fix(go): generic composite literal constructor inference (F33) by evander-wang · Pull Request #1976 · abhigyanpatwari/GitNexus

evander-wang · 2026-06-02T13:19:56Z

What changed

Go generic composite literal constructors (e.g. Box[User]{}) were not captured by the scope-resolution pipeline. The tree-sitter parser produces generic_type nodes for these patterns, but the Go ingestion code only handled plain type_identifier nodes in composite literal positions.

This is Go language coverage gap F33.

Why

Generic composite literals are a core Go 1.18+ feature. Without this fix, any codebase using pkg.T[Type]{} patterns gets no constructor-inferred type binding, producing missing CALLS edges and incomplete call graphs.

How

Extended tree-sitter queries (query.ts): Added (generic_type) to constructor detection patterns in GO_SCOPE_QUERY
New normalization (captures.ts): normalizeGenericConstructorCapture() strips type parameters from generic_type nodes to extract the base type name (Box[User] → Box)
Recursive type extraction (type-binding.ts): Enhanced extractSimpleTypeNameText() to recursively unwrap generic_type and qualified_type nesting

How to verify

cd gitnexus

# Unit tests — Go type binding (16 tests)
npx vitest run test/unit/scope-resolution/go/go-type-binding.test.ts

# Golden snapshots (9 tests)
npx vitest run test/unit/scope-resolution/go/go-captures-golden.test.ts

# Integration tests — Go resolver (136 tests)
npx vitest run test/integration/resolvers/go.test.ts

# Typecheck
npx tsc --noEmit

All pass: 16/16 unit ✅, 9/9 golden ✅, 136/136 integration ✅.

Risk and rollback

Risk: Low — changes isolated to Go ingestion, no shared infrastructure, backward compatible
Rollback: Revert the commit; non-generic constructor captures are unaffected
Follow-ups: Edge-case tests for nested generics (Box[Container[User]]{}) and qualified generics (pkg.Box[User]{}) can be added in a separate PR

Refs #1927

vercel · 2026-06-02T13:20:01Z

@evander-wang is attempting to deploy a commit to the NexusCore Team on Vercel.

A member of the Team first needs to authorize it.

github-actions · 2026-06-02T14:09:41Z

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
10941	10928	0	13	677s

✅ All 10928 tests passed

13 test(s) skipped — expand for details

COBOL 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)
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	80.3%	38239/47620	79.84%	📈 +0.5	🟢 ████████████████░░░░
Branches	68.85%	24315/35314	68.5%	📈 +0.3	🟢 █████████████░░░░░░░
Functions	85.45%	3978/4655	84.94%	📈 +0.5	🟢 █████████████████░░░
Lines	83.9%	34397/40993	83.36%	📈 +0.5	🟢 ████████████████░░░░

_{📋 View full run · Generated by CI}

magyargergo

PR Tri-Review — F33 generic composite-literal constructor inference

Verdict: production-ready with minor follow-ups. No correctness, security, or performance regression was found across three review methods; the implementation is correct and the primary cases are well-tested. The remaining items are maintainability/coverage follow-ups, none merge-blocking.

Methods & engine independence (stated honestly). GitNexus risk lane + 5 Compound-Engineering personas (correctness, adversarial, maintainability, performance, testing) + Codex (gpt-5.5, xhigh). 7 of the 8 perspectives (6 review lanes + this synthesis) are Claude — their agreement is "consistent across personas," not independent confirmation. Only Codex is a different engine, so Codex-corroborated points carry the real cross-engine weight. The coordinator re-verified every load-bearing claim with real tree-sitter parses in a worktree.

What's solid (validated, not merely unchallenged):

Correctness lane refuted all six probed failure modes with real parses; the adversarial lane could not break it (exotic / multi-arg / *T / slice / map type-args, ~2000-level nesting ≈28 ms with no overflow, capture leakage from type-param lists / field types / signatures / new/make/assertions → no spurious edges, container-of-generic correctly non-capturing, no double-emission). Codex independently concluded "no runtime defect."
Strictly additive for non-generic Go: the new normalizeGenericConstructorCapture guard cannot fire for type_identifier / qualified_type nodes, so existing capture output is unchanged — zero regression surface for non-generic repos.
Performance: the new normalization is O(1) per match; the #1848/#1915 hot-path linearity is intact. bench --check → scaling ratio 1.144 (budget 1.5); scaling_budget correctly left unchanged, only the fingerprint rebaselined.
Legacy/registry split is test-accounting, not a production divergence: Go is registry-primary by default (MIGRATED_LANGUAGES), so no production consumer hits the legacy DAG; the LEGACY_RESOLVER_PARITY_EXPECTED_FAILURES entry is correctly scoped to the one new test.
Test runs (correctness/adversarial lanes, worktree sequential parser): Go scope-resolution unit 85/85, constructor-inference integration 9/9 (incl. the new Box test), captures-golden 9/9; bench fingerprint matches the rebaseline.

Inline findings (maintainability, non-blocking):

[P2] extractSimpleTypeNameText is now duplicated verbatim in captures.ts and type-binding.ts — drift hazard.
[P3] extractTypeNode's fallback gained generic_type but its sibling extractCompositeLiteralTypeNode did not — harmless today (the var x = Box[T]{} path is rescued by childForFieldName('type'); coordinator reproduced → {name:b, type:Box}), but a latent inconsistency.

Codex's independent edge case — investigated, NOT a confirmed defect. Codex flagged that generic-qualified models.Box[T]{} drops the package qualifier at capture time (emits Box), while plain models.Box{} keeps models.Box (coordinator reproduced both). Verified consequence: for the type-binding / receiver path this is provably benign — interpretGoTypeBinding → normalizeGoTypeName strips both the qualifier and the [...], so plain and generic both reduce to Box (interpret.ts:37, 93–96); any shadowing would already affect plain models.Box{}, so this PR introduces no regression there. The reference/CALLS path is untested under shadowing (a package that both declares a local Box and instantiates an imported models.Box[T]), but Codex itself downgraded this to "coverage, not a runtime defect" after tracing the call path, and the non-shadowing case is proven correct by the new integration test. Recommend a shadowed-qualified-generic regression test to close it.

Coverage gaps (nice-to-have; behaviors verified to work but untested): var x = Box[T]{} (var-spec path), multi-assign generic (a, b := X{}, Box[int]{}), nested generic (Box[Container[int]]{}), multi-type-arg (Box[K,V]{}), qualified generic at the unit level, and a negative assertion that the type argument models.User does not create a spurious User edge (adversarial confirmed it doesn't). Separately, the scope-capture bench synthetic Go source contains no generic literals, so the F33 path isn't exercised at the 250/800-entity scale — perf for it rests on the O(1)-per-match analysis plus the 1.144 measurement on the small fixture; one generic literal in the bench generator would gate it at scale.

Merge state & branch hygiene. mergeable=MERGEABLE (no conflicts) but mergeStateStatus=BLOCKED → checks failing. Branch hygiene: one merge-from-main commit (3412d4c0) present but harmless and merge-safe (no overlap with the F33 production files); the other three commits are the F33 work.

CI. All substantive gates green — typecheck, lint, format, benchmarks, scope-resolution parity, coverage, tree-sitter ABI (all platforms), CodeQL, both Build & Push, macOS platform-sensitive. Failing: Vercel (deploy-auth gate, not code) and tests / windows-latest (platform-sensitive) → CI Gate. The Windows failure is a pre-existing, unrelated flake: 8 cli-e2e.test.ts failures about CLI stdout / fd-1 / EPIPE handling and the eval-server "No indexed repositories found" startup (cli-e2e.test.ts:1242/1259/1270/1283/1334/289) — this PR touches only Go ingestion code, with no CLI/eval-server/stdout surface, so it cannot be the cause.

Automated multi-tool digest — 6 Claude review lanes + Codex (gpt-5.5, xhigh), one independent engine; coordinator-verified against real parses. Treat as input to your judgment, not a substitute — verify before acting.

magyargergo · 2026-06-02T14:49:21Z

+  }
+}
+
+function extractSimpleTypeNameText(node: SyntaxNode): string {


[P2 · maintainability · code-read] This extractSimpleTypeNameText is byte-identical to the copy in type-binding.ts:247–257 (same qualified_type handling + the new generic_type branch). Two copies will drift independently the next time a node type needs unwrapping. captures.ts already imports from type-binding.ts (line 13), so exporting the existing definition there and importing it here removes the duplicate with zero behavior change. Non-blocking.

magyargergo · 2026-06-02T14:49:21Z

      expr.childForFieldName('type') ??
-      expr.namedChildren.find((c) => ['type_identifier', 'qualified_type'].includes(c.type)) ??
+      expr.namedChildren.find((c) =>
+        ['type_identifier', 'qualified_type', 'generic_type'].includes(c.type),


[P3 · maintainability/consistency · reproduced] This fallback now includes generic_type, but the sibling extractCompositeLiteralTypeNode (line 264, the var-spec path) still lists only ['type_identifier', 'qualified_type']. It's harmless today — var x = Box[T]{} is rescued by childForFieldName('type') (reproduced: yields {name:'b', type:'Box'}), so this fallback is effectively dead code — but the asymmetry is a latent trap and the var-form generic case currently has no test. Suggest mirroring the list here (or adding a var-form generic test). Non-blocking.

…eLiteralTypeNode fallback

…t-phase benchmark

…idance in interface detection

magyargergo · 2026-06-02T17:04:13Z

      expr.childForFieldName('type') ??
-      expr.namedChildren.find((c) => ['type_identifier', 'qualified_type'].includes(c.type)) ??
+      expr.namedChildren.find((c) =>
+        ['type_identifier', 'qualified_type', 'generic_type'].includes(c.type),


Can we move this array into a set so we won't need to recreate it in every iteration? Plus we can reuse it at line 282.

magyargergo · 2026-06-02T17:05:36Z

+    if (result.size === 0) return []; // Early exit: no struct has all required methods
  }
-  return best === undefined ? [...indexes.structsById.keys()] : [...best];
+  return result === undefined ? [...indexes.structsById.keys()] : [...result];


If we are carrying a lot of elements in either case, it will result in stackoverflow because of the spread operator.

fix(go): generic composite literal constructor inference (F33)

c90f028

evander-wang added 2 commits June 2, 2026 21:35

Merge branch 'main' into codex-go-f33-generic-composite-literals

3412d4c

chore: rebaseline Go scope-capture fingerprint for F33

57f7590

fix: skip F33 generic composite literal test in Go legacy DAG parity

5f1f998

magyargergo reviewed Jun 2, 2026

View reviewed changes

magyargergo and others added 4 commits June 2, 2026 16:01

Merge branch 'main' into codex-go-f33-generic-composite-literals

11be578

fix(go): deduplicate extractSimpleTypeNameText, align extractComposit…

b5dd441

…eLiteralTypeNode fallback

bench(go): increase interface detection tripwire to 100×100, add spli…

6c0a3f0

…t-phase benchmark

perf(go): optimize candidate pruning, arity pre-filter, and clone avo…

50b9ff6

…idance in interface detection

magyargergo requested changes Jun 2, 2026

View reviewed changes

magyargergo and others added 3 commits June 2, 2026 19:48

Merge branch 'main' into codex-go-f33-generic-composite-literals

30aec88

Merge branch 'main' into codex-go-f33-generic-composite-literals

8607223

fix(go): address F33 review follow-ups

5712d67

magyargergo approved these changes Jun 3, 2026

View reviewed changes

magyargergo merged commit 5dcffde into abhigyanpatwari:main Jun 3, 2026
30 of 31 checks passed

This was referenced Jun 3, 2026

fix(ingestion): fully-qualified nested-type identity for C++/Ruby — structure (#1978) + resolution (#1982) #1981

Merged

Go: parsing-layer coverage gaps (5 findings) #1927

Closed

magyargergo linked an issue Jun 4, 2026 that may be closed by this pull request

Go: parsing-layer coverage gaps (5 findings) #1927

Closed

5 tasks

magyargergo mentioned this pull request Jun 4, 2026

Tracking: AST/tree-sitter parsing-layer coverage audit — 92 verified findings (16 languages) #1919

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

fix(go): generic composite literal constructor inference (F33)#1976

fix(go): generic composite literal constructor inference (F33)#1976
magyargergo merged 11 commits into
abhigyanpatwari:mainfrom
evander-wang:codex-go-f33-generic-composite-literals

evander-wang commented Jun 2, 2026 •

edited by magyargergo

Loading

Uh oh!

vercel Bot commented Jun 2, 2026

Uh oh!

github-actions Bot commented Jun 2, 2026 •

edited

Loading

Uh oh!

magyargergo left a comment

Uh oh!

magyargergo Jun 2, 2026

Uh oh!

magyargergo Jun 2, 2026

Uh oh!

magyargergo Jun 2, 2026

Uh oh!

magyargergo Jun 2, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Uh oh!

Conversation

evander-wang commented Jun 2, 2026 • edited by magyargergo Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

What changed

Why

How

How to verify

Risk and rollback

Uh oh!

vercel Bot commented Jun 2, 2026

Uh oh!

github-actions Bot commented Jun 2, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

CI Report

Pipeline Status

Test Results

Code Coverage

Tests

Uh oh!

magyargergo left a comment

Choose a reason for hiding this comment

PR Tri-Review — F33 generic composite-literal constructor inference

Uh oh!

magyargergo Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

magyargergo Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

magyargergo Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

magyargergo Jun 2, 2026

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

evander-wang commented Jun 2, 2026 •

edited by magyargergo

Loading

github-actions Bot commented Jun 2, 2026 •

edited

Loading