[devin-ai-integration]

What and why

In the 0.11.0 release-gate measurement internal/exec/runtime BenchmarkNewRootFrame reads 36 → 55 allocs/op against 0.10.1 (12.5 → 13.8 KiB/op). The allocation profile puts the whole difference under Context.holdFeature, which since #1057 asks semantics.Model.AllRedefinedFeatures for every feature a root frame holds (0.10.1 asked the memoized RedefinedFeatures, explicit clauses only). AllRedefinedFeatures rebuilt its visited set and visit closure on every call, and the implicit parameter redefinitions it walks (implicitParameterTargets) rebuilt the owner's owned-parameter list (positionalParameters(ownedParameters(owner))) on every query to find the parameter's position — a list parametersOf(owner) already memoizes, and whose owned entries come first in the same order.

Changes, all in internal/semantic/semantics:

• AllRedefinedFeatures is a journaled per-symbol table (allRedefined, keyed like nestedRedefs and invalidated with the declaration), so the closure of a feature is computed once per model. The walk runs under a resolver lookup frame and is cached only when that frame settled: a closure cut short by a redefinition still resolving its clause, or by an owner whose parameter list is still being computed, is answered but recomputed on the next query — the same rule RedefinedFeatures follows.
• parametersOf records its lookup depth and cuts a re-entrant reader short (computingParams, like computingRedefined) instead of answering it silently with the empty seed, and does not memoize its own list when its frame did not settle.
• implicitParameterTargets reads the parameter's position from m.parametersOf(owner).positional instead of rebuilding the owned list.

Same redefinitions, same positions; only the repeated work goes.

How it was verified

• Failing-before test TestRedefinedFeaturesAreComputedOncePerSymbol (internal/semantic/semantics/implicit_parameter_memo_test.go): AllRedefinedFeatures(Proc::a) is [Base::x] and a repeated query allocates 0 times (7 on develop d8b60ed62), and a repeated ImplicitParameterRedefinitions query allocates at most its result beyond what the owner's ImplicitGenerals costs (a differential bound, so it holds with or without fix(semantics): derive a declaration's implied generals once per symbol #1350).

• Guard tests: TestRedefinitionClosureIsNotCachedWhileAGuardCutsItShort queries the closure under an open RedefinedFeatures guard and checks that the next query, after the guard lifts, returns the target (the first commit of this branch returned []); TestParameterListGuardCutsDependentQueriesShort does the same under an open parametersOf guard.

• A/B over six interleaved rounds on one machine, allocations exact (benchstat p=0.002 on every moved row). The benchmark did not exist on develop before the 0.10.1 back-merge, which is newer than feat(runtime): accepts, counted steps and recursive typed bodies inside action bodies #1057, so the pre-regression column is the 0.10.1 tag itself. Columns: v0.10.1 1c45b6e17 | develop d8b60ed62 + fix(semantics): derive a declaration's implied generals once per symbol #1350 | this change + fix(semantics): derive a declaration's implied generals once per symbol #1350:

  row	allocs/op	B/op
  NewRootFrame	36 | 55 | 36	12.48 Ki | 13.79 Ki | 12.54 Ki
  DerivedReadWriteRead	65 | 73 | 73	2.511 Ki | 2.699 Ki | 2.699 Ki
  MaterializePlainAttributes	635 | 672 | 672	45.36 Ki | 53.88 Ki | 53.88 Ki

  NewRootFrame is back to the 0.10.1 allocation count (36 with the lookup frames too); the other two rows are not this change's (their movement bisects to fix(runtime): type an object by every type its feature inherits through subsetting #1276 and is addressed in perf(runtime): compute a feature's effective types once per model #1351).

• go build ./..., go vet ./..., gofmt -l . empty, go test ./internal/semantic/semantics ./internal/check/... ./internal/ir/... ./internal/exec/runtime, python3 scripts/changelog.py check.

Checklist

• make test and make lint pass locally (the packages above locally; the full suite runs in CI)
• Tests added or updated for the change
• Documentation extended where it already covers the surface (no doc covers this internal)
• 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 (no gate count moved)
• 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/cf22d7a746814d20b2f0eaf11682fef9
Open in Devin Desktop: https://nasa-jpl-demo.devinenterprise.com/desktop/session/cf22d7a746814d20b2f0eaf11682fef9?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