fix(native): stop scoping full-build dataflow-vertex extraction to files with edges

[carlos-alm]

Summary

runDataflowVertexPass's full-build file selection (the P6 pass that backfills dataflow_vertices after the Rust orchestrator's edge-only dataflow pass) only included native-language files that already had dataflow EDGE rows (flows_to/returns/mutates) — on the theory that a file with no edges would produce zero vertices anyway.

That theory is wrong: a plain leaf function with params and a return but no calls to or from any other function has zero inter-procedural edges, yet still has vertex-worthy params/returns — extractDataflowAnalysis's vertex output isn't gated on argFlows/assignments/mutations being non-empty. In practice this meant the native engine silently dropped dataflow_vertices for most files in a typical codebase (confirmed empirically: bind-call-apply.js, class-scope.js, define-property.js, define-property-accessor.js in the javascript fixture, plus most of hierarchy.ts/index.ts/serializer.ts in typescript and several files in pts-javascript, all had zero native-engine vertices while WASM recorded them correctly), while WASM has always recorded vertices unconditionally.

Fix: the filter now includes any native-language file with at least one function/method definition (checked via the nodes table, kind IN CALLABLE_SYMBOL_KINDS) — which is what actually determines whether a file can produce vertices, not edge presence.

Verification

node scripts/parity-compare.mjs --dataflow across the full 34-language fixture suite:

• Before: javascript, typescript, pts-javascript all diverged on df-vertices (17/20/6 missing vertex rows respectively)
• After: 42/42 fixtures OK on df-vertices — the one remaining jelly-micro divergence is a pre-existing, unrelated technique-label mismatch on calls edges (0 node/edge/df-vertex diffs beyond that), filed separately as bug(parity): native engine tags inherited-method call edges as ts-native instead of cha (jelly-micro classes/classes2) #2572

The originally-filed issue's own note ("--dataflow requires migration v18 — worth checking whether the test DB has that migration applied") turned out not to be the cause — this was a genuine extraction-scoping bug, not a migration issue.

Test plan

• New integration test tests/integration/issue-2483-native-dataflow-vertices-leaf-functions.test.ts — a real dual-function leaf-function fixture built end-to-end through buildGraph with the native engine, confirming both functions' param/return vertices are recorded
• Revert-verified: disabling the new filter condition reproduces the pre-fix [] (zero vertices) for both new tests
• Existing dataflow integration tests (dataflow-vertices, dataflow-p4-native, dataflow-incremental) still pass (28 passed)
• npx tsc --noEmit -p ., npm run lint, full npm test (5487 passed)
• Full parity-compare.mjs --dataflow run across all 34 language fixtures (42/42 OK on df-vertices)

Closes #2483

[greptile-apps]

Greptile Summary

The PR corrects native full-build dataflow-vertex selection by processing files containing callable definitions rather than requiring existing inter-procedural dataflow edges.

• Adds callable-kind-based full-build filtering for native files.
• Avoids unnecessary vertex extraction during empty incremental passes.
• Adds native integration coverage for leaf-function vertices and no-op incremental preservation.

Confidence Score: 5/5

The PR appears safe to merge.

No blocking failure remains.

Important Files Changed

Filename	Overview
src/domain/graph/builder/stages/native-orchestrator.ts	Broadens full-build vertex extraction to native files with callable definitions and explicitly skips empty incremental work.
tests/integration/issue-2483-native-dataflow-vertices-leaf-functions.test.ts	Adds end-to-end native coverage for leaf-function vertices and their preservation across a no-op incremental rebuild.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  Build["Native graph build"] --> Mode{"Full build?"}
  Mode -->|Yes| Query["Select files containing callable nodes"]
  Mode -->|No, changed files| Changed["Select changed files"]
  Mode -->|No changes| Skip["Skip vertex pass"]
  Query --> Extract["Extract native dataflow analysis"]
  Changed --> Extract
  Extract --> Persist["Persist parameter, return, and local vertices"]

Loading

_{Reviews (2): Last reviewed commit: "fix: skip dataflow-vertex full-build sca..." | Re-trigger Greptile}

[github-actions]

Codegraph Impact Analysis

2 functions changed → 5 callers affected across 4 files

• runDataflowVertexPass in src/domain/graph/builder/stages/native-orchestrator.ts:323 (4 transitive callers)
• tryNativeOrchestrator in src/domain/graph/builder/stages/native-orchestrator.ts:2829 (4 transitive callers)

[carlos-alm]

@greptileai