Skip to content

fix(native): type map confidence-aware dedup for engine parity #885

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
carlos-alm merged 6 commits into from
Apr 6, 2026
Merged

fix(native): type map confidence-aware dedup for engine parity #885

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
6 commits
Select commit Hold shift + click to select a range
0486cad
fix(native): type map confidence-aware dedup for engine parity
carlos-alm Apr 6, 2026
0fc6559
feat(ci): gate release workflow on resolution precision/recall thresh…
carlos-alm Apr 6, 2026
5b147bf
style: fix biome formatting in regression guard
carlos-alm Apr 6, 2026
062e85e
fix: remove invalid #[napi(default = 0.9)] attributes causing Rust co…
carlos-alm Apr 6, 2026
0cc4360
fix: clarify dedup comment per review — no-op for Map path, needed fo…
carlos-alm Apr 6, 2026
3a0db09
Merge branch 'main' into fix/native-typemap-parity
carlos-alm Apr 6, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions .github/workflows/benchmark.yml
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -106,6 +106,10 @@ jobs:
fi
node $STRIP_FLAG --import ./scripts/ts-resolve-loader.js scripts/resolution-benchmark.ts $ARGS > resolution-result.json

- name: Gate on resolution thresholds
if: steps.existing.outputs.skip != 'true'
run: npx vitest run tests/benchmarks/resolution/resolution-benchmark.test.ts --reporter=verbose

- name: Merge resolution into build result
if: steps.existing.outputs.skip != 'true'
run: |
Expand Down
2 changes: 1 addition & 1 deletion Cargo.lock
View file
Open in desktop

Some generated files are not rendered by default. Learn more about how customized files appear on GitHub.

1 change: 1 addition & 0 deletions crates/codegraph-core/src/build_pipeline.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -877,6 +877,7 @@ fn build_and_insert_call_edges(
.map(|t| TypeMapInput {
name: t.name.clone(),
type_name: t.type_name.clone(),
confidence: t.confidence,
})
.collect();

Expand Down
30 changes: 22 additions & 8 deletions crates/codegraph-core/src/edge_builder.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -57,6 +57,8 @@ pub struct TypeMapInput {
pub name: String,
#[napi(js_name = "typeName")]
pub type_name: String,
/// Confidence: 0.9 = type annotation, 1.0 = constructor, 0.7 = factory.
pub confidence: f64,
}

#[napi(object)]
Expand Down Expand Up @@ -153,10 +155,22 @@ fn process_file<'a>(
.map(|im| (im.name.as_str(), im.file.as_str()))
.collect();

let type_map: HashMap<&str, &str> = file_input
.type_map.iter()
.map(|tm| (tm.name.as_str(), tm.type_name.as_str()))
.collect();
// Build type map keeping the highest-confidence entry per name
// (first-wins on tie), matching the JS setTypeMapEntry behaviour.
let mut type_map: HashMap<&str, (&str, f64)> = HashMap::new();
for tm in &file_input.type_map {
let entry = type_map.entry(tm.name.as_str());
match entry {
std::collections::hash_map::Entry::Vacant(e) => {
e.insert((tm.type_name.as_str(), tm.confidence));
}
std::collections::hash_map::Entry::Occupied(mut e) => {
if tm.confidence > e.get().1 {
e.insert((tm.type_name.as_str(), tm.confidence));
}
}
}
}

let file_nodes: Vec<&NodeInfo> = all_nodes.iter().filter(|n| n.file == *rel_path).collect();
let defs_with_ids: Vec<DefWithId> = file_input.definitions.iter().map(|d| {
Expand Down Expand Up @@ -210,7 +224,7 @@ fn resolve_call_targets<'a>(
call: &CallInfo,
rel_path: &str,
imported_from: Option<&str>,
type_map: &HashMap<&str, &str>,
type_map: &HashMap<&str, (&str, f64)>,
) -> Vec<&'a NodeInfo> {
// 1. Import-aware resolution
if let Some(imp_file) = imported_from {
Expand All @@ -236,7 +250,7 @@ fn resolve_call_targets<'a>(

// 4. Type-aware resolution via receiver → type map
if let Some(ref receiver) = call.receiver {
if let Some(type_name) = type_map.get(receiver.as_str()) {
if let Some(&(type_name, _conf)) = type_map.get(receiver.as_str()) {
let qualified = format!("{}.{}", type_name, call.name);
let typed: Vec<&NodeInfo> = ctx.nodes_by_name
.get(qualified.as_str())
Expand Down Expand Up @@ -296,15 +310,15 @@ fn emit_call_edges(
/// Emit a receiver edge from caller to the receiver's type node (if applicable).
fn emit_receiver_edge(
ctx: &EdgeContext, call: &CallInfo, caller_id: u32, rel_path: &str,
type_map: &HashMap<&str, &str>,
type_map: &HashMap<&str, (&str, f64)>,
seen_edges: &mut HashSet<u64>, edges: &mut Vec<ComputedEdge>,
) {
let Some(ref receiver) = call.receiver else { return };
if ctx.builtin_set.contains(receiver.as_str())
|| receiver == "this" || receiver == "self" || receiver == "super"
{ return; }

let effective_receiver = type_map.get(receiver.as_str()).copied().unwrap_or(receiver.as_str());
let effective_receiver = type_map.get(receiver.as_str()).map(|&(t, _)| t).unwrap_or(receiver.as_str());
let type_resolved = effective_receiver != receiver.as_str();

let samefile = ctx.nodes_by_name_and_file
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/c.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,7 @@ fn match_c_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _dept
symbols.type_map.push(TypeMapEntry {
name: final_name,
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -55,6 +56,7 @@ fn match_c_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _dept
symbols.type_map.push(TypeMapEntry {
name,
type_name: node_text(&type_node, source).to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/cpp.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@ fn match_cpp_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _de
symbols.type_map.push(TypeMapEntry {
name: final_name,
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -54,6 +55,7 @@ fn match_cpp_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _de
symbols.type_map.push(TypeMapEntry {
name,
type_name: node_text(&type_node, source).to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/csharp.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -428,6 +428,7 @@ fn match_csharp_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols,
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -445,6 +446,7 @@ fn match_csharp_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols,
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
1 change: 1 addition & 0 deletions crates/codegraph-core/src/extractors/go.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -330,6 +330,7 @@ fn collect_go_typed_identifiers(node: &Node, source: &[u8], type_map: &mut Vec<T
type_map.push(TypeMapEntry {
name: node_text(&child, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/java.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -39,6 +39,7 @@ fn match_java_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _d
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -54,6 +55,7 @@ fn match_java_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _d
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
8 changes: 5 additions & 3 deletions crates/codegraph-core/src/extractors/javascript.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -58,23 +58,24 @@ fn match_js_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _dep
if let Some(name_n) = node.child_by_field_name("name") {
if name_n.kind() == "identifier" {
let var_name = node_text(&name_n, source);
// Type annotation takes priority
// Type annotation: confidence 0.9
if let Some(type_anno) = find_child(node, "type_annotation") {
if let Some(type_name) = extract_simple_type_name(&type_anno, source) {
symbols.type_map.push(TypeMapEntry {
name: var_name.to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
return; // Skip new_expression check — annotation wins
}
}
// Fall back to new expression inference
// Constructor: confidence 1.0 (overrides annotation in edge builder)
if let Some(value_n) = node.child_by_field_name("value") {
if value_n.kind() == "new_expression" {
if let Some(type_name) = extract_new_expr_type_name(&value_n, source) {
symbols.type_map.push(TypeMapEntry {
name: var_name.to_string(),
type_name: type_name.to_string(),
confidence: 1.0,
});
}
}
Expand All @@ -93,6 +94,7 @@ fn match_js_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _dep
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/kotlin.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -32,6 +32,7 @@ fn match_kotlin_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols,
symbols.type_map.push(TypeMapEntry {
name,
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -44,6 +45,7 @@ fn match_kotlin_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols,
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: node_text(&type_node, source).to_string(),
confidence: 0.9,
});
}
}
Expand Down
1 change: 1 addition & 0 deletions crates/codegraph-core/src/extractors/php.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -396,6 +396,7 @@ fn match_php_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _de
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/python.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -335,6 +335,7 @@ fn match_python_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols,
symbols.type_map.push(TypeMapEntry {
name: name.to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -352,6 +353,7 @@ fn match_python_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols,
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/rust_lang.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -394,6 +394,7 @@ fn match_rust_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _d
symbols.type_map.push(TypeMapEntry {
name: node_text(&pattern, source).to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -410,6 +411,7 @@ fn match_rust_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _d
symbols.type_map.push(TypeMapEntry {
name: name.to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
2 changes: 2 additions & 0 deletions crates/codegraph-core/src/extractors/scala.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -31,6 +31,7 @@ fn match_scala_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _
symbols.type_map.push(TypeMapEntry {
name: node_text(&pat, source).to_string(),
type_name: node_text(&type_node, source).to_string(),
confidence: 0.9,
});
}
}
Expand All @@ -43,6 +44,7 @@ fn match_scala_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _
symbols.type_map.push(TypeMapEntry {
name: node_text(&name_node, source).to_string(),
type_name: node_text(&type_node, source).to_string(),
confidence: 0.9,
});
}
}
Expand Down
1 change: 1 addition & 0 deletions crates/codegraph-core/src/extractors/swift.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -38,6 +38,7 @@ fn match_swift_type_map(node: &Node, source: &[u8], symbols: &mut FileSymbols, _
symbols.type_map.push(TypeMapEntry {
name: name.to_string(),
type_name: type_name.to_string(),
confidence: 0.9,
});
}
}
Expand Down
3 changes: 3 additions & 0 deletions crates/codegraph-core/src/types.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -290,6 +290,9 @@ pub struct TypeMapEntry {
pub name: String,
#[napi(js_name = "typeName")]
pub type_name: String,
/// Confidence: 0.9 = type annotation, 1.0 = constructor, 0.7 = factory.
/// Used to resolve conflicts when the same name appears multiple times.
pub confidence: f64,
}

#[napi(object)]
Expand Down
33 changes: 33 additions & 0 deletions scripts/update-benchmark-report.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -320,6 +320,39 @@ if (prev) {
}
}

// ── Resolution regression detection ─────────────────────────────────────
// Resolution metrics are "higher is better" — warn when they DROP.
const PRECISION_DROP_THRESHOLD = 0.05; // warn if precision drops >5pp
const RECALL_DROP_THRESHOLD = 0.10; // warn if recall drops >10pp

if (prev && latest.resolution && prev.resolution) {
for (const lang of Object.keys(latest.resolution)) {
const cur = latest.resolution[lang];
const prv = prev.resolution?.[lang];
if (!cur || !prv) continue;

const precDrop = prv.precision - cur.precision;
if (precDrop > PRECISION_DROP_THRESHOLD) {
const msg = `[resolution] ${lang} precision: ${(prv.precision * 100).toFixed(1)}% → ${(cur.precision * 100).toFixed(1)}% (−${(precDrop * 100).toFixed(1)}pp, threshold ${(PRECISION_DROP_THRESHOLD * 100).toFixed(0)}pp)`;
if (process.env.GITHUB_ACTIONS) {
console.error(`::warning title=Resolution Regression::${msg}`);
} else {
console.error(`⚠ REGRESSION: ${msg}`);
}
}

const recDrop = prv.recall - cur.recall;
if (recDrop > RECALL_DROP_THRESHOLD) {
const msg = `[resolution] ${lang} recall: ${(prv.recall * 100).toFixed(1)}% → ${(cur.recall * 100).toFixed(1)}% (−${(recDrop * 100).toFixed(1)}pp, threshold ${(RECALL_DROP_THRESHOLD * 100).toFixed(0)}pp)`;
if (process.env.GITHUB_ACTIONS) {
console.error(`::warning title=Resolution Regression::${msg}`);
} else {
console.error(`⚠ REGRESSION: ${msg}`);
}
}
}
}

// ── Patch README.md ──────────────────────────────────────────────────────
if (fs.existsSync(readmePath)) {
let readme = fs.readFileSync(readmePath, 'utf8');
Expand Down
15 changes: 14 additions & 1 deletion src/domain/graph/builder/stages/build-edges.ts
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -342,7 +342,7 @@ function buildCallEdgesNative(
if (!fileNodeRow) continue;

const importedNames = buildImportedNamesForNative(ctx, relPath, symbols, rootDir);
const typeMap: Array<{ name: string; typeName: string; confidence: number }> =
const typeMapRaw: Array<{ name: string; typeName: string; confidence: number }> =
symbols.typeMap instanceof Map
? [...symbols.typeMap.entries()].map(([name, entry]) => ({
name,
Expand All @@ -352,6 +352,19 @@ function buildCallEdgesNative(
: Array.isArray(symbols.typeMap)
? (symbols.typeMap as Array<{ name: string; typeName: string; confidence: number }>)
: [];
// Deduplicate: keep highest-confidence entry per name (first-wins on tie),
// matching JS setTypeMapEntry semantics. The Map branch is already
// deduped by setTypeMapEntry — this loop is only needed for the Array
// branch (pre-rebuilt native addon) but runs unconditionally as
// belt-and-suspenders since it's a cheap O(n) pass.
const typeMapDedup = new Map<string, { name: string; typeName: string; confidence: number }>();
for (const entry of typeMapRaw) {
const existing = typeMapDedup.get(entry.name);
if (!existing || entry.confidence > existing.confidence) {
typeMapDedup.set(entry.name, entry);
}
}
const typeMap = [...typeMapDedup.values()];
Comment on lines 345 to +367
Copy link
Copy Markdown
Contributor

@greptile-apps greptile-apps bot Apr 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

P2 Dedup is a no-op for the Map path

When symbols.typeMap instanceof Map (the WASM/JS extraction path), setTypeMapEntry already enforces highest-confidence-wins at insertion time, so the resulting array has no duplicate names — the dedup loop adds no correctness value for that branch. The fix is only needed for the Array.isArray branch. Consider scoping the dedup to that branch, or add a comment clarifying this is intentional belt-and-suspenders for pre-rebuilt-addon builds.

Copy link
Copy Markdown
Contributor Author

@carlos-alm carlos-alm Apr 6, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Fixed — expanded the comment to clarify that the dedup is a no-op for the Map path (already deduped by setTypeMapEntry) and only needed for the Array branch (pre-rebuilt native addon). Kept it unconditional as intentional belt-and-suspenders since it's a cheap O(n) pass.

nativeFiles.push({
file: relPath,
fileNodeId: fileNodeRow.id,
Expand Down
Loading
Loading