts, already threaded through all three production call sites (features/ast.ts, ast-analysis/engine.ts, domain/wasm-worker-entry.ts).

Closes #1821

Verification (non-tautological)

Per the #1729/#1822 precedent, I verified the new tests actually exercise the fix rather than passing vacuously:

• Reverted the WASM guard (removed php from REQUIRES_NAMED_NODE), rebuilt dist/, reran the PHP WASM tests → 3 of them failed, showing 7 spurious string rows instead of 3 genuine literals.
• Reverted the native guard (requires_named_node: false on PHP_AST_CONFIG), rebuilt + codesigned the native addon, reran the PHP native tests → same 3 failures, same 7-vs-3 mismatch.
• Restored both fixes, rebuilt both sides again — all tests pass.

Test plan

• npx vitest run tests/parsers/ast-nodes.test.ts tests/parsers/ast-all-langs.test.ts tests/engines/ast-parity.test.ts — 87 passed, including new PHP WASM + native regression tests and ast_nodes row-count parity: php
• npm test — full suite green: 3738 passed, 30 skipped, 2 todo, 0 failed
• npm run lint — clean
• cargo clippy --release — no new warnings (checked helpers.rs diff against pre-existing warnings on the base branch)
• Native addon rebuilt via napi build --platform --release and codesigned; native tests ran against the rebuilt binary, not a stale one
• codegraph diff-impact --staged -T — 1 function changed (astRequiresNamedNode), 9 transitive callers across 3 files, all expected (the three production call sites + their test coverage)

Stacked on #1819/#1820 (base branch fix/issue-1819-exported-consts-with-non-recognized) — only the ast-analysis/extractor/test diff is this issue's change.

[greptile-apps]

Greptile Summary

This PR fixes a false-positive in codegraph ast --kind string for PHP, where tree-sitter-php's primitive_type (scalar type-hints like string $x, : string return types) and cast_type ((string)) productions lex the string keyword as an anonymous grammar token whose kind() string is identical to the named string literal node — the same collision previously fixed for TypeScript/JS in #1729.

• Native engine: adds requires_named_node: bool to LangAstConfig and gates classify_ast_node on node.is_named() when set; only PHP_AST_CONFIG sets it true, all 29 other language configs explicitly set false.
• WASM/TS engine: extends REQUIRES_NAMED_NODE (previously JS_REQUIRES_NAMED_NODE) to include 'php', reusing the existing resolveAstKind() guard already threaded through all production call sites.
• Tests: adds WASM and native PHP regression suites (text-content and exact-count assertions) plus parity and multi-lang tests covering parameter type-hints, return types, and casts.

Confidence Score: 4/5

The fix is safe to merge; the core guard logic in both engines is correct and does not touch any language other than PHP.

The Rust and WASM fixes are correct and symmetric. The only wrinkle is in the new PHP test suite: the does not misclassify (string) cast assertion checks line 12, which is a blank line in the fixture — the actual cast lives on line 14 — so that particular test passes vacuously even without the fix. The remaining tests (text-content guards and the exact-count assertion) do catch the regression, so there is no correctness gap in the shipped fix, but the cast-specific test does not exercise what its comment describes.

tests/parsers/ast-nodes.test.ts — the cast-keyword test checks the wrong line number.

Important Files Changed

Filename	Overview
crates/codegraph-core/src/extractors/helpers.rs	Adds requires_named_node: bool to LangAstConfig, threads the Node reference into classify_ast_node, and gates classification on node.is_named() when set. Only PHP_AST_CONFIG sets this true; all 29 other configs explicitly set false to make future additions intentional. Logic is correct and well-scoped.
src/ast-analysis/rules/index.ts	Renames JS_REQUIRES_NAMED_NODE to REQUIRES_NAMED_NODE and adds 'php' to the set, extending the existing TypeScript/JS guard to PHP. Comment is accurate and well-documented.
src/ast-analysis/visitors/ast-store-visitor.ts	Comment-only update to resolveAstKind's doc block, extending the issue reference to include PHP (#1821). No logic changes.
tests/engines/ast-parity.test.ts	Extends the PHP parity fixture to include a typed parameter (string $label), a return type annotation, and a (string) cast, making the parity test exercise both engine paths against the new guard.
tests/parsers/ast-all-langs.test.ts	Adds a native multi-language regression test asserting that no ast_nodes row with kind='string' from the PHP fixture has unquoted text='string', correctly verifying the fix end-to-end.
tests/parsers/ast-nodes.test.ts	Adds WASM and native PHP describe blocks with 5 tests each covering parameter type-hints, return type-hints, casts, genuine literals, and exact counts. The does not misclassify (string) cast test checks line 12 (blank) instead of line 14 where the cast actually lives, making that specific assertion vacuously true — the other tests still catch the regression.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[tree-sitter walks PHP AST node] --> B{requires_named_node?}
    B -- false --> D[classify by kind string]
    B -- true --> C{node.is_named?}
    C -- false --> E[return None — skip anonymous token]
    C -- true --> D
    D --> F{kind matches config?}
    F -- string_types --> G[emit AstNode kind=string]
    F -- new_types --> H[emit AstNode kind=new]
    F -- throw_types --> I[emit AstNode kind=throw]
    F -- no match --> J[skip]

    subgraph WASM side
        K[resolveAstKind] --> L{astRequiresNamedNode langId?}
        L -- php/ts/tsx/js --> M{node.isNamed?}
        M -- false --> N[skip anonymous token]
        M -- true --> O[emit AstNode]
        L -- other lang --> O
    end

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
flowchart TD
    A[tree-sitter walks PHP AST node] --> B{requires_named_node?}
    B -- false --> D[classify by kind string]
    B -- true --> C{node.is_named?}
    C -- false --> E[return None — skip anonymous token]
    C -- true --> D
    D --> F{kind matches config?}
    F -- string_types --> G[emit AstNode kind=string]
    F -- new_types --> H[emit AstNode kind=new]
    F -- throw_types --> I[emit AstNode kind=throw]
    F -- no match --> J[skip]

    subgraph WASM side
        K[resolveAstKind] --> L{astRequiresNamedNode langId?}
        L -- php/ts/tsx/js --> M{node.isNamed?}
        M -- false --> N[skip anonymous token]
        M -- true --> O[emit AstNode]
        L -- other lang --> O
    end

Loading

_{Reviews (1): Last reviewed commit: "fix: exclude PHP scalar type-hint/cast k..." | Re-trigger Greptile}

[greptile-apps]

Cast-test assertion checks a blank line, not the cast line

PHP_FIXTURE_CODE places return (string) $value; on line 14 (counting from the first line of the template literal: <?php → 1, blank → 2, class Greeter { → 3, greet(...) → 4, …, normalize body → 14). Line 12 is a blank line, so nodes.some((n) => n.line === 12) is always false regardless of whether the fix is in place. The cast false-positive (when present) appears at line 14, so this specific assertion would pass vacuously even with the guard reverted. The other tests in the suite (text === 'string' guards and the exact-count test) still catch the regression, so correctness is not broken — but this test does not verify what its comment claims.

[github-actions]

Codegraph Impact Analysis

1 functions changed → 9 callers affected across 3 files

• astRequiresNamedNode in src/ast-analysis/rules/index.ts:417 (9 transitive callers)