@esportsplus/reactivity 0.32.0 → 0.34.0

package/.claude/CHANGELOG.md

@@ -1,43 +0,0 @@
-# Changelog
-
-## Completed
-- [relocate-benchmarks] 567bd82. Relocate 7 benches to bench/ mirror + config, delete tests/ — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- [relocate-test-suites] 26fa1017. Relocate 23 suites to test/ mirror + config — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- [waitfor-test-helper] fa0a934a. waitFor/tick helper replacing 50 fixed-sleep sync sites — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- [compiler-import-order] 2610d50d. Normalize import order across compiler modules — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- [reactive-index-cleanup] 00b21810. reactive() facade unknown-bridges + import order — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- [reactive-array-cleanup] e0aa8b86. ReactiveArray listener typing, sort map, accessor order, imports — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- [system-any-casts] b890f047. Eliminate any casts in system.ts (unknown-bridged pool sentinels) — landed by direct-implement after blocked run 36ee71d0 (harness gate false-block: depless worktrees, issue 20260720T141515Z). Gate: pnpm agent:test green (tsc --noEmit + 369 tests), benches discover 7. · spec: spec-coding-standards
-- abandoned spend: ~$199.51 equivalent API cost across 7 terminal non-COMPLETED item(s) (system-any-casts, reactive-array-cleanup, reactive-index-cleanup, compiler-import-order, waitfor-test-helper, relocate-test-suites, relocate-benchmarks); recorded, not divided into per-item costs (run 36ee71d0)
-- run-level spend: ~$0.98 equivalent API cost (warm implementer runs + unit-scoped + boundary seats; never divided across items; blended rate = mean(input, output, cache_read, cache_creation) per million tokens) (run 36ee71d0)
-- [constants-export-order] 84f0b3613de0076e070ba4ddcd69c3e0d68a9f3a. Deviations: none; ladder: 3 attempts (critic → replanner completed → critic), completed. Cost: ~$20.03 equivalent API cost (blended-rate approximation). · spec: spec-coding-standards
-- [compiler-const-enum] 1b252cd2180063eefc04bf2b221ce4e9eff969a5. Deviations: none; ladder: 3 attempts (critic → replanner completed → critic), completed. Cost: ~$34.13 equivalent API cost (blended-rate approximation). · spec: spec-coding-standards
-- [types-variadic-any] da6b4873d4238db26d0d75f19a67f8c966a01b83. Deviations: none; ladder: 3 attempts (critic → replanner completed → critic), completed. Cost: ~$47.04 equivalent API cost (blended-rate approximation). · spec: spec-coding-standards
-- [reactive-object-cleanup] 769d007e46a9f6b0b299c2c3d48fa55dffe29628. Deviations: none; ladder: 3 attempts (critic → replanner completed → critic), completed. Cost: ~$32.61 equivalent API cost (blended-rate approximation). · spec: spec-coding-standards
-- cost source of truth: ccusage per-model breakdown (local transcripts carry the input/output/cache split the run journal does not) — figures above are a blended-rate approximation per contracts/models.json pricing (run 36ee71d0)
-- [recursion-free-walks] 143d5591. Recursion-free notify(), update(), and dispose()/unlink() walks — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next-2
-- [unobserved-hooks] b7a93aa4. onUnobserved() — last-subscriber lifecycle hook — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next-2
-- [invalidate] 60df87ae. invalidate() — standalone re-derivation of any computed — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next-2
-- [custom-equals] a4cb4e04. Custom equals on signal() and computed() — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next-2
-- follow-up discharged (completed) from spec-signals-next: pending-only-writes: complete full shape — selector recursive fan-out + nextPend… — evidence: 6a9b8e6
-- abandoned spend: ~$206.41 equivalent API cost across 4 terminal non-COMPLETED item(s) (custom-equals, invalidate, unobserved-hooks, recursion-free-walks); recorded, not divided into per-item costs (run 3306873d)
-- run-level spend: ~$2.11 equivalent API cost (warm implementer runs + unit-scoped + boundary seats; never divided across items; blended rate = mean(input, output, cache_read, cache_creation) per million tokens) (run 3306873d)
-- [async-iterable-resolve] 17b3878753d238869fdf6723dff468705d2795f8. Deviations: deviated — resolve()'s 'rejects when the tracked expression throws' was split into its own it() rather than combined into one test with the resolve-success case, matching the file's one-scenario-per-it convention from tests/async-hardening.ts; both scenarios from Design test 6 are covered.; added — Verified async generator .return() timing empirically (a queued return() on a generator suspended mid an unresolved internal await does not settle until that await resolves) via a throwaway Node script before writing tests 1-3, since the Design's prose doesn't spell out that an abandoned generator's pending gate must be released for its finally to run; tests 1-3 release the old generation's gate to observe closure, consistent with test 3's own description.; added — Added a gate()/tick() test-local helper pair (not in Design) to keep the generator-gating boilerplate readable across tests 1-5; mirrors async-hardening.ts's resolver-array convention.; ladder: 1 attempt (critic), completed. Cost: ~$16.19 equivalent API cost (blended-rate approximation). · spec: spec-signals-next-2
-- cost source of truth: ccusage per-model breakdown (local transcripts carry the input/output/cache split the run journal does not) — figures above are a blended-rate approximation per contracts/models.json pricing (run 3306873d)
-- abandoned spend: ~$229.74 equivalent API cost across 5 terminal non-COMPLETED item(s) (custom-equals, invalidate, unobserved-hooks, async-iterable-resolve, recursion-free-walks); recorded, not divided into per-item costs (run 2f191933)
-- run-level spend: ~$0.56 equivalent API cost (warm implementer runs + unit-scoped + boundary seats; never divided across items; blended rate = mean(input, output, cache_read, cache_creation) per million tokens) (run 2f191933)
-- [untrack-peek] 5d2a577c2b941bdc6fe869951c770ed3c12a82f8. Deviations: deviated — pull(node) extracted only the notified-broadcast + observer-nulled update() call (matching design's literal extraction target); drainPending() and the height>=heap_i||NOTIFY_MASK gate stay inline in read() exactly as before for byte-identical behavior.; deviated — peek() has no separate explicit NOTIFY_MASK pre-check before calling pull() — it always drains pending writes then calls pull(node) unconditionally for a STATE_COMPUTED node, relying on pull()'s internal notify-broadcast (which sets DIRTY/CHECK bits) plus update()'s own gv/CHECK/DIRTY branching to no-op cheaply when nothing changed. A literal external NOTIFY_MASK gate in peek() would check the bit *before* the broadcast that sets it, which never fires on a freshly-written-but-undrained dependency and would make peek() return stale values, contradicting Acceptance clause 3's 'up-to-date value for a dirty computed'. Verified via a dedicated test that peek() sees a fresh value synchronously right after write(), with 0 regressions across the full 343-test suite.; ladder: 1 attempt (critic), completed. Cost: ~$20.51 equivalent API cost (blended-rate approximation). · spec: spec-signals-next
-- [flush-batch] 3bc415d5568da53154f59a32069afaebfc381d04. Deviations: deviated — flush() implemented as `while (stabilizer === STABILIZER_SCHEDULED) stabilize();` instead of the Design's single `if`. Traced the RESCHEDULE path: a write() during an effect's recompute only flips stabilizer to RESCHEDULE and queues the *next* microtask when its subscriber's height is at or below the current heap_i already scanned in this pass (drain-before-scan only re-triggers on a later height iteration reached within the same for-loop) — a single stabilize() call leaves that tail unsettled until the next microtask, violating Acceptance clause 1's 'settles before flush() returns'. The while-loop drains it synchronously while still satisfying the no-op cases (RUNNING/RESCHEDULE/IDLE never equal SCHEDULED) and clause 4 (any stray queued microtask fires against an already-empty/IDLE state and no-ops).; ladder: 1 attempt (critic), completed. Cost: ~$11.27 equivalent API cost (blended-rate approximation). · spec: spec-signals-next
-- cost source of truth: ccusage per-model breakdown (local transcripts carry the input/output/cache split the run journal does not) — figures above are a blended-rate approximation per contracts/models.json pricing (run 2f191933)
-- follow-up discharged (completed) from spec-signals-next: read-version-dedup + pending-only-writes: stamp rv (+ nextPending on Signal entr… — evidence: 6a9b8e6
-- [cleanup-hardening] 14a40dd7. cleanup() hardening — run every disposer, never abort the batch — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next
-- [async-computed-hardening] 033b205f. asyncComputed hardening — dirty-gap guard, isPending, leak test, ReactiveObject reuse — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next
-- [async-error-propagation] bf8ad5c0. asyncComputed error propagation (rejections surface, ownership fixed) — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next
-- [computed-error-caching] 79224d08. Computed error caching + propagation (errors never vanish) — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next
-- [benchmark-harness] 9d6d7b2c. Canonical reactivity benchmark suite + glitch-freedom regression tests — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next
-- [test-bench-scripts] 524d07bc. Add test/bench script aliases (agent:test, agent:bench) — reconciled (killed-run recovery). Deviations: unrecorded (the run died before its boundary). Cost: ~$UNMEASURED (manual pre-run, no journal)~ · spec: spec-signals-next
-- [read-version-dedup] dcbaae90. O(1) interleaved repeat-read dedup (read-version stamp) — landed by hand (combined perf integration, bench-gated vs main: 0 regressions). Deviations: SelectorSignal entry-literal rv stamp deferred until signal-is-selector lands (superset-shape rule, follow-up); computed object-size assertion reconciled to 15. Cost: ~$UNMEASURED (manual integration, no journal)~ · spec: spec-signals-next
-- [global-version-fast-path] dcbaae90. globalVersion fast path for clean-graph reads — landed by hand (combined perf integration, bench-gated vs main: 0 regressions). Deviations: computed object-size assertion reconciled to 15 (shared bump with read-version-dedup + pending-only-writes). Cost: ~$UNMEASURED (manual integration, no journal)~ · spec: spec-signals-next
-- [pending-only-writes] dcbaae90. Pending-only writes — defer fan-out off the write hot path (reduced-shape core) — landed by hand (combined perf integration, bench-gated vs main: 0 regressions). Deviations: reduced-shape core only (intrusive nextPending queue, drainPending, per-height + read()-preamble drain); full shape (lazy mark-don't-schedule, selector recursive fan-out + nextPending on the SelectorSignal entry, tests/flush.ts invariant 2) deferred to follow-up pending flush-batch + lazy-computeds + signal-is-selector; SelectorSignal entry-literal nextPending/rv stamp deferred until signal-is-selector; computed object-size assertion reconciled to 15. Cost: ~$UNMEASURED (manual integration, no journal)~ · spec: spec-signals-next
-- [signal-is-selector] 6a9b8e63. signal.selector() per-key selector primitive — O(2) write-path fan-out, lazy per-key entries, eviction on last unsub — landed by hand off main (dcbaae9). Deviations: SelectorSignal entry literal pre-declares nextPending + rv (perf superset-shape, discharges follow-up #1); write-path bench informal (feature, not perf) — one predicted-not-taken keys null-check, write-neutral; suite 328→336 (+8 acceptance tests incl. SameValueZero NaN/object-ref caveats); glitch-freedom green. Cost: ~$UNMEASURED (manual, no journal)~ · spec: spec-signals-next
-- [lazy-computeds] REVERTED (no landing sha). Lazy computeds — compute on first read, sleep while unobserved — removed from scope by user decision 2026-07-19: lazy-by-default was implemented + A/B benchmarked and REGRESSED (full-suite ABBA geomean 1.04; read-class 0.77–0.88× eager; steady-read parity disproven by isolation bench). Do not re-attempt lazy-by-default naively; a non-regressing opt-in approach would be a fresh design. Patch preserved at ~/.claude/storage/lazy-bench/lazy-by-default-full.patch. Feature file removed from the spec per user request. Cost: ~$UNMEASURED (manual, no journal)~ · spec: spec-signals-next