fix(vue): F89 JSDoc fix, F90 dual-script merge, F92 lang plumbing (#1936)

[prajapatisparsh]

F89: JSDoc corrected (0-based → 1-based, behavior unchanged)
F90: merge <script> + <script setup> blocks (was: only setup)
F92: plumb lang through to select JS vs TS captures

• lang="js" → JavaScript grammar/captures
• lang="ts"|"tsx"|"jsx"|"" → TypeScript grammar/captures
  F91 done via feat(vue): migrate Vue SFC to scope-based resolution (RFC #909 Ring 3, closes #940) #1950
  29 unit tests pass, 1 integration test suite blocked by pre-existing tree-sitter-dart import

[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]

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
10605	10589	0	16	580s

✅ All 10589 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	74.3%	34471/46389	N/A%	—	🟢 ██████████████░░░░░░
Branches	62.28%	21337/34257	N/A%	—	🟢 ████████████░░░░░░░░
Functions	79.6%	3692/4638	N/A%	—	🟢 ███████████████░░░░░
Lines	78.16%	31216/39938	N/A%	—	🟢 ███████████████░░░░░

---

_{📋 View full run · Generated by CI}

[magyargergo]

PR #2050 tri-review — Vue SFC dual-script merge & JS/TS routing

Methods (3): GitNexus reviewer swarm + Compound-Engineering personas (both Claude) and Codex (the one independent engine). Engine breakdown: 8 Claude lanes — correctness, adversarial, maintainability, testing, risk, test/CI, plus a Vue-SFC expert and a TypeScript/tree-sitter expert added at your request (risk & test/CI ended mid-investigation; their domains were covered by the other lanes) — and Codex live. Codex ran as a foreground, not-prompted-from-Claude-output call and independently confirmed all three headline findings. Note on independence: the two Claude methods share priors, so Claude-only agreement is "consistent across personas," not independent confirmation; the strong signal here is Codex + a Claude lane + a reproduction.

What's solid (validated — credit where due): the extractVueScript merge logic and the new lang field work as intended on the sequential (<15-file) path, and the extractor unit tests pass. Two suspected problems were refuted and excluded (validation is a feature): (a) feeding a TS-grammar cached tree into emitJsScopeCaptures does not corrupt captures — it is guarded by jsCachedTreeMatchesGrammar and re-parsed; (b) the F89 lineOffset doc change is accurate — countNewlines(precedingText)+1 is genuinely 1-based and the consumer math (+ node 0-based row) is correct, no off-by-one introduced.

Headline (P1 — reproduced)

For a .vue with an explicit <script lang="js"> block followed by a <script setup lang="ts"> block, extractVueScript merges both and returns lang = blocks[0].lang = 'js', so emitVueScopeCaptures routes the combined JS+TS text to the JavaScript grammar. Reproduced with the repo's pinned grammars on that merged content: tree-sitter-javascript → 1 ERROR + 1 missing node, the interface fails to parse; tree-sitter-typescript → 0 errors, clean. This is a regression — before this PR only the <script setup> (TS) block was extracted and parsed as TS. (Trigger needs an explicit lang="js"; an absent lang routes to TS and is safe.) Codex and the adversarial lane independently confirmed/reproduced it.

Inline findings

• [P1] captures.ts:74 / vue-sfc-extractor.ts:269 — mixed-lang merged content routed to one grammar via blocks[0].lang (the headline above).
• [P2] captures.ts:62-63 — setupLang is read but never set by any caller (the bridge passes only { sourceKind }; the LanguageProvider contract type omits it; repo-wide grep finds zero setters). So in worker mode (≥15 files — the large-repo path this branch targets) every .vue routes to emitTsScopeCaptures, silently losing the JS-only passes (CJS require() decomposition, JSDoc type-bindings). Not a crash and not a regression vs base — but F92's routing fix doesn't reach the worker path, and the comment "the parse worker passes langs via sourceMeta.setupLang" is actively misleading.
• [P2] vue-sfc-extractor.ts:266 — lineOffset = blocks[0].lineOffset uses source order while combinedContent is built from ordered (non-setup first); divergent when <script setup> precedes <script>, and the single-newline join collapses the inter-block gap so the second block's lines are off regardless. Affects user-visible line numbers (jump-to-line / call-site annotations), not graph topology.
• [P2/P3] new fixtures — vue-dual-script/App.vue and vue-js-lang/App.vue are referenced by no test and lack the src/ mini-repo layout the integration suite uses; their natural home is test/integration/resolvers/vue-scope.test.ts (which currently runs only vue-composition-api / vue-options-api / vue-cross-file).

Lower-priority / notes

• Absent lang="" routes to TS even though Vue's default script lang is JS — harmless (TS is a superset) but plain <script> JS also misses the JS-only passes; consider routing '' | 'js' | 'jsx' → JS.
• lang="jsx"/"tsx" (newly advertised in the doc) route to the non-JSX TS grammar (grammar is chosen by the .vue extension) → JSX would ERROR-node. Pre-existing; consider trimming the doc claim.
• Stale doc: languages/vue/index.ts:40-42 still says "only <script setup> is processed… the non-setup block is skipped" — contradicts the new merge (that file is not in this diff).
• Comment at vue-sfc-extractor.ts:267-268 ("the template compiler will catch it") — GitNexus never runs the Vue template compiler; mismatched-lang blocks are silently mis-parsed here.

Test gaps

No test exercises the worker (pre-extracted-script) routing, the full-pipeline JS-lang or dual-script paths (the new fixtures would, if wired), <script setup>-before-<script> ordering, or symbol line numbers for merged blocks. The merge test asserts presence (toContain) but not order.

CI: all green except Vercel (deploy authorization, not a code issue) — typecheck/lint/tests/coverage pass. Coverage of this review: the full 5-file diff was read and the headline reproduced; the PR head moved mid-review (a main merge bringing in unrelated MCP work, #2049) but no Vue file changed, so these findings apply to the current head.

Automated multi-tool digest (GitNexus swarm + Compound-Engineering personas + Codex). Verify before acting; inline anchors mark the most load-bearing lines.

[magyargergo]

[P2/P3 — testing, code-read] This fixture (and vue-dual-script/App.vue) is referenced by no test — a repo-wide grep finds zero consumers. The integration suite test/integration/resolvers/vue-scope.test.ts loads fixtures by explicit path and runs only vue-composition-api / vue-options-api / vue-cross-file; these new dirs also lack the src/ mini-repo layout those use. So the JS-lang routing and the dual-script merge are covered only by the pure-extractVueScript unit tests — never end-to-end, and never on the worker path (which would expose the dead setupLang).

Actionable path: add describe blocks to vue-scope.test.ts that run runPipelineFromRepo against these fixtures and assert symbols from both blocks / JS-style captures appear — or remove the fixtures. [code-read]

[prajapatisparsh]

on it

[magyargergo]

There are two failures

  
  ⎯⎯⎯⎯⎯⎯⎯ Failed Tests 2 ⎯⎯⎯⎯⎯⎯⎯
  
   FAIL   default  test/integration/resolvers/vue-scope.test.ts > F90 — dual-script (<script> + <script setup>) > includes symbols from both the non-setup and setup blocks
  AssertionError: expected [] to include 'ref'
   ❯ test/integration/resolvers/vue-scope.test.ts:490:19
      488|   it('includes symbols from both the non-setup and setup blocks', () =…
      489|     const funcs = getNodesByLabel(result, 'Function');
      490|     expect(funcs).toContain('ref'); // from <script setup> import
         |                   ^
      491|   });
      492| });
  
  ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[1/2]⎯
  
   FAIL   default  test/integration/resolvers/vue-scope.test.ts > F92 — JS-lang script block > parses <script lang="js"> content
  AssertionError: expected [] to include 'createApp'
   ❯ test/integration/resolvers/vue-scope.test.ts:507:19
      505|   it('parses <script lang="js"> content', () => {
      506|     const funcs = getNodesByLabel(result, 'Function');
      507|     expect(funcs).toContain('createApp');
         |                   ^
      508|   });
      509| });
  
  ⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯⎯[2/2]⎯
  
  
   Test Files  1 failed | 429 passed | 8 skipped (438)
        Tests  2 failed | 10230 passed | 16 skipped (10248)
     Start at  08:54:17