[drakeo338]

Fixes #4145

Summary

When a memo throws while an async sibling holds the frame, stageError puts the error in _pendingValue and sets CONFIG_STAGED_ERROR. If the memo recovers before that frame commits, recompute stages the value over it but leaves the flag set. At the reveal, serve sees the flag and throws the value. That halts reactivity. The flag only got cleared in commitPendingNode, which runs after that read.

Now the branch that stages a pass's value also clears CONFIG_STAGED_ERROR.

How did you test this change?

New tests/held-error-recovery-4145.test.ts (the issue's repro, recovering to 1 and to 7). Before the fix, both cases halt with the value as the error and the sum stays 2. After it, they show 3 and 9.

pnpm build && pnpm test: signals 5355 passed. The only failure, in web's server-functions-adapter-request.spec.tsx, also fails on next without this change.

Written with help from an AI assistant (Claude). I reviewed and tested the change myself.

[changeset-bot]

🦋 Changeset detected

Latest commit: bda2ad6

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

This PR includes changesets to release 14 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
board-example	Patch
server-dbmon-example	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 degrade performance by 10.57%

⚠️ 4 benchmarks spent significant time in system calls

System calls cannot be consistently instrumented, so they are not included in the measure, which understates the real cost. Please switch to the Walltime instrument to accurately measure system calls.

Measurement and system calls

• build + consume
• propagation:diamond
• readAllowed
• middle: remove at 500 + insert at 500 (1000 rows)

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

❌ 1 regressed benchmark
✅ 195 untouched benchmarks
⏩ 5 skipped benchmarks¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
❌ ⚠️	build + consume	694.8 µs	776.9 µs	-10.57%

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing drakeo338:fix/held-error-recovery (bda2ad6) with next (fda29a4)}

Footnotes

1. 5 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 for this! Clearing CONFIG_STAGED_ERROR where recompute stages a pass's value is the right fix for #4145, and error → value → error inside one hold still throws the last error as it should. Two other paths stage a value over a held error and keep the stale flag, though:

1. Async settle. setSignal(el, () => value) in async.ts stages the value without clearing the flag. Repro: a memo that throws at step 1 and returns inner.promise at step 2, beside the issue's held slow sibling. setStep(1), setStep(2), inner.resolve(5), request.resolve(2) ends in [REACTIVITY_HALTED] 5 (either resolve order).
2. Writable memo set inside an action. action(function* () { setStep(1); yield; setValue(5); }) where step() === 1 throws: the effect's error arm receives 5.

Clearing the flag in setSignal right after el._pendingValue = v fixed both in our probes, with the signals suite still green. An option that might also be smaller: drop the flag and have the readers test _pendingValue instanceof StatusError, which covers every staging path at once.

Size: the Size check fails. It's +14 B minified everywhere, which puts signals: core floor and signals: + until/refresh at +22 B, 2 B past the 20 B allowance. Please trim under it, or add a Size-Exception: line to the PR body with the reason.

Could you add the two repros above as tests too? Note that #4143 touches the same staged-error area (setMemo), so whichever lands second may need a small follow-up.

— Claude via Claude Code

---

Generated by Claude Code

[ryansolid]

Thank you for this, and for the clear repro. #4156, merged to next, now clears CONFIG_STAGED_ERROR at the start of every pass, which includes the clear this PR adds. #4145 no longer reproduces on next, and this PR's test passes there without the change, so I'm closing this one as superseded. The action-write case from the review is being fixed together with #4143.

— Claude via Claude Code

---

Generated by Claude Code