Repository navigation
[2.0 / next] Recovering before a held error commits throws the successful value #4145
Description
Describe the bug
A memo throws while an async sibling keeps the rendered update held. Before the request finishes, the memo recomputes successfully. When the request then resolves, Solid throws the successful recovery value as an error, reports [REACTIVITY_HALTED], and leaves the old content on screen.
Your Example Website or App
https://s.olid.uk/id/df_wYAiPRlmgfLmezgxTag
Steps to Reproduce the Bug or Issue
Click each button once, in order; rebuild or reload to repeat.
- Initially the page shows
Sum: 2: both memos return1. - Click 1. Start update.
slowbecomes pending andvaluethrowsError("boom"). The previous sum stays visible. - Click 2. Recover.
valuesuccessfully returns1while the request is still pending. - Click 3. Finish request. The request resolves to
2. - The page stays at
Sum: 2. The console reports[REACTIVITY_HALTED]with the successful value 1 as the error.
Control: remove if (step() === 1) throw new Error("boom"); and rebuild. The same clicks produce Sum: 3.
Changing the recovery value to 7 makes Solid throw 7 instead; the sum stays 2 instead of becoming 9.
Expected behavior
The successful recomputation should replace the proposed error. Once the request resolves, the page should show Sum: 3 without throwing the recovered value or halting reactivity.
Screenshots or Videos
No response
Platform
- OS: Linux
- Browser: Chromium 154.0.8037.57
- Solid: playground pinned to
next@d2ad8b288886a5c7171e6746ccc019ad1d08d8ab, also the currentnexthead when checked on 2026-10-19. Reproduces with development and production builds.
Additional context
Follow-up to the error-outcome audit in #3886, which was closed in favor of the D-41 implementation in #4085. The related S-43 test in error-is-an-answer.test.ts covers recovery after the previous error has committed. This case replaces an error proposal with a successful value before that error commits, while the same frame is still held.
Related reports, with different triggers:
- [2.0 / next] A visible button's condition throws in its click handler during a held update #4109 concerns ordinary outside reads throwing before a held frame reveals. This repro fails during the eventual reveal, after the memo has recovered.
- [2.0 / next] A direct write does not clear a writable memo's previous error #4110 concerns a direct write to a failed writable memo. This repro uses successful recomputation of an ordinary memo; there is no writable memo setter.
- [2.0 / next] Outside latest reads of a derivation expose errors and values at different times #4106 was fixed by fix(signals): a held error reaches a derivation of latest when a value would #4129 for error/value publication through
latest()derivations. This repro uses nolatest()and still fails on a commit containing that fix.
Likely cause: stageError stores the proposed error in _pendingValue and sets CONFIG_STAGED_ERROR. The successful recomputation replaces _pendingValue with 1 but leaves that flag set. A subsequent tracked read in serve therefore throws the successful value. The flag is normally cleared by commitPendingNode, after the failing read in this schedule.
The component uses one signal, two memos, a native Promise and ordinary JSX, with no timers or explicit effects/flushes. Source-level controls reproduce both same-value and changed-value recovery failures with immediate and native delivery; the no-error controls succeed.
Activity
fix(signals): a later pass replaces a staged error (#4109)
fix(signals): stage a pass's error before the flush joins its hold (#…
ryansolid commentedon Oct 20, 2026
Generated by Claude Code