solidjs / solid / 35842357635

74%

main: 89%

push

github

feat(signals): warn in dev when an async computation reads after await (#3602)

* feat(signals): warn in dev when an async compute reads after await

Reads made after the first await in an async computation are untracked, so the computation keeps a stale result when they change. Dev builds now attribute such reads through V8 async stack traces and report UNTRACKED_READ_AFTER_AWAIT.

* fix(signals): warn once per store and release unsettled flights

Store reads after await warned for every array index and prototype method; the check now skips non-own keys and reports once per store object per computation. Tail-flight entries hold their computation and promise weakly and are swept when superseded or disposed, so a flight that never settles no longer pins its node.

* fix(signals): check inherited store keys against the committed backing

* fix(signals): warn for refetching sources and bound the flight sweep

A refetching source serves its old value to an untracked read instead of throwing, so exempting every pending source missed that stale read; only a source still loading its first value is exempt now. The weak-entry sweep ran on every registration, making N concurrent flights cost O(N^2); it now runs when the registry doubles. Also lists the new code in the solid-js diagnostics skill.

* fix(signals): UNTRACKED_READ_AFTER_AWAIT ignores Solid-run callbacks on the continuation's stack

Effect callbacks, cleanups, and action bodies run with no owner, exactly like
an async continuation. When a continuation called flush(), dispose(), or an
action, the reads those callbacks made were on the continuation's async
stack and were blamed on the computation that had awaited.

Two dev-only depth counters in core.ts mark those scopes: `callbackDepth`
(effect callbacks in effect.ts, the synchronous slice of an action body in
action.ts; both sites already have try/finally) and `disposalDepth` (the
`_disposal` drain and the effect-returned cleanup in owner.ts). The d... (continued)

592 of 865 branches covered (68.44%)

Branch coverage included in aggregate %.

873 of 1121 relevant lines covered (77.88%)

19.81 hits per line