fix(signals): a run that committed undefined is not WASTED_RECOMPUTE

[drakeo338]

Fixes #3715.

Summary

The WASTED_RECOMPUTE diagnostic counted a re-run as wasted whenever the computed's committed value was unchanged. A projection's internal computed always commits undefined (its work is draft writes or reconciling a returned value), so every projection re-run, and any memo that works by writing, was reported as producing the same value. The change passes a noValue flag into the check and skips the wasted count for runs that committed undefined.

How did you test this change?

Added attribution-wasted-recompute.test.ts with three cases (draft mutation, returned value to reconcile, memo that writes). Before the fix 3 of 8 tests fail; with it all 8 pass.

---

Maintainer addition (commit 7d723f8, pushed on top of the contributor's commit):

Maintainer follow-up

The undefined exemption now applies to changed itself in recomputeEnd, mirroring the existing effect-output rule, instead of a separate noValue flag read only by checkWastedRecompute. Every consumer of the per-run record now agrees with the finding. The draft-projection test also asserts that its rerun records report changed: true and that costs().wastedMs is 0.

Public API Changes

• RerunEvent.changed is now true for a plain run that commits undefined (projections, memos that do their work by writing). This moves every consumer of that field:
  • costs().scopes[].wastedMs no longer counts these runs;
  • expectNoWaste in @solidjs/diagnostics no longer fails on them;
  • the performance tracks no longer colour them as unchanged / wasted, and the why-run log no longer marks them unchanged.
• WASTED_RECOMPUTE no longer fires for these runs.

This matches the documented behavior (documentation/solid-2.0/08-dev-diagnostics.md, WASTED_RECOMPUTE: "a side-effect-only compute (undefined output) reports changed: true and is exempt"). No new exports, options or diagnostic codes.

[changeset-bot]

🦋 Changeset detected

Latest commit: 7d723f8

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 12 packages

Name	Type
@solidjs/signals	Patch
test-integration	Patch
@solidjs/web	Patch
@solidjs/babel-plugin	Patch
@solidjs/compiler	Patch
@solidjs/diagnostics	Patch
@solidjs/element	Patch
@solidjs/h	Patch
@solidjs/html	Patch
solid-js	Patch
@solidjs/universal	Patch
todos-server-example	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[codspeed]

Merging this PR will not alter performance

✅ 185 untouched benchmarks
⏩ 3 skipped benchmarks¹

---

_{Comparing drakeo338:claude/3715-fix (7d723f8) with next (da84bd9)}

Footnotes

1. 3 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩

[ryansolid]

Thanks @drakeo338 — the diagnosis in #3715 was spot on, and the tests (including the constant-memo control) made this easy to finish.

We pushed one follow-up commit on top of yours (7d723f8) rather than asking for another round:

• The exemption moved onto changed itself. With the separate noValue flag, only WASTED_RECOMPUTE skipped these runs; the per-run record still said changed: false, so costs().wastedMs, expectNoWaste in @solidjs/diagnostics, the performance tracks and the why-run log kept counting every projection run as waste — and the finding disagreed with costs() about the same runs. recomputeEnd now reports a run that committed undefined as changed, mirroring the existing effect-output rule, so every consumer reads the same fact. This is what 08-dev-diagnostics.md already documents ("a side-effect-only compute (undefined output) reports changed: true and is exempt").
• One added assertion: the draft-projection case now also checks that its rerun records report changed: true and that costs().wastedMs is 0.

We accepted the blind spot you'd expect from this rule: a memo that's genuinely stuck at undefined is no longer flagged. We looked for a precise signal instead — a "this run wrote" marker catches writing memos and draft writes, but a returning projection's reconcile lands after its run has ended, and writes to unread store keys fire no hook, so it can't cover projections without after-the-fact accounting. Identity of undefined proving nothing is already the rule for effects, so we kept it.

This lands in the next release, 2.0.0-rc.14.

— Claude via Cursor