solidjs / solid / 35961628461

74%

main: 89%

push

github

feat(signals): WASTED_RECOMPUTE — runs the equality gate discarded (#3613)

* feat(signals): WASTED_RECOMPUTE — a scope whose equality gate closes almost every time

The mirror of UNSTABLE_MEMO_OUTPUT. There the gate never closes; here it
closes on nearly every run: an input changed, the scope re-ran, computed,
compared equal to its last value and told nobody — the compute bought
nothing. `costs().wastedMs` already sums this for whoever imports the fold;
the finding names it while it happens, with the input that keeps
triggering it, and points the fix upstream: an equality boundary on the
part of the input the scope reads, or a narrower read.

Fires once per window per scope when, within `windowMs`, the scope ran
`minRuns`+ times, `ratio`+ of them unchanged, for `budgetMs`+ of compute
(defaults 5 / 0.8 / 2ms / 1000ms). Plain runs only — held and overlay runs
may be replayed and are never blamed. The sixth cost check `checks: false`
folds off; the per-scope window is an engine WeakMap.

Size: engine +294 B (cap 29.00 -> 29.30 KB); tier +47 B of mangler drift
with byte-count-identical core artifacts (cap 17.45 -> 17.50 KB).

Co-Authored-By: Claude via Cursor <cursoragent@cursor.com>
Co-authored-by: Cursor <cursoragent@cursor.com>

* perf(signals): WASTED_RECOMPUTE on node fields; an engine-cost tripwire; the lean-posture proposal

The check runs on every re-run, so its window moves from a WeakMap plus a
clock read (~15 ns/re-run) to `_devWaste*` fields on the node and the
run's own `at` (~2 ns) — the shape hotRuns/hotTime already use.

`attribution-engine-cost.test.ts` caps the enabled/idle per-re-run ratio
the way observe-idle-cost caps the tier: same workload, interleaved,
best-of-k, cap 14 against a measured ~10. Nothing else in the suite would
notice a check that grew a map lookup or an allocation on the hot path.

The plan gains a Cost section with the measurements — the six checks are
~50 ns of the engine's ~490 ns/re-run; the RerunEvent itself is... (continued)

606 of 894 branches covered (67.79%)

Branch coverage included in aggregate %.

883 of 1142 relevant lines covered (77.32%)

24.87 hits per line