fix(native): write dynamic_kind for sink edges in native orchestrator path

[carlos-alm]

Summary

• Root cause: insert_call_edge_rows in pipeline.rs mapped ComputedEdge → EdgeRow but silently dropped dynamic_kind. Native sink edges (eval, computed-key, unresolved-dynamic) were inserted with dynamic_kind IS NULL. The benchmark filter added in 39b9a00 (OR e.dynamic_kind IS NULL) then included these rows as false positives, breaking precision for dynamic-javascript, dynamic-typescript, and dynamic-scala.
• Rust fix: Add dynamic_kind: Option<String> to EdgeRow; update do_insert_edges to a 6-column INSERT (CHUNK 199→165 to stay under SQLite's 999 param limit); wire e.dynamic_kind.clone() through insert_call_edge_rows so the field reaches the DB.
• Benchmark fix: Add AND tgt.kind != 'file' to extractResolvedEdges — the correct semantic filter since sink edges are the only calls edges that ever target the file node, removing the dependency on dynamic_kind being set.

Test plan

• npm test passes locally (all 206 tests green, including the 4 previously failing dynamic-* precision gates)
• dynamic-javascript precision 100%, recall ≥ 75%
• dynamic-typescript precision 100%, recall ≥ 75%
• dynamic-scala precision 100%, recall 100%

[greptile-apps]

Greptile Summary

This PR fixes a silent field-drop bug in the native orchestrator path where dynamic_kind was never written to the edges table, causing sink edges to appear as false positives in the benchmark precision filter. The fix is two-pronged: a Rust-side schema change that propagates dynamic_kind through EdgeRow and do_insert_edges, and a benchmark SQL change that uses tgt.kind != 'file' as the authoritative semantic guard against sink edges.

• Rust fix (edges.rs, pipeline.rs): EdgeRow gains a 6th field dynamic_kind: Option<String>; do_insert_edges is updated to a 6-column INSERT with CHUNK reduced from 199 → 165 (165×6=990, safely under the 999 param limit); insert_call_edge_rows clones e.dynamic_kind into the EdgeRow so the value reaches SQLite.
• Benchmark fix (resolution-benchmark.test.ts): extractResolvedEdges adds AND tgt.kind != 'file' — the correct semantic definition of a sink edge — while retaining the compound confidence/dynamic_kind clause as belt-and-suspenders for databases built before this fix; an inline comment documents both guards.

Confidence Score: 5/5

Safe to merge — the change is a targeted one-field fix in the Rust native orchestrator path with no behavioural changes to the JS/WASM path, and the benchmark filter gains a semantically-correct authoritative guard.

The root-cause field drop is definitively fixed in insert_call_edge_rows; the CHUNK math is verified (165×6=990≤999); as_deref() is the correct way to bind Option<String> as a nullable SQLite text parameter; the JS path was already handling dynamic_kind via a post-insert UPDATE and is untouched; and the benchmark SQL change is semantically correct and well-commented.

No files require special attention — all three touched code locations (edges.rs, pipeline.rs, resolution-benchmark.test.ts) are clean and consistent with each other.

Important Files Changed

Filename	Overview
crates/codegraph-core/src/db/repository/edges.rs	Adds dynamic_kind: Option<String> to EdgeRow, updates INSERT to 6 columns, and reduces CHUNK from 199 to 165 (165×6=990 ≤ 999 param limit). Binding uses as_deref() correctly for Option→Option<&str>.
crates/codegraph-core/src/domain/graph/builder/pipeline.rs	One-line fix: adds dynamic_kind: e.dynamic_kind.clone() to the ComputedEdge → EdgeRow mapping in insert_call_edge_rows, closing the field-drop bug on the native orchestrator path.
tests/benchmarks/resolution/resolution-benchmark.test.ts	Adds AND tgt.kind != 'file' as the authoritative semantic guard against sink edges in extractResolvedEdges; retains the compound confidence/dynamic_kind clause as a belt-and-suspenders guard for pre-fix DBs; comment explains the rationale clearly.
CHANGELOG.md	Adds v3.15.0 release entry documenting the full dynamic-call feature set and associated bug fixes.
README.md	Adds codegraph roles --dynamic example and ignoreAdditionalDirs config snippet; minor roadmap phase renumbering.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
    participant Extractor as Rust Extractor
    participant Builder as build_edges (Rust)
    participant Pipeline as pipeline.rs
    participant EdgeRepo as edges.rs (do_insert_edges)
    participant DB as SQLite edges table

    Extractor->>Builder: "ComputedEdge { dynamic_kind: Some(eval) }"
    Builder->>Pipeline: "insert_call_edge_rows(&edges)"
    Pipeline->>Pipeline: map ComputedEdge to EdgeRow
    note over Pipeline: was: dynamic_kind dropped
    note over Pipeline: now: dynamic_kind: e.dynamic_kind.clone()
    Pipeline->>EdgeRepo: "do_insert_edges(conn, &edge_rows)"
    EdgeRepo->>DB: INSERT INTO edges (source_id, target_id, kind, confidence, dynamic, dynamic_kind)
    note over EdgeRepo,DB: 165 rows x 6 params = 990 under 999 limit

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"}}}%%
sequenceDiagram
    participant Extractor as Rust Extractor
    participant Builder as build_edges (Rust)
    participant Pipeline as pipeline.rs
    participant EdgeRepo as edges.rs (do_insert_edges)
    participant DB as SQLite edges table

    Extractor->>Builder: "ComputedEdge { dynamic_kind: Some(eval) }"
    Builder->>Pipeline: "insert_call_edge_rows(&edges)"
    Pipeline->>Pipeline: map ComputedEdge to EdgeRow
    note over Pipeline: was: dynamic_kind dropped
    note over Pipeline: now: dynamic_kind: e.dynamic_kind.clone()
    Pipeline->>EdgeRepo: "do_insert_edges(conn, &edge_rows)"
    EdgeRepo->>DB: INSERT INTO edges (source_id, target_id, kind, confidence, dynamic, dynamic_kind)
    note over EdgeRepo,DB: 165 rows x 6 params = 990 under 999 limit

Loading

_{Reviews (3): Last reviewed commit: "fix(benchmark): add comment explaining c..." | Re-trigger Greptile}

[greptile-apps]

Redundant filter after Rust fix

With dynamic_kind now correctly written to the DB for sink edges, the new AND tgt.kind != 'file' guard already excludes all sink edges semantically. The residual third clause OR e.dynamic_kind IS NULL was the original root-cause condition and is now effectively dead for correctly-built databases — any edge with confidence = 0 and tgt.kind != 'file' that also has dynamic_kind IS NULL would be a DB anomaly rather than a sink edge.

Both filters together are harmless and the belt-and-suspenders approach is reasonable as defence-in-depth, but a brief comment clarifying why the compound condition is kept (e.g., for databases built before the Rust fix) would prevent future confusion.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[carlos-alm]

Added a 4-line comment above the two guards explaining: (1) tgt.kind != 'file' is the authoritative semantic guard for sink edges (the only calls edges that target file nodes), and (2) the compound confidence/dynamic_kind clause is kept as belt-and-suspenders for databases built before the Rust fix in #1698 where dynamic_kind was NULL for sink edges.

[carlos-alm]

@greptileai