refactor: decompose runPerfBenchmarks in token-benchmark.ts#1908
Merged
Conversation
BATCH_COMMANDS entries with sig: 'dbOnly' (currently only complexity) always wrote their target into opts.target. complexityData treats opts.target as a symbol-name filter and opts.file as the file-path filter, so batch complexity <file> never matched anything and the function list silently fell back to empty with a whole-repo summary regardless of which file was requested. Add an optional targetKey on BatchCommandEntry so each dbOnly command can declare which opts key its targets map to, and set it to 'file' for complexity. batchData and multiBatchData now route the target through the declared key instead of assuming opts.target. Fixes #1721 Impact: 4 functions changed, 6 affected
…ssification codegraph roles --role dead flagged every function parameter as dead-leaf and every interface/type member as dead-unresolved, regardless of actual usage. Root cause: the role classifier's fan-in-based "no callers = dead" heuristic is meaningless for these two symbol categories, since neither can ever have inbound call edges by construction: - Parameters: `kind = 'parameter'` nodes were unconditionally force-assigned dead-leaf via a fast-path bypass in classifyNodeRolesFull/Incremental (TS) and do_classify_full/do_classify_incremental (native), before any fan-in was even computed. A parameter's liveness is a local dataflow question (is it referenced within its own function body), not a call-graph reachability question, so this produced ~90% noise in --role dead output. - Interface/type members: every language extractor qualifies interface/type members as `Owner.member` top-level definitions (mirroring class method qualification), and they never receive inbound call edges (they're consumed via type annotations, not calls). Lacking any recognition of this, they fell through the normal fan-in/fan-out path and landed in dead-unresolved. Fix: - Parameters are now fully excluded from role classification (role stays NULL), the same treatment already given to file/directory nodes. - Interface/type members are now recognized in the classifier by resolving the Owner.member prefix against same-file TYPE_DEF_KINDS declarations (interface/type/struct/enum/trait/record) and classified `leaf` unconditionally. Class methods use the identical Owner.member naming convention but are unaffected since `class` is not in TYPE_DEF_KINDS, so real dead methods/functions remain detected. Applied to both the TS/WASM classifier (graph/classifiers/roles.ts, features/structure.ts) and the native Rust classifier (graph/classifiers/roles.rs) per the dual-engine parity requirement; verified both engines produce identical results end-to-end against this repo's own graph. Fixes #1723 Impact: 6 functions changed, 5 affected
codegraph exports <file> --json (and the --unused dead-export filter it
feeds) reported zero consumers for interfaces/types that are demonstrably
imported and used elsewhere via `import type { X }`, even though
`codegraph deps <file>` correctly showed the importing file in
`importedBy`. Example: ChaContext in src/domain/graph/builder/cha.ts is
imported (type-only) by build-edges.ts and native-orchestrator.ts, but
`codegraph exports` showed consumerCount: 0.
Root cause: exportsFileImpl's per-symbol consumers query
(domain/analysis/exports.ts) only looked at kind = 'calls' edges. The
builder already emits a symbol-level `imports-type` edge for `import
type { X }` statements (source = importing file node, target = the
specific imported symbol -- see emitTypeOnlySymbolEdges in
build-edges.ts/incremental.ts), which `codegraph deps` reads from, but
the exports consumer query never looked at this edge kind. Role
classification (features/structure.ts) already includes 'imports-type'
in its fan-in formula, so `codegraph roles --role dead` was unaffected --
this was purely a gap in exports's independent consumer list.
Fix: widen the consumers query to `kind IN ('calls', 'imports-type')`,
matching the edge-kind set already used everywhere else in the codebase
for cross-file usage credit (structure.ts, graph-enrichment.ts,
boundaries.ts, dependencies.ts). No native Rust changes needed --
domain/analysis/exports.ts is pure query-layer code that reads the
already-built edges table and has no engine-specific mirror.
Deliberately does NOT add `extends`/`implements`: investigation found
those edges are resolved by symbol name only, with no file/import
scoping (buildClassHierarchyEdges in both build-edges.ts/incremental.ts
and the native emit_hierarchy_edges), so they link same-named
declarations across unrelated files (verified: this repo's own graph has
false `implements`/`extends` edges between unrelated fixture classes
across languages and a `Repository` interface in src/types.ts). Filed
as #1812. Also filed #1813 for a related but distinct gap: `import {
type X }` inline per-specifier modifiers aren't tracked as type-only in
either engine's extractor, so such X still gets no credit even after
this fix.
Added tests/integration/exports.test.ts coverage: an interface consumed
only via a symbol-level `imports-type` edge gets consumerCount >= 1 and
is excluded from --unused, while a genuinely unreferenced interface
still shows 0 consumers.
Verified against this repo's own graph: `codegraph exports
src/domain/graph/builder/cha.ts -T --json` now credits ChaContext with
2 consumers (build-edges.ts, native-orchestrator.ts). Full test suite
(201 files, 3359 tests) and lint pass clean.
docs check acknowledged: internal bug fix to exports consumer
computation, no new feature/language/CLI surface/architecture change --
README.md, CLAUDE.md, and ROADMAP.md are unaffected.
Fixes #1724
Impact: 1 functions changed, 3 affected
The CLI's `-f/--file` option is a repeatable Commander accumulator (collectFile) that always produces a string[], even on first use. The native composite bindings backing fn-impact (findNodesWithFanIn) and query (fnDeps) forwarded that array straight into napi-rs functions whose Rust signatures only accepted a single String, so any use of -f/--file crashed with "Failed to convert JavaScript value ... into rust type `String`" regardless of engine defaults. context worked only because its code path never touches the native repository for symbol lookup. Widen the native Rust signatures (find_nodes_with_fan_in, fn_deps) to accept Vec<String> and build an OR-of-LIKE clause for multiple files, mirroring buildFileConditionSQL/NodeQuery.fileFilter on the JS side. Normalize the file option before calling into the native binding on the TS side, and thread the widened string | string[] type through QueryOpts, fnDeps's opts, and the call chain down to findMatchingNodes. This also makes repeated -f/--file genuinely multi-file end to end for fn-impact/query, matching context's existing behavior, instead of crashing on any use. findNodesByScope shares QueryOpts but has no CLI caller and its native binding still only accepts one file; it now takes the first value defensively rather than crash or fail to type-check. Fixes #1726 docs check acknowledged: pure bug fix, no new CLI options/languages/ architecture — the -f/--file "repeatable" docs are now accurate rather than needing correction. Impact: 14 functions changed, 34 affected
… exports Two compounding root causes made `where --file`'s `exported` array (and `codegraph exports`) unreliable for entire classes of exports: 1. The JS/TS export-statement handler (WASM query path, WASM walk path, and the mirrored native Rust extractor) only recognized `export function`, `export class`, `export interface`, and `export type` declarations. It never matched `lexical_declaration`/`variable_declaration`, so `export const/let/var …` was never added to the extractor's export list — regardless of the initializer's shape — leaving the `exported=1` DB column permanently unset for every exported constant in the codebase (including ones that happened to look "correct", like a `new Set(...)`-initialized constant referenced as a call argument elsewhere). 2. `where --file`'s `exported` list ignored the `exported` DB column entirely and computed membership from `findCrossFileCallTargets` (symbols targeted by a cross-file `calls` edge) — a heuristic that only "worked" by coincidence for symbols that happened to be passed as call arguments elsewhere. `codegraph exports` already had the right idea (prefer the `exported` column, fall back to the heuristic only for pre-migration DBs) but duplicated that logic locally instead of sharing it. Fixes: - src/extractors/javascript.ts: extract the function/class/interface/type kind map plus a new lexical/variable-declaration branch into a shared `collectExportedDeclarations`, used by both `handleExportCapture` (query path) and `handleExportStmt` (walk path) so they can't drift apart again. Declarator values are classified with the same predicate already used to build the matching Definition (function-valued -> kind 'function', literal/array/object/new-expression-valued `const` -> kind 'constant'). - crates/codegraph-core/src/extractors/javascript.rs: mirror the same fix in `handle_export_declaration` via a new `collect_exported_var_declarations`. - src/db/repository/nodes.ts: add `findExportedNodesByFile`, the single shared implementation of "prefer exported=1, fall back to cross-file calls for pre-migration DBs" — re-exported through db/repository/index.ts and db/index.ts. - src/domain/analysis/exports.ts: use the shared helper instead of a locally duplicated hasExportedCol/exportedNodesStmt implementation. - src/domain/analysis/symbol-lookup.ts: `whereFileImpl` now uses the shared helper instead of `findCrossFileCallTargets` directly, so `where --file` and `exports` agree on what "exported" means. Tests: - tests/parsers/javascript.test.ts: unit coverage for bare-literal, new Set(...), object-literal-with-methods, and arrow-function exported consts, plus a non-exported-const negative case and a function/class regression guard for the shared-helper refactor. - tests/engines/query-walk-parity.test.ts, tests/engines/parity.test.ts: cross-path/cross-engine regression cases for the same shapes. - tests/integration/queries.test.ts: new fixture node with exported=1 but zero incoming edges of any kind, proving `where --file`'s exported list no longer depends on cross-file call presence. No user-facing command/language/architecture changes, so README/CLAUDE.md/ ROADMAP.md are unaffected — docs check acknowledged. Fixes #1728 Impact: 8 functions changed, 17 affected
tree-sitter-typescript's predefined_type production (string, number, boolean, ... primitive type keywords) lexes its keyword as an anonymous grammar token whose type string is identical to the named `string` node type used for real string literals. Both engines matched by node type alone, so `codegraph ast --kind string` spuriously matched bare `string` type annotations (interface fields, parameter types, return types) as if they were string literals. Add an isNamed/is_named() guard scoped to the JS/TS/TSX family on both engines: WASM via astRequiresNamedNode() in ast-analysis/rules/index.ts feeding the shared resolveAstKind() in ast-store-visitor.ts, native via a match guard in javascript.rs's walk_ast_nodes_depth. Genuine string, template, and string-literal-type nodes are always named and unaffected. No user-facing command/language/architecture changes, so README/CLAUDE.md/ ROADMAP.md are unaffected — docs check acknowledged. Fixes #1729 Impact: 8 functions changed, 13 affected
codegraph exports <file> reported zero consumers for an exported
function/symbol that is genuinely imported and used elsewhere, whenever
the importing file renamed the binding at the import site
(import { X as Y } from '...'). Example: collectFiles in
src/domain/graph/builder/helpers.ts is imported (as collectFilesUtil)
and called by collect-files.ts and native-orchestrator.ts, but
`codegraph exports helpers.ts -T --json` showed consumerCount: 0.
Root cause (extractor, both engines): extractImportNames (javascript.ts)
and its mirrored scan_import_names_depth (javascript.rs) handled
import_specifier nodes with `node.childForFieldName('name') ||
childForFieldName('alias')`. Per the tree-sitter grammar, `name` is
*always* present on import_specifier (the name as declared by the source
module); `alias` only exists for `X as Y` and holds the *local* binding
actually referenced by call sites. Preferring `name` unconditionally
meant the local alias was silently dropped for every renamed import —
`imp.names` recorded "collectFiles" instead of "collectFilesUtil", so
`importedNames` (keyed by call-site text) never had a matching entry and
the call fell through unresolved.
Fixing only the extractor was not sufficient: once `imp.names` correctly
holds the local alias, `resolveCallTargets` (call-resolver.ts) /
resolve_call_targets (build_edges.rs) still searched the *target file*
for a symbol literally named "collectFilesUtil" — which doesn't exist
there (only "collectFiles" does). This required threading a second
piece of information end-to-end: for each renamed specifier, the local
alias's *original* exported name, so target-file/barrel lookups search
by the right name while importedNames stays keyed by the call-site text.
Changes (both engines, full-build + incremental/native-orchestrator +
native-hybrid paths):
- types.ts / types.rs: new `Import.renamedImports` /
`Import.renamed_imports` field — { local, imported } pairs, populated
only for specifiers that actually rename a binding.
- extractors/javascript.ts, .rs: extractImportNames /
extract_import_names_with_renames now prefer `alias` (local binding)
for import_specifier and record the rename pair. export_specifier is
deliberately left unchanged (see scope notes below). Fixed in both the
walk path (handleImportStmt) and the query/WASM-worker path
(handleImportCapture) on the TS side.
- build-edges.ts, build_edges.rs, import_edges.rs, pipeline.rs,
call-resolver.ts, incremental.ts: buildImportedNamesMap /
collect_imported_names_for_file now also produce a local-alias ->
original-name map; resolveCallTargets / resolve_call_targets consult
it to search the target file by the original name. Applied
consistently to the primary call-resolution path, the
Object.defineProperty post-pass, points-to alias resolution, barrel
edge / import-type edge emission, and cross-file return-type
propagation — every place that previously assumed a call site's local
name equals the target file's declared name.
Native Rust: touched and verified. Rebuilt locally via
`npx napi build --platform --release` (crates/codegraph-core), codesigned
(`codesign --sign - --force`), and installed over the loaded
node_modules/@optave/codegraph-darwin-arm64/codegraph-core.node so the
fix was exercised by the actual native engine, not the prebuilt npm
binary. `cargo test --lib`: 427 passed.
Tests added:
- tests/parsers/javascript.test.ts: extractor-level coverage for the
local-name/rename-pair extraction, a mixed renamed+non-renamed
specifier list, and confirmation export_specifier is unaffected.
- crates/codegraph-core/src/extractors/javascript.rs: matching Rust unit
tests.
- tests/integration/issue-1730-renamed-import-consumer.test.ts: new
end-to-end test building real files through buildGraph() on both wasm
and native engines, asserting the calls edge exists, no edge is
created against a nonexistent "collectFilesUtil" symbol, and
`codegraph exports` credits the consumer — on both engines.
tests/integration/exports.test.ts was not extended: that file
hand-inserts DB rows and only exercises the query layer, which is
unaffected by this bug (the missing piece was edge creation, not the
consumer query added for #1724).
Verified against this repo's own graph (native engine, rebuilt):
`codegraph exports src/domain/graph/builder/helpers.ts -T --json` now
credits collectFiles with 2 consumers (collect-files.ts,
native-orchestrator.ts); WASM engine agrees exactly. Without -T, also
picks up tests/unit/builder.test.ts via the (unrenamed) builder.ts
barrel re-export, which already worked pre-fix and was only hidden by
-T in the original repro.
Scope notes — filed as separate issues rather than expanding this fix:
- #1823: export { X as Y } from '...' (barrel re-export with rename) is
not tracked — resolveBarrelExport/resolve_barrel_export key off the
original name, and export_specifier extraction was deliberately left
unchanged here since barrel/reexport tracing is a distinct mechanism.
- #1824: dynamic import destructuring rename
(const { a: b } = await import(...)) has the same class of bug in
extractDynamicImportNames, a different extraction function.
- #1825: a renamed import used as a call receiver
(import { X as Y }; Y.method()) still fails to resolve — confirmed
empirically (no calls edge created) — a different resolution strategy
(resolveByReceiver/resolveViaDirectQualifiedMethod in
resolver/strategy.ts) than the one this fix addresses.
Full test suite (202 files, 3398 tests) and lint pass clean.
docs check acknowledged: internal resolver/extractor bug fix, no new
feature/language/CLI surface/architecture change -- README.md,
CLAUDE.md, and ROADMAP.md are unaffected.
Fixes #1730
Impact: 29 functions changed, 47 affected
… builds insertNodes committed file_hashes for changed files in the same transaction as node insertion, before resolveImports/buildEdges rebuilt those files' edges (a separate, later stage/transaction) — in both the JS/WASM pipeline and the native Rust orchestrator. Any exception, crash, or interruption between the two left a hash that claimed a file was "up to date" while its edges still reflected an older revision. Since change-detection trusts file_hashes exclusively, that divergence was never self-healed by later builds. Defer the file_hashes commit so it only runs once edges have been rebuilt to match: - insert-nodes.ts: insertNodes no longer writes file_hashes for changed files; new commitFileHashes() does it, called from pipeline.ts after buildEdges. - insert_nodes.rs / pipeline.rs / connection.rs: do_insert_nodes no longer upserts file_hashes (only removed-file cleanup, which has no coupling risk); new commit_file_hashes() runs after Stage 7 (edges). Watch-mode's rebuildFile never wrote file_hashes at all, leaving it permanently stale after every edit — also fixed by writing/deleting the hash once a file's edges have been rebuilt or the file is deleted. Adds tests/integration/issue-1731-hash-edge-coupling.test.ts: a fault-injection test that throws inside buildEdges mid-incremental-build and asserts the hash does not advance until edges genuinely match (and that a retry self-heals), plus coverage for rebuildFile keeping file_hashes in sync with its edge rebuilds. Fixes #1731 Impact: 8 functions changed, 13 affected
checkNoSignatureChanges compared def.line (post-change, new-file coordinates, from the rebuilt graph) against diff.oldRanges (pre-change, old-file coordinates). Any diff that shifts line counts before a given point in a file — virtually all deletions/insertions — could make an untouched symbol's new line number coincidentally fall inside the old hunk's range, producing a false "signature change" violation. checkMaxBlastRadius already used diff.changedRanges (new-file positions) against the same db; checkNoSignatureChanges now does the same. The parameter is renamed from oldRanges to changedRanges to make the expected coordinate space explicit. diff.oldRanges remains computed by parseDiffOutput and is still exercised directly by its own tests. No README/CLAUDE.md/ROADMAP updates needed — internal predicate logic fix, no new commands, languages, or architecture changes (docs check acknowledged). Fixes #1732 Fixes #1737 Impact: 2 functions changed, 5 affected
…g WASM grammars Every git worktree gets its own untracked node_modules/ and grammars/, so a worktree set up before a host Node upgrade (or where npm install was interrupted) can end up with a better-sqlite3 binary compiled for the wrong Node ABI, or an incomplete grammars/ directory — both fail deep inside a build or test run with confusing, unrelated-looking errors rather than a clear diagnosis. Adds src/infrastructure/doctor.ts with two checks whose decision logic is pure and unit-tested in isolation from the I/O: - native ABI compatibility, via a real require() attempt - grammar completeness against the full LANGUAGE_REGISTRY, split by the registry's own required flag: a missing required (JS/TS/TSX) grammar fails the check, but a missing optional grammar only warns. Non-required parsers are designed to fail gracefully at runtime (per this repo's own CLAUDE.md), so a worktree missing one rarely-used language's grammar must stay able to run npm test, not get hard-blocked before a single test starts. scripts/doctor.ts is the CLI entry point (npm run doctor, optionally with --fix for a scoped, worktree-local repair covering both blocking and non-blocking findings) and is also wired as pretest so npm test fails fast on a genuinely blocking problem instead of a wall of unrelated failures. Fixes #1733 docs check acknowledged: CLAUDE.md already covers the new npm run doctor command and infrastructure/doctor.ts (added in this same change); README.md does not document npm-run dev scripts (only the shipped codegraph CLI), and ROADMAP.md has no phase this bug-fix-sized change affects. Impact: 6 functions changed, 3 affected
`codegraph communities --drift` produced different modularity and community assignments across separate full rebuilds of byte-identical source. Two independent, compounding causes in the native Rust engine: 1. The build pipeline collected parsed file symbols into a `std::collections::HashMap` (pipeline.rs), whose iteration order is randomized per-process. That order drove node/edge insertion order into SQLite, so the same file could get a different autoincrement `id` (and therefore a different position in the in-memory graph) on every rebuild. Fixed by switching `file_symbols` to `BTreeMap` throughout the pipeline, import-edge, and structure-metrics stages, so insertion order is always sorted by file path. 2. The native Louvain local-move phase (louvain.rs) accumulated per-candidate-community weights in a `HashMap`. A genuine tie in modularity gain between candidate communities was broken by hashmap bucket order instead of a reproducible rule -- non-deterministic even with a fixed random seed, since the seed only controls visitation order, not this tie-break. Fixed by switching `cur_edges`/`comm_w` to `BTreeMap`. Also added `ORDER BY` to the node/edge read queries used to build the community-detection graph (both native `graph_read.rs` and the JS/WASM `graph-read.ts` mirror), as defense in depth: SQLite's row order for a bare `WHERE` scan is otherwise unspecified. Verified via 10+ full rebuilds of this repo's own ~900-file graph producing byte-identical `communities --drift --json` output, versus differing modularity/community counts on every rebuild beforehand. This is an internal determinism fix with no new features, commands, or language support changes, so no README/CLAUDE.md/ROADMAP.md updates are needed (docs check acknowledged). Fixes #1734 Impact: 4 functions changed, 17 affected
The PostToolUse hook's hardcoded extension case-statement was a hand-copied subset of EXTENSIONS (src/shared/constants.ts) that had drifted out of sync — .mjs/.cjs were missing, so editing those files silently skipped the incremental rebuild and left .codegraph/graph.db stale. The hook now prefers dist/hook-extensions.txt, a plain-text snapshot of EXTENSIONS generated by scripts/gen-hook-extensions.mjs as part of `npm run build`, checked with a native `grep -qxF` (no extra Node startup per edit). A synced hardcoded case list remains as a fallback for before the first build. tests/unit/hook-extensions.test.ts fails if that fallback ever drifts behind EXTENSIONS again. This only touches internal dev-tooling (a Claude Code hook and its build-time codegen script) — no language support, CLI feature, or architecture surface changed, so README/CLAUDE.md/ROADMAP do not need updates. docs check acknowledged. Fixes #1736
…n incremental rebuild codegraph structure --depth 2 --json reported stale fileCount/symbolCount/ fanIn/cohesion/density for a directory after an incremental rebuild added or removed a file in it. Only a full rebuild (--no-incremental) produced correct numbers. Root cause: the small-incremental fast path (updateChangedFileMetrics in domain/graph/builder/stages/build-structure.ts, mirrored by update_changed_file_metrics in crates/codegraph-core/src/features/ structure.rs) only ever updated per-FILE node_metrics rows. It never touched directories at all -- no directory-metrics recompute, no `contains` edge for the new file, and no directory node for a brand-new directory. This path triggers whenever an incremental build touches at most smallFilesThreshold (5) files and the repo already has more than 20 files -- i.e. almost every normal edit-and-rebuild cycle on a non-trivial repo, including a pure-removal build (0 parsed files, which trivially satisfies the "<=5" gate). The full (non-fast-path) incremental branch was already correct in both engines -- it always recomputes every current directory's metrics unconditionally. Fix: added refreshAffectedDirectoryMetrics (build-structure.ts) and its mirror refresh_affected_directory_metrics (structure.rs), which run alongside the existing per-file fast path whenever it's taken. They recompute metrics for the ancestor directories of every file touched by the build (added, removed, or modified), plus any directory reachable from them via a live cross-directory import edge (a changed file gaining/losing an import into a sibling package shifts that package's fan-in/fan-out too, even though none of its own files changed -- mirrors the one-hop neighbour-expansion classifyNodeRolesIncremental already does for role classification). Directory/edge bookkeeping (node creation, `contains` edges) is wired up idempotently, so a file landing in a brand-new (possibly multi-level) directory is handled too. All of this uses indexed point queries against the live DB state, bounded by (changed files x path depth) rather than repo size, so it stays cheap enough to run unconditionally on the fast path. getAncestorDirs was promoted from a features/structure.ts-private helper to shared/constants.ts so both the fast path and the full path use the same implementation. Filed #1839 for a narrower residual gap this does not cover: a directory whose only link to the touched file set was an edge to/from a file that was itself just removed can't be discovered here, since that edge's evidence is already deleted by the purge step that runs earlier in the pipeline. Verified against the real incremental pipeline (not just unit-level): a throwaway repro script confirmed the stale fileCount/symbolCount and a missing `contains` edge for the new file before the fix, and correct, full-rebuild-matching output after, on both engines (native rebuilt via napi build + codesign for local verification). Added tests/integration/issue-1738-structure-metrics-incremental.test.ts, which diffs incremental-rebuild output against a from-scratch full build of the identical final file set across add/remove/new-nested-directory/ cross-directory-neighbour scenarios, run against both WASM and native. npm test: 207 files / 3444 tests passed, 0 failed. npm run lint: clean. cargo check / cargo test -p codegraph-core: clean. This is an internal bug fix to incremental-build correctness -- no new language support, CLI commands, or architecture surface changed, so README/CLAUDE.md/ROADMAP.md do not need updates (docs check acknowledged). Fixes #1738 Impact: 3 functions changed, 8 affected
…n DEFAULTS Moves the remaining inline maxBuffer magic numbers for git subprocess calls into DEFAULTS, following the same pattern already applied to resolveSecrets's apiKeyCommand execFileSync options: - DEFAULTS.coChange.execMaxBufferBytes (50 MB) — git log in cochange.ts - DEFAULTS.check.execMaxBufferBytes (10 MB) — git diff in check.ts and diff-impact.ts (same operation, shared constant) - DEFAULTS.build.execMaxBufferBytes (100 MB) — git check-ignore in native-orchestrator.ts Call sites that already had a resolved config in scope (check.ts, diff-impact.ts, native-orchestrator.ts) now read the value off it so .codegraphrc.json overrides apply; cochange.ts reads DEFAULTS directly, matching its existing convention for other coChange fields. No behavioral change — numeric values are unchanged. docs check acknowledged: no README/CLAUDE.md/ROADMAP.md updates needed — this is a config-registration fix with no language/architecture/roadmap impact; docs/guides/configuration.md was updated with the three new keys. Fixes #1739 Impact: 10 functions changed, 16 affected
…ng fan-in checkMaxBlastRadius previously failed any staged change touching a function whose absolute transitive-caller count exceeded the threshold, regardless of whether the diff actually changed anything risky. A function reachable only through a near-universally-called "spine" function (e.g. resolveSecrets via loadConfig) would fail on any edit at all, including fully behavior-preserving ones like replacing an inline literal with a named constant. checkMaxBlastRadius now exempts a touched function from the threshold unless the diff changed its call graph shape: its own declaration line was touched (signature/name risk), or the set of paren-preceded tokens referenced in its body changed (a callee was added, removed, or swapped). This is a mechanical, diff-text heuristic rather than a full pre/post call-graph reconstruction — parseDiffOutput now pairs each added-line run with whatever removed text it replaced (scoped to a single hunk, never crossing hunk boundaries) so the comparison needs no re-parsing and stays fully synchronous. Missing edit data (e.g. hand-built ranges) conservatively falls back to the old always-gate behavior, so this is purely opt-in via real diff data. Known limitation: paren-less call syntax (Ruby's `foo x, y`, Lua's `foo "arg"`) is invisible to the token-set comparison, so a newly introduced paren-less call could be missed and its function wrongly exempted. This is a deliberate, documented trade-off for a non-parsing check. Also updates titan-gate SKILL.md Step 8 (both the internal and docs/examples copies) to defer to Step 1's now-authoritative, shape-aware blast-radius predicate instead of re-deriving a pass/fail from diff-impact's raw absolute counts. No user-facing feature, command, or language changed -- README/ROADMAP left as-is; docs check acknowledged. Fixes #1740 Impact: 17 functions changed, 10 affected
…g callees extractCallbackReferenceCalls (and its native mirror) emitted a dynamic call edge for every bare identifier argument passed to any call expression, with no gating on the callee — unlike member_expression args, which were already gated by CALLBACK_ACCEPTING_CALLEES/HTTP_VERB_CALLEES (#974/#1191). When an identifier's name collided with an unrelated exported function elsewhere in the repo, the resolver's global-fallback confidence scoring bound the two together, fabricating a call edge and, in codegraph's own graph, a phantom cycle between src/features/communities.ts and src/presentation/communities.ts (analyzeDrift/communitiesData both call analyzeDrift(communities, ...), where `communities` is a plain parameter — not a call to the unrelated `communities` CLI command). Apply the same allowlist gate to identifier args that member_expression args already use, in both the TS/WASM extractor (extractCallbackReferenceCalls, shared by the walk and query extraction paths) and the native Rust mirror (extract_callback_reference_calls). Legitimate callback-by-reference patterns (e.g. arr.forEach(myCallback), router.use(handleToken)) are preserved since their callees are already in the allowlist. On codegraph's own repo this eliminates the two fabricated communities.ts edges and the phantom 3-node cycle, and reduces the broader dynamic=1/confidence=0.5/kind=calls edge signature from 342 to 14 (native) across the whole graph. Recalibrates the jelly-micro `classes` recall floor (6/31 -> 5/31): the one lost edge (f4 -> f1, from `function f4(x) { return x(f1); }`) was only ever matched because the removed heuristic fabricated it — `x` is an unresolvable parameter call, syntactically identical to the false-positive shape this fix removes, and only Jelly's points-to analysis can resolve it precisely. Fixes #1741 Internal extractor/resolver bug fix — no new language support, CLI surface, architecture, or roadmap changes. docs check acknowledged. Impact: 1 functions changed, 6 affected
Follow-up refinement to #1741's identifier-argument gating. A peer audit of the resolution-benchmark fixtures found two edges the naive name-only gate would break: - tests/benchmarks/resolution/fixtures/pts-javascript/array-from.js: `Array.from(arr, mapCallback)` — a legitimate, well-known stdlib callback pattern, not a name-collision false positive. `Array.from`'s callee name is `from`, which wasn't in CALLBACK_ACCEPTING_CALLEES at all, so the callback arg (mapCallback) was dropped. - tests/benchmarks/resolution/fixtures/typescript/callbacks.ts: `processEach(users, logUser)` — logUser passed to a project-defined higher-order function. No name allowlist can enumerate arbitrary user code, so this class of edge is a genuine, harder gap (see below). Fix Array.from properly rather than just adding 'from' to the general allowlist: `Array.from(arrayLike, mapFn, thisArg)` puts the callback at argument index 1, not 0 — naively allowlisting 'from' the same way as '.map'/'.forEach' would treat `arrayLike` (plain data at index 0) as a callback candidate too, reintroducing the exact name-collision false-positive class #1741 fixes. Added POSITIONAL_CALLBACK_ARG_INDEX (TS) / positional_callback_arg_index (Rust) so a callee can restrict eligibility to one specific argument index instead of "any position" (the existing behavior, still needed for variadic Express/Router middleware chains like `app.get(path, mw1, mw2, handler)`). Applies uniformly to every TypedArray constructor's .from (Uint8Array.from, Int32Array.from, etc.) since they share the same signature convention. This restores pts-javascript to 13/13 (100% recall/precision, matching its pre-#1741 baseline exactly). The processEach/filterThen case (arbitrary user-defined higher-order functions) is NOT fixed here — recognizing it needs the callee's own parameter type (function-shaped?), which is a resolver-level, cross-engine feature, not a name/position extension of this gate. Left the 3 affected expected-edges.json entries in typescript/expected-edges.json unchanged (they're real, decidable facts, not fabricated edges) and documented the gap in resolution-benchmark.test.ts's THRESHOLDS.typescript comment: recall is now 44/47 (93.6%), still well above the 0.72 floor. Tracked as a follow-up in issue #1845 (not fixed by this commit). Verified via tests/benchmarks/resolution/resolution-benchmark.test.ts (--reporter=verbose): pts-javascript 100%/100% (was 92.3% recall with the naive gate), typescript 95.7%/93.6% (unchanged by this commit, gap tracked). Cross-engine parity (native rebuilt + codesigned) and query/walk-path parity both re-verified with new dedicated test cases. Internal extractor/resolver refinement — no new language support, CLI surface, architecture, or roadmap changes. docs check acknowledged. Impact: 1 functions changed, 6 affected
`codegraph exports <file>` treated any file-level `reexports` edge as
proof that every export of the target file was re-exported, so a single
`export { X } from 'Y'` caused `reexportedSymbols` to dump all of Y's
exports (including symbols never mentioned in any reexport clause, and
symbols only ever imported as a type). The file-level edge only proves a
reexport *relationship* exists with a target file — it never carried the
specific symbol name(s).
Both engines now also emit a symbol-level `reexports` edge straight to the
specifically-named symbol for `export { X }` / `export { X as Z }`
clauses, mirroring the existing `imports-type` symbol-level edge from
#1724 (JS/WASM: build-edges.ts + incremental.ts watch-mode path; native:
import_edges.rs primary pipeline + build_edges.rs FFI fallback). The query
layer prefers these precise edges when present and only falls back to a
target's full export list when none exist — i.e. a genuine
`export * from 'Y'` wildcard, which really does re-export everything.
Fixes #1742
docs check acknowledged — internal query/edge-emission bug fix, no new
commands, languages, or architecture to document.
Impact: 14 functions changed, 14 affected
… complexity storeCfgResults / storeNativeCfgResults in ast-analysis/engine.ts (and a duplicate copy in domain/wasm-worker-entry.ts) overwrote the correctly computed cyclomatic complexity with a CFG-derived value (edges - blocks + 2). That formula doesn't model short-circuit logical operators (&&, ||, ??), optional chaining (?.), or nested function/closure bodies, all of which the AST-based cyclomatic walk correctly counts (matching cognitive complexity's treatment of closures). The override ran on every WASM build and on native builds whenever the Rust orchestrator was bypassed (e.g. an engine switch or schema/version change triggers forceFullRebuild), silently corrupting cyclomatic for any function using those constructs while leaving cognitive, Halstead, and LOC untouched. The native engine never had this override, so it was already correct. Cyclomatic is now always the single AST-derived value on both engines. Fixes #1743 Impact: 5 functions changed, 13 affected
Full builds always tag directly-resolved calls edges technique='ts-native' (applied by both engines, not just native), either inline or via a post-insert backfill. Two incremental-rebuild paths left it NULL instead: - codegraph watch (rebuildFile/emitIncrementalCallEdges in builder/incremental.ts) never set technique at all when inserting a calls edge. Fixed with a scoped post-rebuild backfill, mirroring applyEdgeTechniquesAfterNativeInsert. - codegraph build's native-orchestrator incremental path (backfillEdgeTechniquesAfterNativeOrchestrator) scoped its backfill to only the directly-changed files reported by Rust, missing one-hop reverse dependents whose outgoing edges into a changed file are reconnected by Rust's own reverse-dep cascade and so also carry a fresh, untagged technique. Fixed by expanding the scope to include them. Rust itself never writes the technique column on any edge-insertion path (confirmed across crates/codegraph-core) — it has always been the JS-side backfill's job, so no native crate changes are needed. No README/CLAUDE.md/ROADMAP.md updates needed — bug fix only, no new architecture, commands, or language support (docs check acknowledged). Fixes #1744 Impact: 4 functions changed, 9 affected
response.json() in embedRemote sat outside the try/catch guarding the fetch call, so a 200 OK response with a malformed/non-JSON body (e.g. an HTML error page from a misconfigured proxy) threw a raw uncaught SyntaxError instead of the descriptive EngineError used by every other failure mode in this function (timeout, network failure, non-2xx status, bad response shape, dimension mismatch). Fixes #1745 Impact: 1 functions changed, 9 affected
getDefaultUserConfigPath() and resolveUserConfigPath() each implemented the same XDG_CONFIG_HOME / APPDATA / ~/.config three-way branch, so a future change to the priority order or path shape had to be applied in two places. Extract the branch into computePlatformDefaultConfigPath(), called by both; resolveUserConfigPath() still layers its own existsSync checks on top. Pure dedup — no behavior change. Fixes #1746 Impact: 3 functions changed, 38 affected
…onfig.ts Extract-method decomposition of the 5 functions in src/infrastructure/config.ts flagged for excessive Halstead effort (>15000) while staying within warn-only range on cognitive/cyclomatic/maxNesting/mi: loadConfig, loadConfigWithProvenance, resolveConsent, promptForConsentIfNeeded, and resolveSecrets. Each function's natural sub-steps (merge layers, consent-resolution stages, provenance-tracking stages, prompt gating/IO, secret-command validation/exec) are pulled into named private helpers, with the original function reduced to a thin orchestrator. No control flow, priority ordering, or error handling was changed — verified via the full test suite and manual sanity checks of config loading, global-config consent resolution, and apiKeyCommand secret resolution. Fixes #1747 Impact: 20 functions changed, 136 affected
… check acknowledged) Extract-method refactor only, no behavior change. Both functions exceeded the halstead.effort FAIL threshold (>15000) and were flagged WARN on cognitive/cyclomatic complexity. findDbPath: extracted resolveCustomDbPath (--db directory/`.codegraph` handling), resolveDbSearchCeiling and resolveDbSearchStartDir (realpathSync normalization), and walkUpForDbPath (the parent-directory walk with git ceiling/no-ceiling stop conditions). openRepo: extracted wrapInjectedRepo (opts.repo validation), tryOpenRepoNative (native rusqlite attempt with busy/locked re-throw and fallback-on-failure), and openRepoSqliteFallback (better-sqlite3 fallback construction). halstead.effort: findDbPath 30574.33 -> 1710.82, openRepo 15118.12 -> 2055.6. No newly extracted helper exceeds any threshold. Internal refactor only — no public API, CLI, or language-support changes, so README/CLAUDE.md/ROADMAP do not need updates. Fixes #1748 Impact: 11 functions changed, 159 affected
Extracts the /\b(busy|locked|SQLITE_BUSY|SQLITE_LOCKED)\b/i regex check into a single isBusyOrLockedError(msg) helper, replacing the identical inline check duplicated between openRepo's native path (tryOpenRepoNative) and openReadonlyWithNative's native path. No change to matching behavior or the busy_timeout value. DEFAULTS.db.busyTimeoutMs (this issue's other flagged cleanup) was already added and wired into openDb/openReadonlyOrFail by an earlier commit in this stack (8f23020); this commit covers the remaining regex duplication only. docs check acknowledged: internal dedup-only refactor, no CLI surface, language support, or documented architecture/design decision changed. Fixes #1749 Impact: 3 functions changed, 27 affected
…ng them The catch block in resolveCustomDbPath (part of the findDbPath flow) caught all fs.statSync errors — including unexpected ones like EACCES or symlink loops — indistinguishably from the expected "path doesn't exist yet" case, leaving no diagnostic trail. Add a debug() call matching this file's existing convention (18 other catch blocks already do this). Fixes #1750 Impact: 1 functions changed, 68 affected
…oseDbDeferred closeDbPair, closeDbPairDeferred, and closeDbDeferred in src/db/connection.ts had zero direct test coverage despite being core resource-lifecycle primitives in a fanIn-55 file — the same category of gap where the openReadonlyWithNative leak went undetected (see openReadonlyWithNative-leak.test.ts). Adds three describe blocks to tests/unit/db.test.ts, alongside the existing openDb/closeDb coverage: - closeDbPair: native handle closes before the better-sqlite3 handle; a native close failure doesn't prevent the better-sqlite3 close; works with no native handle present. - closeDbPairDeferred: native closes synchronously within the call, while the better-sqlite3 close is deferred via closeDbDeferred (including when the native close throws). - closeDbDeferred: the advisory lock releases synchronously (verified via a real lock file), the handle itself closes on the next tick, and flushDeferredClose() closes it synchronously when called first — with the originally scheduled callback correctly skipping a second close. Fixes #1751
…arations on incremental rebuild
Root cause: reconnectReverseDepEdges (build-edges.ts, WASM/JS engine) and
its native mirror reconnect_reverse_dep_edges (crates/codegraph-core's
detect_changes.rs) re-attach a reverse-dependency caller's edge to its
purged-and-reinserted target using only (name, kind, file) plus "nearest
to the old line" as a tiebreak. When a file contains multiple distinct
symbols sharing the same name and kind -- e.g. several object-literal
close() {} methods returned from different functions in the same file, a
pattern this repo's own src/db/connection.ts uses four times -- nearest-
line is not a reliable way to tell them apart: once unrelated code
shifts the whole same-named group, an old reference line can end up
numerically closer to a different sibling's new line than to its own,
silently re-attaching the edge to the wrong symbol (and collapsing
distinct edges together via INSERT OR IGNORE while leaving another
candidate untargeted). A full rebuild is immune because it re-resolves
every call from scratch using real call-site information, not line
proximity.
Root-caused by replaying this repo's own last 35 real commits as a
sequence of incremental builds and diffing the result against a full
rebuild of the identical final source: node tables came out byte-
identical, but 5 reverse-dep callers ended up wired to close@line 433
instead of the correct close@line 580.
Fixed by recording each target's 1-based ordinal rank (by line) among
its same-(name,kind) siblings at save time, and using that ordinal --
not line proximity -- to re-select the correct candidate after
purge+reinsert, falling back to nearest-line only when the sibling
count itself changed since save (a genuinely ambiguous case).
Changes:
- src/domain/graph/builder/context.ts: extend savedReverseDepEdges with
tgtOrdinal/tgtSiblingCount.
- src/domain/graph/builder/stages/detect-changes.ts: compute each
target's ordinal/sibling-count before purge (computeNodeOrdinals).
- src/domain/graph/builder/stages/build-edges.ts: reconnect using the
saved ordinal instead of nearest-line (pickReconnectTarget).
- crates/codegraph-core/.../detect_changes.rs: identical fix on the
native engine (compute_ordinals, pick_reconnect_target), plus Rust
unit/integration tests covering the ordinal match, the sibling-count-
changed fallback, and a full save/purge/reconnect round trip.
Native engine: fixed in source and manually verified by full read-
through (borrow-checker/type correctness), but NOT compiled or run via
cargo test in this session -- the environment's disk ran critically low
(shared machine, other concurrent sessions) and a from-scratch napi/
cargo build for this crate was not safe to attempt. The TS-side fix is
complete, tested (full suite green), and independently verified via a
controlled A/B swap against the pre-fix code reproducing the exact
divergence this fix resolves.
A separate, pre-existing bug was found and filed independently
(#1863): resolveByGlobal's receiver-less call
resolution matches every same-named candidate clearing a loose
directory-proximity confidence threshold and creates an edge to each of
them, rather than picking the best match. That bug reproduces on a
from-scratch full build too (not incremental-specific) and is out of
scope for this fix.
docs check acknowledged: internal correctness fix to incremental-build
edge reconnection; no CLI surface, language support, or documented
architecture/design decision changed.
Fixes #1752
Impact: 5 functions changed, 12 affected
…sis.pointsToMaxIterations MAX_SOLVER_ITERATIONS was a hardcoded 50 in both the WASM points-to solver (points-to.ts) and the native Rust solver (build_edges.rs), duplicating but never reading DEFAULTS.analysis.pointsToMaxIterations. Threads a maxIterations parameter from the pipeline's resolved config through buildPointsToMap -> buildPointsToMapForFile -> buildCallEdgesJS/buildCallEdgesNative on the TS side, and through build_call_edges -> process_file -> build_file_context -> build_pts_map_for_file -> build_points_to_map on the Rust side, sourced from a new BuildConfig.analysis.points_to_max_iterations field deserialized from the JSON config payload already passed to the native engine. Default value (50) is unchanged when no override is configured. docs check acknowledged: no README/CLAUDE.md/ROADMAP.md updates needed — purely internal plumbing for an already-documented, already-accepted config key with no new CLI flags or user-facing surface. Fixes #1753 Impact: 7 functions changed, 6 affected
generator.ts's embedding-progress messages now go through logger.info() (unconditionally visible, matches the info()-based "Reusing previously- stored embedding model" message already used in cli/commands/embed.ts) instead of console.log/stdout. semantic.ts's dimension-mismatch message is a genuine warning (same severity class as the file's existing warnOnSimilarQueries), so both of its console.log lines are merged into a single warn() call, following the fix already applied to prepare.ts. cli-formatter.ts is the actual data-output layer for `codegraph search` (including the --json contract), directly analogous to presentation/queries-cli/ — CLI display wrappers for query functions that already live in presentation/ and call console.log directly. No other domain/ file has precedent for presentation code living there, so it moves to presentation/search.ts rather than being kept as a domain-layer exception. Only two import sites needed updating (cli/commands/search.ts, the domain/search/index.ts barrel) plus one test import. docs check acknowledged: internal logging-layer refactor + file move, no new feature/language/CLI/architecture surface — README.md, CLAUDE.md, and ROADMAP.md do not need updates. Fixes #1754 Impact: 2 functions changed, 9 affected
computeDeltaModularityDirected exceeded the cyclomatic threshold (11 vs
warn=10) from four repeated "newC < arr.length ? fget(arr, newC) [|| 0] : 0"
bounds-checked reads. Two of the four (inFromNew/outToNew, plus the oldC
siblings inFromOld/outToOld) included a `|| 0` fallback; the other two
(totalInStrengthNew/totalOutStrengthNew, plus totalInStrengthOld/
totalOutStrengthOld) did not.
Traced the asymmetry to its root: all six arrays involved are dense,
zero-initialized Float64Arrays populated purely by +=/-= over edge weights
and node strengths that are scrubbed of NaN/undefined at adapter.ts's
ingestion point (`+linkWeight(attrs) || 0`). No code path can put NaN or a
sparse "hole" into any of them, so `|| 0` is a no-op today for both array
families. The `|| 0` presence instead traces back to which arrays already
had a getter-based public API: getNeighborEdgeWeightToCommunity/
getOutEdgeWeightToCommunity/getInEdgeWeightFromCommunity always bake in
`|| 0`, and this function hand-inlined that same convention for the
edge-weight arrays it reads directly, but never established an equivalent
convention for the community-strength-total arrays (which is itself
inconsistently applied elsewhere in this file, e.g. computeDeltaCPM's
sizeOld vs. sizeNew on the same communityTotalSize array) — confirmed by
diffing against the original vendored ngraph.leiden source, where this
exact asymmetry already existed unchanged since the initial vendoring
commit.
Since the two families are provably equivalent here, extracted a single
shared `fgetOrZero(arr, i)` helper (bounds check + `|| 0`) into
typed-array-helpers.ts and applied it uniformly to all eight reads
(newC/oldC x edge-weight x strength-total), documenting why the fallback
is currently redundant but retained for defense-in-depth and consistency.
Split the two extracted read-groups into computeDirectedEdgeWeightTerms/
computeDirectedStrengthTerms helper functions.
cyclomatic 11 -> 3, cognitive 10 -> 2 for computeDeltaModularityDirected;
new helpers are cyclomatic 1 (computeDirectedEdgeWeightTerms/
computeDirectedStrengthTerms) and 3 (fgetOrZero) — none exceed threshold.
Verified behavior preservation two ways:
- Direct detectClusters(graph, { directed: true }) comparison across 12
seed/resolution/file-vs-function-level combinations on this repo's own
dependency graph (701 file nodes / 8833 function nodes) between the
pre-refactor and post-refactor build: byte-for-byte identical quality()
and community assignments.
- codegraph communities -T --json (undirected path) before/after,
controlling for native-vs-JS engine selection: byte-for-byte identical.
Added unit tests for fgetOrZero's bounds/zero/NaN-squashing contract, plus
two exact-value regression tests pinning computeDeltaModularityDirected's
quality()/assignment output for the existing directed-modularity fixtures,
so a future accidental re-divergence of the `|| 0` handling would be
caught.
Fixes #1755
docs check acknowledged
Impact: 4 functions changed, 7 affected
…pact commands `renderAuditFunction` (presentation/audit.ts) and `printFnImpactLevels` (presentation/queries-cli/impact.ts) both rendered the same `Record<number, ImpactLevelEntry[]>` transitive-caller-levels shape (produced by the shared `bfsTransitiveCallers` BFS) but with two different, independently-maintained text formats. Extract a single `renderImpactLevels(levels, opts)` helper into a new presentation/impact-levels.ts module and adopt it in both call sites, using the richer icon+truncation format from fn-impact as the canonical one (per issue #1756's recommendation). `opts.emptyMessage` lets audit.ts suppress the "No callers found." line, since its "Impact: N transitive dependent(s)" line already conveys a zero count and none of its other subsections (Calls/Called by/Tests) print an explicit empty message either. Investigated downstream dependents before making this change: - No test asserts on audit.ts's exact text output; tests/integration/audit.test.ts only exercises the auditData data layer, and all CLI-level audit/fn-impact tests (tests/integration/cli.test.ts, tests/unit/queries-unit.test.ts) use --json. - No skill or hook in .claude/ parses this text; every `codegraph audit` invocation across .claude/skills/ already passes --json. - docs/examples/CLI.md and MCP.md do document an audit output example, but it was already stale relative to the current implementation independent of this change (wrong header/complexity/impact wording, single-line "Calls:" list) -- filed as #1873 rather than fixing inline. - README.md/CLAUDE.md/ROADMAP.md contain no example output text for audit's impact-level rendering (checked directly) -- docs check acknowledged. BEHAVIOR CHANGE: `codegraph audit`'s impact-level text output now matches `codegraph fn-impact`'s icon+truncation format (per-level header with count, `^ <icon> <name> <file>:<line>` per entry, truncated at 20 with "... and N more") instead of the old plain `Level {n}: name1, name2, ...` comma list. --json/--ndjson output is unaffected. Fixes #1756 Impact: 9 functions changed, 6 affected
…ken-benchmark results computeSavings reimplemented the percentage-reduction formula inline instead of using the pct helper already defined for computeAggregate. Both call sites now share pct — verified byte-identical output across normal, zero-denominator, negative-savings, and null-guard cases. main()'s per-issue loop only serialized results to stdout after the full loop completed, so a crash partway through (another issue throwing, or the optional --perf benchmarks failing) discarded every already-computed result. The loop now overwrites token-benchmark.partial.json after each issue; the file is removed once the full run succeeds and the final JSON has been printed. Fixes #1757 Impact: 5 functions changed, 17 affected
The rust fixture had no main.rs, so trace_rust() failed immediately trying to inject `mod trace_support;` into a nonexistent file. This silently skipped the rust same-file recall assertion in tracer-validation.test.ts as "toolchain not available" even when cargo was installed, masking the fact that the rust dynamic tracer had never actually been exercised. Adds main.rs exercising the full models/repository/service/validator call chain through build_service()/add_user()/get_user()/remove_user() plus a direct_repo_access() helper, and documents the edges it introduces in expected-edges.json (mirroring how the swift/dart/zig fixtures already catalog every call sourced from their own main driver) — precision stays at 100%, recall is 58.3% against a larger, more complete manifest (24 edges vs. 14). Getting the tracer to actually run past the missing file surfaced three more bugs in the same trace_rust() code path that had never been reached before, since compilation was never previously attempted: - The dump_trace() injection used a GNU-only inline `i\text` sed form that BSD sed (macOS) rejects; switched to the portable `i\` + newline form. - The impl-block context regex captured the trait name instead of the concrete type for `impl Trait for Type` blocks, mislabeling e.g. EmailValidator.validate as Validator.validate in the trace output. - trace_support.rs's trace_call() held an immutable borrow from t.stack.last() across mutations of t.seen/t.edges, which doesn't compile under the borrow checker; clones the needed fields out first instead. The fixture itself also never compiled as a real crate: User didn't derive Clone (needed by find_by_id's .cloned()), and service.rs called Repository trait methods on self.repo without importing the trait. Both fixed. Verified end-to-end: native-tracer.sh now compiles and runs, producing 23/24 expected edges (the only miss, User.display_name, is legitimately unreachable since the repo stub's save() never persists into its HashMap). tracer-validation.test.ts's rust case now actually runs (rather than skipping) at 100% same-file recall (7/7) against the 50% threshold. Filed #1876 for the pre-existing static-resolution gap this also confirmed (receiver-typed/trait-dispatch calls through locally-typed variables aren't resolved by either engine). Fixes #1759 Impact: 1 functions changed, 1 affected Impact: 3 functions changed, 2 affected
parseDiffOutput tested every diff line against the file-header regexes unconditionally, including lines inside an active hunk body. A removed line whose original content starts with "-- " (e.g. a Markdown horizontal rule) becomes a "--- " line once diff-prefixed and was misdetected as a source-file header, silently dropped, and desynced the old-line cursor for the rest of the hunk. Symmetrically, an added line starting with "++ b/" becomes a "+++ b/..." line and was misdetected as a new-file header, flushing the in-progress run under a phantom file key and misattributing every later line in the hunk to it. DiffLineTracker now derives an insideHunk() state from the old/new line-count bounds declared by the most recent hunk header, and parseDiffOutput only attempts file-header matching while that is false: before a file's first hunk, or once the previous hunk's declared counts are fully consumed. Real diffs never have hunk-header counts that lie about their body length, so this makes header detection position-aware without weakening it for any real diff shape. No user-facing behavior, commands, or architecture changed. docs check acknowledged. Fixes #1761 Impact: 4 functions changed, 5 affected
… object shape
codegraph roles --role dead --json has always returned { count, summary,
symbols } (never a bare array, confirmed back to the command's original
commit) but three Node one-liners in titan-grind's SKILL.md assumed a bare
array, calling .length/.reduce()/.filter() directly on the parsed JSON.
Steps 0.12 and 4 threw "items.reduce is not a function"; the Step 2c
symbol-level duplicate scan silently swallowed the same TypeError in a
try/catch and always emitted an empty candidate list.
Read .count and .summary directly (summary is already the per-role
breakdown) and read .symbols for the duplicate-scan candidate list.
Fixes #1762
…y call sites Extends the config-driven db.busyTimeoutMs wiring (started in #1748/#1749 for openDb()/openReadonlyOrFail() and the call sites that already held a loaded config) to the ~30 ad-hoc read-only query call sites across features/*, domain/analysis/*, and domain/search/* that previously called openReadonlyOrFail() directly and silently fell back to DEFAULTS.db.busyTimeoutMs. Adds resolveBusyTimeoutMs() to src/db/connection.ts (exported via db/index.ts), sharing rootDir derivation with resolveDbSettings() via a new deriveRootDirFromDbPath() helper. For call sites that never loaded config in their path, this new call is threaded in directly. Fixing withReadonlyDb() in domain/analysis/query-helpers.ts centrally also covers its callers (exports.ts, dependencies.ts, context.ts) for free. For call sites that already loaded config but did so AFTER opening the DB (diffImpactData, checkData, complexityData, manifestoData, moduleBoundariesData), reordered config-load-before-db-open, mirroring the precedent already established by resolveDbSettings()/openReadonlyWithNative() (see the phase-15 gauntlet handle-leak fix). Verified this reorder is safe: loadConfig() only throws ConfigError for one narrow case (a non-string llm.apiKeyCommand), unrelated to db.busyTimeoutMs and independent of whether the DB exists — no existing test pins the previous error ordering for any of these call sites. Made every new config.db.busyTimeoutMs access optional-chained (config.db?.busyTimeoutMs) after this surfaced a real crash in tests/integration/complexity.test.ts, which mocks loadConfig() to return a partial object without a db key. Verified end-to-end against a real built graph with a custom .codegraphrc.json (db.busyTimeoutMs override) via the CLI (owners, check, structure --modules, map, audit, complexity, dataflow, search, co-change, ast, flow, cfg, roles, where, children), confirming the configured value reaches the PRAGMA busy_timeout call and the default still applies with no config override. Added tests/unit/busy-timeout-query-sites.test.ts covering resolveBusyTimeoutMs directly plus a representative sample of call sites from each category (ownersData, cfgData, manifestoData, hybridSearchData, withReadonlyDb), spying on Database.prototype.pragma to assert the configured value is actually applied. Filed two follow-ups discovered while completing this wiring pass, both intentionally out of scope here: - #1881: several functions (manifestoData, hybridSearchData, moduleBoundariesData, diffImpactData, complexityData, auditData) resolve their config via process.cwd() rather than the resolved --db path, unlike resolveDbSettings()/checkData() — a pre-existing inconsistency, not introduced by this change, that affects cross-repo --db usage for several config fields (not just busyTimeoutMs). - #1882: the Rust native DB layer (NativeDatabase::open_readonly/ open_read_write in crates/codegraph-core/src/db/connection.rs) still hardcodes PRAGMA busy_timeout = 5000. Investigated and confirmed this does NOT require giving Rust config-loading capability (the JS side already resolves the value) — it requires threading an already-resolved number across the napi FFI boundary plus a native rebuild across platforms, which is a heavier-weight change than this wiring pass. Refs #1763 (TS read-only call-site wiring is complete; the Rust portion of that issue's scope is deferred to #1882, so this does not close it). docs check acknowledged: internal config wiring only, no new config keys or CLI flags — db.busyTimeoutMs was already documented by the original phase-7 commit that introduced it. Impact: 36 functions changed, 131 affected
…n (docs check acknowledged)
extractObjectLiteralFunctions's pair branch only stripped quotes for
plain string-typed keys, so a computed string-literal key like
{ ['foo']: () => {} } fell through to the raw bracket/quote text,
emitting a definition named obj.['foo'] instead of obj.foo. This
broke resolution of obj.foo() call sites.
Extract the computed_property_name-unwrapping logic already used by
resolveMethodDefinitionName into a shared resolveComputedKeyName
helper and reuse it in the pair branch, skipping the pair (like the
adjacent method_definition branch already does) when the computed key
isn't a resolvable string literal (e.g. [Symbol.iterator]).
Mirrors the same fix in the Rust native engine: extract_object_literal_functions
(const) and match_js_objlit_qualified_method_defs (let/var) both had the
identical gap, now routed through shared resolve_computed_key_name /
resolve_pair_key_name helpers alongside resolve_method_def_name.
No README/CLAUDE.md/ROADMAP.md changes needed — internal extractor bug
fix, no new language, command, or architecture surface.
Fixes #1764
Impact: 3 functions changed, 17 affected
…ocs check acknowledged) incremental.ts's applyThisReceiverFallbacks (renamed applyCallFallbacks) only implemented the this.method() and Object.defineProperty fallbacks, missing the same-class bare-call fallback that the full-build path has (resolveSameClassBareCallFallback in stages/build-edges.ts). For class-scoped languages (non-JS/TS), a bare call with no receiver now retries qualified as <CallerClass>.<callName> in the incremental path too, guarded by isModuleScopedLanguage exactly as the full-build path is. Reuses the already-shared resolveSameClassQualifiedMethod from call-resolver.ts rather than duplicating its lookup logic. Fixes #1765 Impact: 2 functions changed, 6 affected
…ween build paths The full-build path's Object.defineProperty accessor fallback (resolveDefinePropertyAccessorFallback in stages/build-edges.ts, and its native-engine post-pass buildDefinePropertyPostPass) and the incremental path's (applyCallFallbacks in incremental.ts) diverged in their final same-file fallback tier: full-build returned any same-file node named call.name regardless of kind, while incremental filtered to function/method kinds only. A getter/setter registered via Object.defineProperty always dispatches to callable code, and every resolved target is unconditionally emitted as a 'calls' edge downstream — so the incremental path's kind filter is the correct behavior. Extracted the shared logic into resolveDefinePropertyAccessorTarget in call-resolver.ts (applying the kind filter) and updated all three call sites - the full-build fallback, its native-engine post-pass, and the incremental fallback - to use it, removing the divergent duplicated implementations. docs check acknowledged: internal call-resolution bug fix/refactor, no new feature/language/architecture/command surface to document in README/CLAUDE.md/ROADMAP.md. Fixes #1766 Impact: 4 functions changed, 14 affected
… structure batchUpdateRoles in features/structure.ts re-implemented the same chunk-size-keyed prepared-statement cache already used by getNodeStmt/getEdgeStmt/getExportStmt in domain/graph/builder/helpers.ts (check cache by chunk size, prepare + cache on miss) — the SQL and cache scope differed, but the caching mechanics were identical. Extracted the shared mechanics into src/shared/chunked-stmt-cache.ts: getOrCreateChunkStmt (flat Map<chunkSize, stmt>) and its per-database wrapper getOrCreatePerDbChunkStmt (WeakMap<db, Map<chunkSize, stmt>>). builder/helpers.ts's three statement getters now delegate to getOrCreatePerDbChunkStmt against their existing WeakMap caches (same persisted-across-calls semantics as before); batchUpdateRoles now calls getOrCreateChunkStmt against a Map created fresh per call (same per-call-only lifetime as before). No SQL, chunking, or caching-scope changes on either side — one implementation instead of two copies of the same shape. Placed the primitive in shared/ rather than exporting it from builder/helpers.ts: nothing outside domain/graph/builder/ currently imports from that directory directly — index.ts, cli/commands/build.ts, and features/branch-compare.ts all go through the domain/graph/builder.ts barrel, which only re-exports the builder's public API (buildGraph, collectFiles, etc). Adding a generic, domain-agnostic statement-cache helper to that barrel's surface for a features/ consumer would be the first such exception; shared/ already hosts this class of cross-cutting, domain-free utility (paginate.ts, globs.ts, sleep.ts). docs check acknowledged: pure internal dedup refactor, no new feature, language, command, or architecture surface to document. Fixes #1767 Impact: 6 functions changed, 26 affected
…Insert chunk loop The technique/confidence backfill loop in applyEdgeTechniquesAfterNativeInsert re-prepared its two UPDATE statements on every chunk iteration instead of caching by chunk size. Apply the shared getOrCreatePerDbChunkStmt primitive (src/shared/chunked-stmt-cache.ts) with a persistent per-db WeakMap cache, matching the batchInsertEdges/batchInsertNodes caches in builder/helpers.ts — this function can run twice within a single buildEdges() call against the same db (once from insertNativeBulkEdges, once from reconnectReverseDepEdges), so a fresh per-call cache would still miss across that second invocation. Pure performance refactor: SQL text and behavior are unchanged. Fixes #1768 Impact: 1 functions changed, 4 affected
…tance (docs check acknowledged) computeConfidence (and its Rust mirror compute_confidence) scored call-edge resolution confidence using a fixed-depth check: same-directory, or dirname(dirname(caller)) === dirname(dirname(target)) for a "sibling directory" tier. That equality only holds when both files sit at the same depth, so a file in a subdirectory calling a method on a class declared in its direct parent directory (e.g. graph/algorithms/bfs.ts calling a method on the CodeGraph class in graph/model.ts) was scored as maximally distant (0.3) even though the receiver's type had already been correctly resolved via typeMap (from the parameter's `graph: CodeGraph` type annotation). That score fell below the 0.5 threshold used by resolveViaTypedMethod, silently dropping the call edge. Replaced both fixed-depth checks with a symmetric, depth-independent directory-tree distance (hops to the nearest common ancestor), preserving the existing same-directory (0.7) and sibling (0.5) tiers exactly and adding a new tier for direct parent/child directory nesting (0.6). Verified via the resolution-benchmark suite before/after the fix: all 42 language fixtures produce byte-identical precision/recall/TP/FP/FN numbers (aggregate recall 63.8%, 355/556 edges) — no regression. Fixes #1769 Impact: 3 functions changed, 26 affected
…thods optimiser.ts's computeQualityGain (the only caller of CPM/modularity delta computation in the Leiden hot loop) calls the standalone diffCPM/diffModularity functions directly — it never calls partition.deltaCPM(...), partition.deltaModularityDirected(...), or partition.deltaModularityUndirected(...). Confirmed via repo-wide grep that these three Partition interface methods, plus getCandidateCommunityCount, have zero callers in src/, tests/, or scripts, and that Partition/makePartition are not part of the public API (src/index.ts). Verified dead empirically, not just statically: replaced each of the four functions' bodies with a throw and ran the full test suite (3549 tests) — all passed, proving no code path anywhere reaches them. This also revealed that the "directed modularity delta — exact regression" tests added by #1755 (pinning computeDeltaModularityDirected's assumed output) actually exercise a different function — diffModularityDirected in modularity.ts, reached via computeQualityGain — since detectClusters never calls the Partition interface method. Corrected that test's comment accordingly instead of leaving it pointing at a deleted function. Removed: - The four dead Partition interface methods (deltaCPM, deltaModularityDirected, deltaModularityUndirected, getCandidateCommunityCount) and their backing implementations (computeDeltaCPM, computeDeltaModularityDirected, computeDeltaModularityUndirected, and the CPM/directed-term helper functions extracted in #1755/phase-22 that existed solely to serve them). - fgetOrZero (typed-array-helpers.ts), which becomes fully unreachable once computeDeltaModularityDirected (its only caller) is gone — it was extracted in #1755 specifically for that function. Removed its direct unit-test block along with it. Chose removal over wiring computeQualityGain to the Partition methods (deduplicating diffCPM/diffModularity vs. computeDeltaCPM/computeDeltaModularityDirected/ computeDeltaModularityUndirected) because the two implementations read through different data-access layers (raw PartitionState typed arrays vs. the PartitionView getter interface) — swapping which one runs in the Leiden hot loop would need rigorous before/after numerical verification on real data to rule out subtle divergence, on a file with a track record of exactly that kind of bug (#1734, #1755). Removal of genuinely unreachable code carries none of that risk. Verified no behavioral change: built this repo's own dependency graph (920 files / 19285 function-level nodes) and ran detectClusters directly (bypassing louvainCommunities' native-Rust preference to exercise the changed JS path) at file- and function-level granularity, both modularity and CPM quality functions, 3 resolutions x 3 random seeds (36 combinations) — before/after membership + quality() output is byte-for-byte identical (matching SHA-256). Also confirmed `codegraph communities -T --json` and `npm run typecheck`/ `npm run lint`/`npm test` (216 files, 3545 passed, 30 skipped, 2 todo) are unaffected. Filed #1895 to track codegraph's own dead-code detector blind spot (an object-literal-property-value reference counts as liveness without checking whether the resulting property is ever invoked) — out of scope to fix here. diff-impact --staged: 1 function changed (makePartition) -> 4 transitive callers across 2 files, all of which only use surviving Partition methods. Fixes #1770 docs check acknowledged Impact: 1 functions changed, 4 affected
…eral values
codegraph roles --role dead inconsistently flagged dispatch-table handler
functions (`{ matches, resolve: someFunction }`-style arrays) as dead,
depending on an unrelated, incidental property of the referenced function:
whether it happened to call another tracked symbol internally (fanOut > 0).
Root cause: codegraph created no edge at all for a bare function identifier
used as an object-literal property VALUE. The role classifier's
`node.kind === 'function' && node.fanOut > 0` heuristic in
classifyUnreferencedNode incidentally rescued whichever handlers happened to
have nonzero fanOut, leaving the rest misclassified dead-unresolved — a
false positive that risks deleting live dispatch-table handlers during
dead-code cleanup.
Fix: both engines now extract a dynamic `calls` edge (dynamicKind /
dynamic_kind = 'value-ref') for every bare-identifier object-literal
property value and shorthand property (`{ resolve: fn }` / `{ fn }`),
attributed to the enclosing scope via the existing findCaller machinery
(falls back to the widest enclosing constant/variable binding for top-level
dispatch tables, same as any other call). Resolution is restricted to
function/method-kind targets only, so plain data references
(`{ name: SOME_CONSTANT }`) are silently dropped rather than fabricating
nonsensical edges to constants.
Reuses the existing `calls` edge kind + dynamic flag (per ADR-002's "no new
edge kind" precedent, and the existing extractCallbackReferenceCalls
mechanism from #1741) rather than inventing a new edge kind — fan-in/fan-out
(scoped to `kind IN ('calls','imports-type')` / `kind = 'calls'`) already
account for it with zero changes to structure.ts/roles.rs's median
computations.
The `fanOut > 0` heuristic in classifyUnreferencedNode is kept, narrowed by
comment: it's no longer load-bearing for the object-literal case but remains
the only rescue for sibling value-reference patterns not yet extracted as
edges (logical-or fallback defaults, ternary defaults, array-of-functions
elements).
Applied to both engines:
- WASM/TS: src/extractors/javascript.ts (collectObjectLiteralValueRefCall,
wired into the shared runCollectorWalk used by both the walk and query
extraction paths), src/domain/graph/builder/stages/build-edges.ts
(resolveFallbackTargets kind filter), src/domain/graph/builder/incremental.ts
(mirrored filter for the incremental/watch-mode path), src/types.ts (new
'value-ref' DynamicKind variant).
- Native: crates/codegraph-core/src/extractors/javascript.rs
(handle_object_literal_pair_value_ref /
handle_object_literal_shorthand_value_ref, wired into match_js_node),
crates/codegraph-core/src/domain/graph/builder/stages/build_edges.rs
(matching kind filter in process_file).
Verified against the exact repro (this repo's own PARAM_NODE_HANDLERS
dispatch table in src/ast-analysis/visitor-utils.ts): all 5 handlers now
show totalDependents=1 and role=core on both engines, with zero dead-flagged
symbols in that file. Resolution-benchmark precision/recall is unchanged
across all 34 language fixtures (javascript 100/100, typescript 95.7/93.6,
aggregate 63.8% recall, identical before and after).
Fixes #1771
docs check acknowledged — internal edge-emission/resolver bug fix, no new
commands, languages, or architecture to document.
Impact: 9 functions changed, 19 affected
… inspect commands Extracts the duplicated "Calls"/"Called by"/"Tests" section rendering from presentation/audit.ts's renderCallRefs/renderRelatedTests and presentation/queries-cli/inspect.ts's renderExplainEdges into shared helpers in the new presentation/call-ref-sections.ts (renderCallRefsSection, renderNoCallEdgesFallback, renderRelatedTestsSection), following the same pattern established by #1756's impact-levels.ts unification. inspect.ts's richer format (indent-aware, "no call edges" fallback, singular/plural "Tests" label) is adopted as canonical in both commands. BEHAVIOR CHANGE: `codegraph audit`'s plain-text output (not --json/--ndjson) now matches `codegraph audit --quick`/`codegraph explain` exactly for these sections: - Prints "(no call edges found -- may be invoked dynamically or via re-exports)" when a function has neither callers nor callees. Previously audit printed nothing in this case. - "Tests (N):" is now "Tests (N file):"/"Tests (N files):" (singular/plural), matching explain's existing label. No changes to `codegraph audit --quick`/`codegraph explain` output — inspect.ts keeps its exact current format, just via the shared helper. docs check acknowledged: this is an internal presentation-layer refactor (no new command, language, or architecture change). CLAUDE.md's presentation/*.ts catch-all row already covers the new shared helper file, matching the precedent set by #1756's impact-levels.ts, which likewise did not get its own table row. Fixes #1772 Impact: 14 functions changed, 6 affected
…heck acknowledged)
Object-pattern destructuring targets (const { a, b } = x; and renamed
const { a: b } = x;) were hardcoded to kind: 'function' in both the TS/WASM
extractor (extractDestructuredBindings, shared by the walk and query paths)
and its native Rust mirror (extract_destructured_bindings), regardless of
what the destructured value actually held. Non-function bindings (e.g.
const { dbPath } = workerData) had no call-graph edges pointing at them by
name, so the dead-code classifier risked flagging them dead-unresolved even
when read repeatedly elsewhere in the file.
Both engines now emit kind: 'constant', matching the existing convention for
plain `const x = <literal>` bindings and array-pattern destructuring (which
was already correct). Every call site of these functions is const-gated, so
'constant' is unconditionally correct — no let/var case exists here. Call-
target resolution is kind-agnostic and constant is already included in the
top-level-binding caller-attribution fallback, so destructured
callback-style bindings (e.g. `const { handler } = router; handler(req)`)
still resolve correctly; verified via the resolution-benchmark suite showing
identical precision/recall across all 40 fixture languages before and after.
Array-pattern destructuring was checked and does not have this bug (already
kind: 'constant' in both engines' TS path); a separate, pre-existing native
vs WASM parity gap for array patterns was found and filed as #1901.
Docs check acknowledged: kind-classification bug fix only, no new
languages/features/architecture changes — README/ROADMAP tables unaffected.
Fixes #1773
Impact: 2 functions changed, 7 affected
Deduplicates the "loop N times, time each run, compute median" pattern
across the 7 remaining call sites that predated scripts/lib/bench-timing.ts:
- query-benchmark.ts's benchDepths()/benchDiffImpact() and
incremental-benchmark.ts's parent-process nativeBatchMs/jsFallbackMs
computations and benchmark.ts's benchQuery() now adopt timeMedian()
directly (converted to async; propagation stops at each call site
since all callers are already at ESM top-level/top-level-await scope).
- incremental-benchmark.ts's and benchmark.ts's "1-file rebuild"
measurements need the phases of the median-timed run, not just the
numeric median, so bench-timing.ts gains timeMedianWithValue(fn, runs,
beforeEach?) — returns the {ms, value} pair for the median-duration
run, with an optional untimed beforeEach(i) hook for per-iteration
setup (writing the probe file) that must stay outside the timed
window.
Verified no methodology regression: ran query-benchmark.ts and
incremental-benchmark.ts before/after on this repo (3 runs and 1 run
respectively) with consistent latencies, plus an isolated microbenchmark
quantifying the await-on-sync-value overhead at ~1 microsecond/call —
negligible against the millisecond-to-hundred-millisecond latencies
these scripts measure.
Fixes #1774
Impact: 4 functions changed, 4 affected
runPerfBenchmarks mixed build benchmarking, stats collection, hub-selection, and query benchmarking in one function, exceeding complexity thresholds (cognitive 15, cyclomatic 13) and the titan-gate halstead.bugs FAIL threshold (1.2559 > 1.0). Extract runBuildBenchmarks(engines, dbPath, nextjsDir, buildGraph) and runQueryBenchmarks(hubName, dbPath, fnDepsData, fnImpactData) as named async helpers; runPerfBenchmarks becomes a thin orchestrator. Pure decomposition, verified with a throwaway call-sequence harness — no control-flow or value changes. Fixes #1775 Impact: 3 functions changed, 3 affected
Contributor
Contributor
Codegraph Impact Analysis3 functions changed → 3 callers affected across 1 files
|
Base automatically changed from
fix/issue-1774-adopt-timemedian-sync-sites
to
main
July 8, 2026 08:57
docs check acknowledged — conflict resolution only, no new functionality. Impact: 46 functions changed, 183 affected
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to subscribe to this conversation on GitHub.
Already have an account?
Sign in.
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Summary
Pure decomposition (no behavior change) of
runPerfBenchmarks, which exceeded cognitive/cyclomatic thresholds and the titan-gatehalstead.bugsFAIL limit (1.2559 vs 1.0):runPerfBenchmarks(before)runPerfBenchmarks(after)runBuildBenchmarks(new)runQueryBenchmarks(new)Extracted
runBuildBenchmarks(per-engine full-build + no-op-rebuild timing loop) andrunQueryBenchmarks(fnDeps/fnImpact depth-scaling timing loop), both moved verbatim;runPerfBenchmarksis now a thin orchestrator.Closes #1775
Test plan
npm test— full suite green (3564 passed, 0 failed)npm run lint— clean (scripts/confirmed outside Biome's scope)node --experimental-strip-types --check— syntax OKgit diffconfirms moved blocks are byte-identical to the originalsbuildGraph/fnDepsData/fnImpactData— same 12/18-call sequences, same order, same args, same return shapes, before and afterFiled #1906 (sibling
runSessionalso exceeds thresholds, pre-existing, untouched) and #1907 (discoveredrunPerfBenchmarks'/buildCodegraph's dynamic imports reference flat paths that don't exist in the real compiled output —--perfhas likely never worked against this repo's actual build; identical before and after this change, left untouched).Stacked on #1905 (base branch
fix/issue-1774-adopt-timemedian-sync-sites) — only the token-benchmark.ts diff is this issue's change.