solidjs / solid / 41125818647

78%

main: 89%

push

github

fix(signals): net staged rewrites a custom equals calls equal to the committed value (D-8, #4483) (#4488)

* fix(signals): net staged rewrites a custom equals calls equal to the committed value (D-8, #4483)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015hPxMkC3jg3Uivh65gvaf2

* fix(signals): gate the custom-equals netting on the verdict engine

Only isPending/latest read the netted test before a commit, so the
write-back staging is gated on observeFlightHook and bundles without the
verdict engine pay 0 B. Drops the floor re-records and Size-Exception.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015hPxMkC3jg3Uivh65gvaf2

* chore(size): re-record over next @ d0fc2851

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015hPxMkC3jg3Uivh65gvaf2

* perf(signals): skip the custom-equals netting under the default comparator

A rewrite loop over a staged signal (CodSpeed propagation:avoidable,
updateSignals:update1to1000) ran the uninitialized/shown-error checks and
isSame on every write. Under the default comparator only the committed
object is equal and `v` is already it, so one load of `_equals` decides.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_015hPxMkC3jg3Uivh65gvaf2

* perf(signals): share setSignal's comparator test between the gate and D-8 netting

CodSpeed (simulation, --no-opt so setSignal runs as Sparkplug code) put
`propagation:avoidable` 5.3% behind next: every staged rewrite paid a
fresh `_equals` load and a strict compare for the netting, ~120
instructions a write under callgrind, on a bench that is 2,000 staged
rewrites per iteration.

With the verdict engine installed, `custom` (`_equals !== isEqual`) is now
taken once: the gate runs the default comparator's `===` inline instead of
calling `isSame`, and the nett... (continued)

790 of 1090 branches covered (72.48%)

Branch coverage included in aggregate %.

1125 of 1362 relevant lines covered (82.6%)

29.69 hits per line