resolver: stop framework synthesizers from full-scanning the graph#267
Merged
Conversation
Two of the framework-dispatch synthesizers (fn-value-callback, factory-chain) scanned every edge in the graph via AllEdges() and filtered by kind in Go, instead of using the existing kind-scoped EdgesByKind/edgesByKinds helpers that push the filter into SQL. On a large multi-repo graph this meant decoding several times more edges (including their Meta blobs) than the pass could ever act on. fn-value-callback additionally re-fetched a file's node list via a fresh GetFileNodes SQL round-trip for every candidate in that file, rather than once per file — turning a file with a large candidate count (a generated parser, an ORM-generated source file) into hundreds of thousands of redundant queries against a handful of nodes. Caching the fetch per file removes the redundancy. Also add per-synthesizer and per-tail-pass elapsed-time logging to RunFrameworkSynthesizers so a future regression like this shows up in the existing per-pass log line instead of hiding behind one aggregate duration.
This was referenced Jul 8, 2026
vitaliyslion
pushed a commit
to vitaliyslion/gortex
that referenced
this pull request
Jul 8, 2026
Same treatment as the attribution sweep in the prior commit: the six lang-dispatch passes (relative/Lua/Razor imports, non-Go module attribution, Java/PHP override dispatch) shared one opaque duration. Broken out so a slow one has a breadcrumb instead of hiding behind the aggregate. Investigated a red herring while measuring this: several of these passes independently EdgesByKind-scan the same import edges, which looked like the redundant cross-pass pattern the framework-synthesizer fix (zzet#267) addressed. Measured cost was negligible (~1.5s combined on a 26-repo store) — not worth touching. The real, previously-invisible cost turned out to be the untimed gap around the pending-edge collection itself; see the follow-up investigation for that.
zzet
added a commit
that referenced
this pull request
Jul 10, 2026
resolver: stop framework synthesizers from full-scanning the graph
zzet
added a commit
that referenced
this pull request
Jul 10, 2026
Same treatment as the attribution sweep in the prior commit: the six lang-dispatch passes (relative/Lua/Razor imports, non-Go module attribution, Java/PHP override dispatch) shared one opaque duration. Broken out so a slow one has a breadcrumb instead of hiding behind the aggregate. Investigated a red herring while measuring this: several of these passes independently EdgesByKind-scan the same import edges, which looked like the redundant cross-pass pattern the framework-synthesizer fix (#267) addressed. Measured cost was negligible (~1.5s combined on a 26-repo store) — not worth touching. The real, previously-invisible cost turned out to be the untimed gap around the pending-edge collection itself; see the follow-up investigation for that.
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 join this conversation on GitHub.
Already have an account?
Sign in to comment
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
framework dispatch synthesisglobal pass took 9,667.76s (161 min) — 97.6% of the 165-minute post-resolve settle phase, with most of the ~46 registered synthesizers landing zero edges on this corpus.ResolveFnValueCallbacks,ResolveFactoryChains) scanned every edge in the graph viaAllEdges()and filtered by kind in Go, instead of using the existing kind-scopedEdgesByKind/edgesByKindshelpers every other synthesizer already uses (which push the filter into SQL).ResolveFnValueCallbacksadditionally re-issued a freshGetFileNodesSQL round-trip per candidate edge, not per file — on this real graph, 522,679 candidate placeholder edges triggered that many round-trips, with a single generated file (tree-sitter-swift/src/parser.c) alone responsible for 199,117 of them against a file holding only 24 nodes.RunFrameworkSynthesizersso a regression like this shows up directly in the existing log line instead of hiding behind one aggregate duration.Measured
Isolated before/after against a
.backup'd snapshot of the real 26-repo store (live daemon untouched throughout):Same resolution results (edge counts match within noise), dramatically faster.
Test plan
GOWORK=off go build ./...GOWORK=off go vet ./internal/resolver/... ./internal/indexer/...GOWORK=off go test ./internal/resolver/... ./internal/indexer/...GOWORK=off go test -race ./internal/resolver/... ./internal/indexer/...