fix(php): import decomposition, enum cases, anonymous class scope — F53,F54,F55 (#1931)

[prajapatisparsh]

F53: iterate all namespace_use_clause + namespace_name children
F54: enum_case captured as @declaration.const
F55: anonymous_class @scope.class
F56 closed by #1937
7 new tests, golden regenerated.

Refs #1931

[vercel]

@prajapatisparsh is attempting to deploy a commit to the NexusCore Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

✨ PR Autofix

Found fixable formatting / unused-import issues across 14 changed lines. Comment /autofix on this PR to apply them, or run npm run lint:fix && npm run format locally.

{"schema":"gitnexus.pr-autofix/v2","state":"fixes-available","pr_number":1989,"changed_lines":14,"head_sha":"bcb05ae8f6e83763ab2a29fc44d5460458d95f39","run_id":"26936470830","apply_command":"/autofix"}

[github-actions]

CI Report

✅ All checks passed

Pipeline Status

Stage	Status	Details
✅ Typecheck	success	tsc --noEmit
✅ Tests	success	unit tests, 3 platforms
✅ E2E	success	gitnexus-web changes only

Test Results

Tests	Passed	Failed	Skipped	Duration
10602	10586	0	16	587s

✅ All 10586 tests passed

16 test(s) skipped — expand for details

• COBOL pipeline benchmark > scales with file count
• C++ ADL emit benchmark > emit phase scales sub-quadratically with co-scaled files and sites
• C++ pipeline benchmark > scales with file count
• C# pipeline benchmark > scales with file count — namespaces spread across the solution
• C# pipeline benchmark > scales with file count — all types in one (global) namespace bucket
• C# pipeline benchmark > scales with file count — all types in one (named) namespace bucket
• Go pipeline benchmark > scales with file count (workers enabled)
• Go pipeline benchmark — worker pool (issue Worker idle timeout kills long Go scope extraction and surfaces as Napi::Error during analyze #1848) > does not quarantine the large generated Go file on sub-batch idle timeout
• Go structural interface detection benchmark > scales linearly with interface × struct count
• Go structural interface detection split-phase benchmark > separates index-build and detection time
• PHP pipeline benchmark > scales with file count (workers enabled)
• Ruby pipeline benchmark > scales with file count (workers enabled)
• Rust pipeline benchmark > scales with file count (workers enabled)
• Vue pipeline benchmark > scales with component count
• run.cjs direct-exec entrypoint (fix(cli): steer docs, skills, and hooks through a CLI-neutral project-local runner (#1939) #1945) > resolves a .cmd shim via the Windows shell branch, passing args and exit code
• buildTypeEnv > known limitations (documented skip tests) > Ruby block parameter: users.each { |user| } — closure param inference, different feature

Code Coverage

Tests

Metric	Coverage	Covered	Base	Delta	Status
Statements	80.38%	37361/46476	N/A%	—	🟢 ████████████████░░░░
Branches	69.04%	23744/34389	N/A%	—	🟢 █████████████░░░░░░░
Functions	85.59%	3917/4576	N/A%	—	🟢 █████████████████░░░
Lines	84.04%	33628/40011	N/A%	—	🟢 ████████████████░░░░

---

_{📋 View full run · Generated by CI}

[magyargergo]

🔍 Tri-Review — PHP import / enum / anonymous-class coverage (F53 / F54 / F55)

Methods & engines. GitNexus swarm (risk, test/CI) + Compound-Engineering personas (correctness, adversarial, maintainability, testing) + Codex. Engine breakdown: 6 Claude reviewer lanes + Codex (the only independent engine). Codex investigated the PR but — its sandboxed shell was blocked, forcing GitHub-fetch-only analysis — it did not return a final structured verdict. Its interim conclusions corroborated the grammar shape (namespace_use_declaration = commaSep1(namespace_use_clause)) and the anonymous-class path, but the headline finding below rests on a Claude lane plus my own end-to-end reproduction, not an independent Codex confirmation. Two of the methods are Claude under different personas (correlated), so agreement here is "consistent across personas," not three independent confirmations.

Bottom line. The F53/F54/F55 changes are correct and well-validated. I reproduced correct output through the real tree-sitter-php parser (use A, B, C; → 3 imports; use Foo\Bar, Baz\Qux as Quux, X\Y; → 3 with the alias resolved; enum cases and anonymous-class scopes captured), and the new behavior is byte-enforced by the golden-captures digest, which CI passes. No regression to existing PHP resolution. The actionable items are one pre-existing P1 bug in the same function being edited (a one-line fix worth folding in) plus minor cleanup and a test-strength nit.

Final verdict	production-ready with minor follow-ups — F53/F54/F55 are correct and CI-validated; the P1 is pre-existing and non-blocking; the rest is P2/P3 cleanup.
Branch hygiene	Clean — the PR's own change is 7 files, all scoped to PHP coverage, mergeable: MERGEABLE. The branch carries merge-from-main commits (head 3741ee1d); the most recent merge pulled in unrelated #1990 (cpp ADL) but touched no PHP file — all reviewed PHP sources/tests/golden are byte-identical to the tree I reviewed.
Merge state	All functional CI checks green; the only non-passing check is Vercel (“Authorization required to deploy” — a deploy-auth gate, not a code test). mergeStateStatus: UNSTABLE is attributable to that Vercel gate alone.

Note: the PR head advanced from 8309a533 to 3741ee1d (a main merge) while this review ran; the merge brought in unrelated #1990 work only — every PHP file under review is unchanged, so all findings and line anchors below remain valid.

🔴 Headline — P1, pre-existing, strongly recommend fixing here

Grouped use imports silently drop their namespace prefix. decomposeGrouped() (import-decomposer.ts:195) reads the group prefix with findNamedChild(stmtNode, 'qualified_name') ?? findNamedChild(stmtNode, 'name'), but tree-sitter-php 0.23.x emits that prefix as a namespace_name node — so both lookups return null, prefix = '', and every inner clause loses its namespace.

• Reproduced (real parser; coordinator + adversarial lane): use App\Models\{User, Post as P}; → source="User" / source="Post" (should be App\Models\User / App\Models\Post); use A\B\C\{D}; → source="D".
• Downstream impact (reproduced by the adversarial lane): with two same-simple-name classes in different namespaces, the prefix-stripped bare name resolves to the wrong file via the PSR-4 suffix fallback. The existing php-grouped-imports fixture hides this because its class names (User/Repo/Main) are globally unique.
• Not introduced by this PR (decomposeGrouped is untouched) and it does not block F53/F54/F55. But it lives in the exact function being rewritten, and the new code already identifies the namespace_name node type — so the fix is one line at 195: … ?? findNamedChild(stmtNode, 'namespace_name'). Add a fixture with two same-simple-name classes in different namespaces so the suffix fallback can't mask a regression. (Inline note left at the nearest in-diff line.)

🟡 Inline findings (verified, anchored in-diff)

1. import-decomposer.ts:66 — dead code + misleading grammar comment (P2). The new "First child: namespace_name" block never executes. A direct namespace_name child of namespace_use_declaration appears only in the grouped form (use Prefix\{…}, where namespace_name is the prefix) — and that form is detected and returned earlier (lines 54–57, namespace_use_group). Every declaration that reaches this non-grouped block therefore has only namespace_use_clause children, so findNamedChild(stmtNode, 'namespace_name') is always null here (confirmed by real parse across 12 probes — every input with a direct namespace_name child also had namespace_use_group; Codex independently derived the non-grouped grammar as commaSep1(namespace_use_clause)). The clause loop below does all the work, so output is correct, but lines 64–79 are unreachable and the comment ("the grammar puts the first import path as a namespace_name child") is wrong for the non-grouped form it describes. (Note the symmetry with the headline: this dead block looks for a namespace_name prefix node that only the grouped path has — yet the grouped path at line 195 fails to read it.)
2. php-coverage.test.ts:82/89 — F55 test strength (P2). The first anonymous-class test proves the new rule (counterfactual: removing (anonymous_class) @scope.class → 0 class scopes → fails). The second (fnScopes ≥ 1) asserts pre-existing method_declaration → @scope.function, which holds regardless of the new rule. Pin classScopes to toBe(1) and document the second as a method-scoping regression guard.

🟢 Lower-priority / notes

• parseSingleUseClause (lines 115–134) is now dead code — its only caller was replaced by the rewrite. ESLint warns; CI still exits 0 (no --max-warnings). Remove it. (P3)
• Duplicated alias-detection between parseUseClause (155–168) and parseInnerClause (248–261) — extractable to one helper. (P2, maintainability)
• Enum implements I emits no inheritance edge — the heritage synthesizer skips enum_declaration; pre-existing but adjacent to the new enum work, so a reader might assume enum heritage is now handled. (P3)
• Aliased use function Foo\f as g; collapses to kind='alias', dropping the function qualifier — harmless (interpret.ts normalizes alias/function/const identically) and pre-existing; not a defect.

✅ Validated / refuted suspicions (credit where due)

• Double-count from the dead block — refuted: use A, B, C; → exactly 3.
• detectQualifier mis-firing on use functionFoo\bar; — refuted: the regex requires a trailing \s word boundary.
• Trailing comma use A, B,; / leading separator — refuted: no phantom imports.
• Nested anonymous classes — refuted: two distinct, correctly-nested scopes.
• enum_case @declaration.const colliding with class constants — refuted: PHP has no class-const capture, so enum cases are the sole producer; no shadowing.

Coverage & caveats

Focused 7-file PHP PR; the full source diff was read and exercised through the real tree-sitter-php parser. Test gaps noted (no use function f1, f2; KIND-propagation assertion — though I verified that path works; no grouped-prefix fixture). Codex did not return a final verdict, so cross-engine confidence is lower than a clean three-method run — weight the Claude-lane agreement as "consistent," not independently confirmed.

Automated multi-tool digest (GitNexus swarm + CE personas + Codex). Verify before acting.

[magyargergo]

Can you test if the $service->execute(); will connect the CALLS edge properly to the right nodes? 🙏

[prajapatisparsh]

CALLS edge won't resolve. $service has no type binding (anonymous_class doesn't match the constructor pattern). The scope/capture-level fix is what F55 was about; full CALLS resolution for anonymous class instances is a separate gap.

[prajapatisparsh]

okay i am working on it , but i have a doubt regarding grouped prefix bug , it is pre-existing, not from this PR. Do you still want the fixture added here, or should that be a separate issue?

[magyargergo]

okay i am working on it , but i have a doubt regarding grouped prefix bug , it is pre-existing, not from this PR. Do you still want the fixture added here, or should that be a separate issue?

You’re correct! The grouped prefix bug is pre-existing on main, not from F53/F54/F55. Please keep the one-line namespace_name fix in this PR. A dedicated new fixture isn’t required to merge; if you want extra safety, add a small inline test in php-coverage.test.ts asserting @import.source for use App\Models\{User}.

[prajapatisparsh]

@magyargergo should i re apply the wiped fixes ?

[magyargergo]

@magyargergo should i re apply the wiped fixes ?

You pushed changes that were conflicting so i removed that commit, and give you the option to properly rebase.

[prajapatisparsh]

@magyargergo should i re apply the wiped fixes ?

You pushed changes that were conflicting so i removed that commit, and give you the option to properly rebase.

okay okay