[devin-ai-integration]

What and why

Develop's CircleCI race and coverage jobs exceed their 30-minute package budget in internal/exec/runtime and cmd/sysml since d8b60ed (usages subset the base usage of their kind): go test ./cmd/sysml -run TestExploreTablesTheSpacecraftRaceWithinItsBudget went from 155 s to 211 s on one machine. #1350 memoized Model.ImplicitGenerals per symbol, which brought that test back to 158 s on develop; what remained is that the memo is per model, and the runtime builds a model per execution, so a library symbol's implied generals were derived again — candidateBaseFeatures, declaredGeneralizationReaches over the library — by every model that asked (statedTypes/effectiveTypes through DeclaredSupertypes). On the internal/exec/runtime suite that derivation was 11.7 s of 415 s of CPU (ImplicitGenerals 12.2 s cumulative), with unitConversion, isBound and longName derived thousands of times each.

A library symbol's facts already record its direct supertypes; they now record its implied generals too, and ImplicitGenerals answers from the record whenever DirectSupertypes does, memoizing the answer like the derived one:

• LibraryFacts.ImpliedSupers lists every general the kind implies — the base definition (Parts::Part) as well as the base usage (Parts::parts) — in derivation order; it listed the base usages only, so answering from it lost the implied typing (the reason ImplicitGenerals read facts for Recorded() symbols alone).
• The per-document cache record (factRecord.Implied, impliedOf) carries the same entries, since the loader installs those facts before the interface records are written; a record whose implied general no reference reaches leaves the whole edge set to be derived on load, as a supertype does.
• internal/workspace/libs/stdlib.snapshot regenerated (go generate ./internal/workspace/libs with a fresh XDG_CACHE_HOME; a stale cache record would have been read back as the fact).

No behaviour change: the same generals in the same order, read instead of derived.

Baseline movement (re-recorded over develop with usages subsetting their base usage)

• docs/project/pilot-rejection-baseline.json (go run -C tools ./cmd/pilot-reject -update): 334 cases, 323 both reject, 11 only we reject, 0 only the pilot rejects, 0 both accept — every bucket keeps its verdict. One row moves: semantic/k49-binding-connector-with-three-ends.kerml ours-errors 2 → 3, the new one connector-ends ("b has 3 ends but specializes a binary link (Links::BinaryLink)"): the connector now conforms to Links::binaryLinks, so the end-count rule of KerML 1.1 §8.3.4.5.3 fires beside binding-binary; the pilot rejects it as before.
• docs/project/api-json-differential-baseline.json (go run -C tools ./cmd/pilot-apijson-diff -update, all nine roots provisioned): headline unchanged — 920 files, 86974 matched, 29497 ours-only, 40387 pilot-only, 596664 present-equal, 47633 present-different, 1021444 ours-missing, 65920 pilot-missing, 2020649 absent-both, 20867 indeterminate — so docs/project/api-json-differential.md does not move; within spec-examples one element changes bucket to a type mismatch (typeMismatch 390 → 391, total 1742 → 1743) and extensionProperties 140676 → 140679. make docs-counts reports every figure current.

How it was verified

• Failing-before test TestLibraryImpliedGeneralsAreRecordedWithTheDefinition (internal/workspace/model): a library usage's record carries an implied definition, ImplicitGenerals matches the record, a repeated query allocates 0 times and a fresh model answers the same — fails on d8b60ed (no definition recorded), passes here. TestEveryPersistedFactFamilyIsCompared now compares the new family.

• Timings on this machine (go test -count=1, same VM, runs sequential; "before" b60a18c, "after" d8b60ed, "develop" 5b0c985 = develop with fix(semantics): derive a declaration's implied generals once per symbol #1350, "fix" this branch):

  package / test	before	after	develop	fix
  cmd/sysml -run TestExploreTablesTheSpacecraftRaceWithinItsBudget	154.9 s	211.1 s	158.4 s	157.5 s
  internal/exec/runtime (whole package, no race)	193 s wall	238 s wall	190 s wall	209 s wall (shared the machine with another test for ~1 min; profile below)

  CPU profile of TestExecutionConformance: ImplicitGenerals 0.63 s / computeImplicitGenerals 0.54 s of 52 s sampled (develop: 12.2 s / 11.7 s of 415 s on the whole package); what remains under statedTypes is declaredTypes resolution, which perf(runtime): compute a feature's effective types once per model #1351 memoizes.

• gofmt -l . empty, go build ./..., go vet, go test of internal/semantic/..., internal/workspace/..., internal/check/..., internal/exec/runtime, cmd/sysml; python3 scripts/changelog.py check. Full gate on the branch: go build ./..., go vet ./..., gofmt -l . empty, go test -timeout 60m ./... (every corpus provisioned) exit 0, go test -C tools ./... exit 0, make lint passed.

Checklist

• make test and make lint pass locally
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved (compliance rows need nothing: the census is counted at docs build)
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

Link to Devin session: https://nasa-jpl-demo.devinenterprise.com/sessions/7cd14420fabe46389e22657ea286e92d
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/7cd14420fabe46389e22657ea286e92d?variant=devin
Requested by: @HuiJun

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring