fix(normalizr): Track INVALID and null nested entities as memo dependencies by ntucker · Pull Request #4460 · reactive/data-client

ntucker · 2026-10-24T21:54:37Z

Requested by Nathaniel

Motivation

Bug fix. A component reads a parent entity, and a nested entity of that parent is INVALID (it failed validation) or null in the store.

Example: an Article has author: '2', and User '2' is INVALID. Then a valid User '2' arrives. Expected: the article shows the new author. Actual: the cached article keeps the old, empty author.

Solution

Behavior: a parent recomputes when a nested INVALID or null entity changes, as it already did for a missing one. unvisitEntity records these entities as dependencies. Blank ids are still not dependencies, because they cannot be looked up.
Internal: WeakMap keys must be objects. depKey() gives each non-object value a stable object stand-in.

Assumptions and validation

High: the INVALID and null tests in MemoCache.ts fail on master and pass here. The unit suite passes.
High: the memo invariants, now written in packages/normalizr/AGENTS.md, hold. fix(normalizr): Keep memo WeakMaps one level deep #4459 builds on them.
Medium: I ran coverage on the memo and denormalize files only. Each new branch is covered.
High: speed, measured. Six alternating master/fix(normalizr): Track INVALID and null nested entities as memo dependencies #4460 rounds of examples/benchmark and three of benchmark-react (data-client only) on one quiet machine (Ryzen 9 7950X, WSL2, Node 26). Between runs, a case moves by up to about ±5%, so only larger changes are real. Full tables are under "Benchmarks" below.
- Cached reads of lists with nested null or INVALID children cost 16–21%, with 25% of children affected. They are now dependencies, so each cached read checks them, as it already checks missing children on master. This cost is the fix.
- Cold reads of those lists cost 6–8%. A child that is not an object now records only its dependency: no closure and no entity-cache entry, because there is no value worth caching. This also makes the 25%-missing cold read 10% faster than master. Measured in three more alternating rounds (master, the previous head, this head); see "Non-object children" under "Benchmarks".
- Other cached reads are 0–6% slower (worst: denormalizeLong All withCache -6.4%, Query-sorted -4.5%, queryShort 500x -4.4%). That is the depKey() call on each dependency check. A smaller depKey for V8 to inline measured no better, so I kept this one.
- Cold reads, normalize and writes are within ±2%.
- benchmark-react: all scenarios are within their run-to-run spread (5–17%). update-entity-sorted (-10.9%) and delete-item (-7.7%) are the largest changes, but those two vary the most between runs.

GOALS impact

Data integrity: positive. Views stay consistent with the store.
Speed: slightly negative. Cached reads of lists with nested null or INVALID children are 16–21% slower, because those children are now tracked (the fix); cold reads of them are 6–8% slower, and missing children read 10% faster cold. Other cached reads are 0–6% slower from depKey(). Other cold reads and writes are unchanged.
Bundle size: +80 B brotli to @data-client/normalizr (+2.3%), against current master. The non-object path adds 8 B of that.
V8: neutral. No deopt loops and no new deopt reasons; the last cleanup measured 617 deopt events against 629 before it, in one run each. No inline-cache site is worse than on master, and the memo functions reach the same tiers. For the non-object path: dependOn optimizes (Maglev) with no deopts, its call site stays monomorphic, and getEntity has fewer deopts than before (no "wrong feedback cell").

Benchmarks

Non-object children, examples/benchmark, ops/s, median of 3 alternating rounds

case	master	before (`de8c335`)	now	now vs master
denormalizeLong nested 25% missing	495	506	546	+10.3%
denormalizeLong nested 25% null	587	519	552	-6.0%
denormalizeLong nested 25% INVALID	602	507	554	-8.0%
denormalizeLong nested 25% missing withCache	11,951	11,466	11,491	-3.8%
denormalizeLong nested 25% null withCache	14,923	11,781	11,889	-20.3%
denormalizeLong nested 25% INVALID withCache	14,731	11,683	11,559	-21.5%
denormalizeLong	496	479	477	-3.8%
denormalizeLong withCache	11,879	11,016	11,415	-3.9%

examples/benchmark, ops/s, median of 6 alternating rounds

case	master	4460	change	rounds
normalizeLong	767	772	+0.7%	6/6
normalizeLong Values	674	678	+0.5%	6/6
normalizeLong Scalar	664	664	+0.0%	6/6
normalizeLong Scalar update	1,651	1,628	-1.4%	6/6
denormalizeLong	537	527	-1.9%	6/6
denormalizeLong Values	480	476	-0.8%	6/6
denormalizeLong donotcache	1,896	1,914	+0.9%	6/6
denormalizeLong nested 25% missing	560	556	-0.8%	6/6
denormalizeLong nested 25% null	672	558	-16.8%	6/6
denormalizeLong nested 25% INVALID	670	556	-17.0%	6/6
denormalizeLong Values donotcache	1,320	1,323	+0.2%	6/6
denormalizeLong Scalar donotcache	2,067	2,014	-2.6%	6/6
denormalizeShort donotcache 500x	2,439	2,480	+1.7%	6/6
denormalizeShort 500x	1,310	1,312	+0.1%	6/6
denormalizeShort 500x withCache	13,206	13,066	-1.1%	6/6
queryShort 500x withCache	6,308	6,029	-4.4%	6/6
buildQueryKey All	96,040	96,362	+0.3%	6/6
query All withCache	9,882	10,104	+2.3%	6/6
denormalizeLong with mixin Entity	498	501	+0.6%	6/6
denormalizeLong withCache	11,959	11,890	-0.6%	6/6
denormalizeLong withCache (Scalar churn)	11,900	11,872	-0.2%	6/6
denormalizeLong nested 25% missing withCache	12,222	12,328	+0.9%	6/6
denormalizeLong nested 25% null withCache	15,155	12,494	-17.6%	6/6
denormalizeLong nested 25% INVALID withCache	14,868	12,448	-16.3%	6/6
denormalizeLong Values withCache	9,120	8,942	-2.0%	6/6
denormalizeLong Scalar withCache	13,876	13,398	-3.4%	6/6
denormalizeLong Scalar update withCache	7,706	7,454	-3.3%	6/6
denormalizeLong All withCache	11,964	11,194	-6.4%	6/6
denormalizeLong Query-sorted withCache	10,962	10,466	-4.5%	6/6
denormalizeLongAndShort withEntityCacheOnly	2,863	2,888	+0.9%	6/6
denormalize bidirectional 50	9,330	9,183	-1.6%	6/6
denormalize bidirectional 50 donotcache	78,436	78,418	-0.0%	6/6
getResponse	6,966	6,722	-3.5%	6/6
getResponse (null)	18,735,359	18,569,344	-0.9%	6/6
getResponse (clear cache)	464	466	+0.5%	6/6
getSmallResponse	7,400	7,085	-4.3%	6/6
getSmallInferredResponse	5,587	5,482	-1.9%	6/6
getResponse Collection	8,073	8,114	+0.5%	6/6
get Collection	7,020	7,084	+0.9%	6/6
get Query-sorted	8,532	8,207	-3.8%	6/6
setLong	781	788	+0.9%	6/6
setLongWithMerge	465	462	-0.6%	6/6
setLongWithSimpleMerge	492	491	-0.2%	6/6
setSmallResponse 500x	1,662	1,660	-0.2%	6/6
setMany 50x one-per-row	226	222	-1.6%	6/6
setMany 50 batch	6,519	6,428	-1.4%	6/6
setMany 500x one-per-row	23	23	-0.4%	6/6
setMany 500 batch	2,782	2,790	+0.3%	6/6

benchmark-react (data-client), ops/s, median of 3 alternating rounds

case	master	4460	change	rounds
getlist-100	21	21	+0.2%	3/3
getlist-500	13	13	-0.9%	3/3
update-entity	955	1,000	+4.8%	3/3
update-user	1,000	1,000	+0.0%	3/3
getlist-500-sorted	13	13	+0.8%	3/3
update-entity-sorted	801	714	-10.9%	3/3
update-entity-multi-view	833	801	-3.8%	3/3
list-detail-switch-10	33	35	+6.1%	3/3
update-user-10000	141	137	-2.8%	3/3
invalidate-and-resolve	10	10	+0.2%	3/3
unshift-item	588	572	-2.8%	3/3
delete-item	833	769	-7.7%	3/3
move-item	426	426	+0.0%	3/3

🤖 Generated with Claude Code

Note

Medium Risk
Changes core denormalize memo dependency tracking and cache lookup keys; correctness is heavily tested but cached-read paths see a small perf cost from depKey() on each dependency step.

Overview
Fixes stale parent denormalization when a nested child was INVALID, null, or missing and later becomes a valid entity—cached parents now recompute instead of keeping the old nested shape.

Memo / denormalize: unvisitEntity calls new cache.dependOn() for non-object store values (still skips blank ids). That records them in WeakDependencyMap without the old UNDEF + getEntity shortcut. depKey() maps undefined, null, INVALID, and other primitives to stable object keys for WeakMap chains.

Docs & tests: “Memo invariants” land in packages/normalizr/AGENTS.md (with agent rule pointers). Coverage adds MemoCache nested-dependency cases, WeakDependencyMap / depKey tests, React concurrent-memo.web.tsx, and benchmark variants for 25% nested missing|null|INVALID.

^{Reviewed by Cursor Bugbot for commit 595ead9. Bugbot is set up for automated code reviews on this repo. Configure here.}

…encies unvisitEntity returned early for a nested entity whose stored value was INVALID or null, without recording it as a dependency. A cached parent kept returning its old result after that entity became valid again. Both now go through cache.getEntity like a missing entity does, keyed by depKey(), which gives each non-object value a stable object stand-in. Documents the memo invariants in packages/normalizr/AGENTS.md, with the two accepted limitations, and adds a scoped agent rule pointing at them. Adds a ReactDOM test that counts memo misses under concurrent rendering. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

changeset-bot · 2026-10-24T21:54:41Z

🦋 Changeset detected

Latest commit: 595ead9

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 11 packages

Name	Type
@data-client/normalizr	Patch
@data-client/core	Patch
@data-client/react	Patch
@data-client/vue	Patch
example-benchmark	Patch
normalizr-github-example	Patch
normalizr-redux-example	Patch
normalizr-relationships	Patch
example-benchmark-react	Patch
test-bundlesize	Patch
coinbase-lite	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ntucker · 2026-10-24T21:56:13Z

Staff engineer (Cursor agent): Staff review at 6d2dbfb1 (Draft). Verdict: LGTM with follow-ups, nothing blocking.

The fix is the right size. Recording INVALID and null children through cache.getEntity, the same way a missing child already goes through UNDEF, closes the gap without a new mechanism, and depKey() is applied on both sides of WeakDependencyMap (set and both get paths), so stored and looked-up keys agree. depKey is idempotent on objects, so the deps GlobalCache records with an already-converted key round-trip fine.

Things to settle before this lands, in order of importance:

Landing order with fix(normalizr): Keep memo WeakMaps one level deep #4459. Both PRs add .cursor/rules/memo-invariants.mdc, .claude/rules/memo-invariants.md, the "Memo invariants" section in packages/normalizr/AGENTS.md, concurrent-memo.web.tsx, and a blog line, and both edit the get/set paths in WeakDependencyMap.ts. The texts already disagree: fix(normalizr): Keep memo WeakMaps one level deep #4459 says "all seven" and lists this bug as a "Known gap" under invariant 2, while this PR says "all of them" and has six. The PR body says fix(normalizr): Keep memo WeakMaps one level deep #4459 builds on this one, but fix(normalizr): Keep memo WeakMaps one level deep #4459 is Ready and this is Draft. Whichever lands second needs to drop its copies of the shared files, re-apply depKey() to the other PR's WeakDependencyMap shape, and remove the "Known gap" sentence (or add invariant 7), so the doc doesn't describe a bug that's already fixed.
Benchmark before Ready. depKey() now runs on every dependency compare in WeakDependencyMap.get, which is the hottest path in denormalize. The object case returns on the first check, so I expect the cost to be noise, but I'd like the Node numbers from the open question here before this goes Ready.
Optional simplification, your call. unvisitEntity now has three near-identical "record this as a dependency" blocks (missing, INVALID, null). They could be one branch: when inputIsId and the entity is null, a symbol, or undefined with a usable id, call cache.getEntity(entityOrId, schema, depKey(entity), ...) and store schema.denormalize(entity, delegate) for the symbol case or entity otherwise. The UNDEF comment block would then read as the explanation for all three. Separately, depKey's module-level Map could be two fixed constants (null and INVALID are the only non-object values that reach it from the store), which drops a lookup and any chance of the map growing if someone stores primitives in an entity table. Neither changes behavior, so it's fine to leave as is.

What I checked: I read the full diff, plus globalCache.ts on master and unvisit.ts at this head, to confirm the new branches feed getEntity the converted key and that paths() and GC only see {key, pk} paths (the same as the existing UNDEF case). I compared the shared files against #4459 at ecea0af6. I didn't run the tests or benchmarks; that the INVALID and null tests fail on master comes from the PR body, not from me.

github-actions · 2026-10-24T21:57:33Z

Docs preview: https://docs-site-pr-4460-data-client.vercel.app

Built from 595ead9 merged into master (this build). Locales: en (the merge queue builds every locale).

unvisitEntity handled missing, null and INVALID nested entities in three near-identical branches. One branch now records all three, and WeakDependencyMap maps its root key through depKey(), so callers pass the raw value. Tests reuse the shared Article and makeGetTodo fixtures. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ntucker · 2026-10-24T22:15:03Z

Staff engineer (Cursor agent): LGTM + FOLLOW_UP at 1be7879.

The refactor is a real entropy win: three near-identical unvisitEntity branches become one, and depKey() is now applied only inside WeakDependencyMap (root key in get, every key in set), so callers and the recorded dependencies[].entity carry the raw value and there's one normalization point instead of callers remembering to pass UNDEF/depKey(...). Blank-id behavior is unchanged (a blank id can't resolve to INVALID, so moving the symbol check after the dependency branch loses nothing).

Follow-up (no change needed here): this commit exists only on #4460. #4459 already merged an earlier cut of this branch, so the two now diverge again (this branch is 1 ahead, 10 behind #4459's head). Whichever lands first, the other needs this refactor folded in, or the three-branch version and the caller-side depKey calls come back on master. Small note for later: the merged branch now also records any other non-object entity value (e.g. a stray string) as a dependency through depKey's strong Map. That's harmless for real entity tables, but worth keeping in mind if one ever holds primitives.

Validated: I read the commit diff plus WeakDependencyMap.ts, UNDEF.ts and GlobalCache.getEntity at 1be7879, and code-searched for other UNDEF consumers (none outside those files). I did not run tests. The tightened avoidableMisses(...).toBe(0) assertions rely on CI, and CircleCI was still pending when I checked.

chatgpt-codex-connector · 2026-10-24T22:51:50Z

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-27T02:38:04.317255Z	`595ead9`	Manual request

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

github-actions · 2026-10-24T22:53:35Z

Size Change: +113 B (+0.1%)

Total Size: 112 kB

📦 View Changed

Filename	Size	Change
`examples/test-bundlesize/dist/rdcClient.js`	10.9 kB	+37 B (+0.34%)
`examples/test-bundlesize/dist/rdcNextjs.js`	12.4 kB	+34 B (+0.27%)
`examples/test-bundlesize/dist/rdcPipeableStream.js`	9.69 kB	+42 B (+0.44%)

ℹ️ View Unchanged

Filename	Size
`examples/test-bundlesize/dist/App.js`	1.45 kB
`examples/test-bundlesize/dist/polyfill.js`	301 B
`examples/test-bundlesize/dist/rdcEndpoint.js`	8.04 kB
`examples/test-bundlesize/dist/react.js`	68.6 kB
`examples/test-bundlesize/dist/webpack-runtime.js`	698 B

_{compressed-size-action}

chatgpt-codex-connector

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 1be7879abb

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

chatgpt-codex-connector · 2026-10-24T22:53:47Z

+/** `WeakMap` key for an entity value: the object itself, `UNDEF` for
+ * `undefined`, or a stable stand-in for any other value (`null`, `INVALID`). */
+export function depKey(entity: unknown): object {
+  if (typeof entity === 'object' && entity !== null) return entity;


Preserve function references as weak dependency keys

When an Entity class is passed to MemoCache.buildQueryKey(), BaseDelegate.tracked() records that constructor as the first dependency. Functions are valid WeakMap keys, but this guard excludes them, so every queried Entity constructor is now stored strongly in the module-global valueKeys map. Dynamically created schemas and anything their closures retain consequently survive even after their Controller and MemoCache are discarded; a forced-GC check confirmed this retention. Return functions unchanged alongside non-null objects rather than allocating permanent stand-ins for them.

AGENTS.md reference: packages/normalizr/AGENTS.md:L22-L22

Useful? React with 👍 / 👎.

github-actions

Benchmark React

Details

Benchmark suite	Current: `595ead9`	Previous: `8817f47`	Ratio
`data-client: getlist-100`	`215.08` ops/s (`± 7.6%`)	`222.22` ops/s (`± 8.5%`)	`1.03`
`data-client: getlist-500`	`80.97` ops/s (`± 4.4%`)	`73.27` ops/s (`± 6.4%`)	`0.90`
`data-client: update-entity`	`500` ops/s (`± 12.1%`)	`666.67` ops/s (`± 6.1%`)	`1.33`
`data-client: update-user`	`526.32` ops/s (`± 8.0%`)	`625` ops/s (`± 6.8%`)	`1.19`
`data-client: getlist-500-sorted`	`67.8` ops/s (`± 9.6%`)	`64.78` ops/s (`± 9.3%`)	`0.96`
`data-client: update-entity-sorted`	`384.62` ops/s (`± 5.6%`)	`444.66` ops/s (`± 10.7%`)	`1.16`
`data-client: update-entity-multi-view`	`454.55` ops/s (`± 9.5%`)	`555.56` ops/s (`± 7.0%`)	`1.22`
`data-client: list-detail-switch-10`	`22.78` ops/s (`± 9.9%`)	`20.64` ops/s (`± 8.0%`)	`0.91`
`data-client: update-user-10000`	`119.06` ops/s (`± 7.5%`)	`125.79` ops/s (`± 11.2%`)	`1.06`
`data-client: invalidate-and-resolve`	`65.36` ops/s (`± 5.4%`)	`60.61` ops/s (`± 7.1%`)	`0.93`
`data-client: unshift-item`	`357.14` ops/s (`± 4.8%`)	`384.62` ops/s (`± 0.0%`)	`1.08`
`data-client: delete-item`	`555.56` ops/s (`± 11.2%`)	`555.56` ops/s (`± 9.1%`)	`1`
`data-client: move-item`	`289.92` ops/s (`± 10.3%`)	`307.77` ops/s (`± 6.0%`)	`1.06`

This comment was automatically generated by workflow using github-action-benchmark.

github-actions

Benchmark

Details

Benchmark suite	Current: `595ead9`	Previous: `7adefee`	Ratio
`normalizeLong`	`694` ops/sec (`±6.35%`)	`693` ops/sec (`±3.56%`)	`1.00`
`normalizeLong Values`	`609` ops/sec (`±1.13%`)	`607` ops/sec (`±1.11%`)	`1.00`
`normalizeLong Scalar`	`547` ops/sec (`±3.02%`)	`544` ops/sec (`±2.39%`)	`0.99`
`normalizeLong Scalar update`	`1587` ops/sec (`±1.60%`)	`1574` ops/sec (`±1.86%`)	`0.99`
`denormalizeLong`	`346` ops/sec (`±6.09%`)	`349` ops/sec (`±5.61%`)	`1.01`
`denormalizeLong Values`	`314` ops/sec (`±4.52%`)	`322` ops/sec (`±3.99%`)	`1.03`
`denormalizeLong donotcache`	`1777` ops/sec (`±1.27%`)	`1820` ops/sec (`±2.03%`)	`1.02`
`denormalizeLong nested 25% missing`	`356` ops/sec (`±5.24%`)
`denormalizeLong nested 25% null`	`354` ops/sec (`±5.02%`)
`denormalizeLong nested 25% INVALID`	`380` ops/sec (`±4.59%`)
`denormalizeLong Values donotcache`	`1253` ops/sec (`±1.74%`)	`1151` ops/sec (`±5.18%`)	`0.92`
`denormalizeLong Scalar donotcache`	`1725` ops/sec (`±0.72%`)	`1909` ops/sec (`±0.48%`)	`1.11`
`denormalizeShort donotcache 500x`	`2481` ops/sec (`±1.03%`)	`2462` ops/sec (`±1.91%`)	`0.99`
`denormalizeShort 500x`	`905` ops/sec (`±6.47%`)	`899` ops/sec (`±7.30%`)	`0.99`
`denormalizeShort 500x withCache`	`11934` ops/sec (`±2.20%`)	`9181` ops/sec (`±2.20%`)	`0.77`
`queryShort 500x withCache`	`4629` ops/sec (`±1.15%`)	`5359` ops/sec (`±0.74%`)	`1.16`
`buildQueryKey All`	`88743` ops/sec (`±2.19%`)	`98489` ops/sec (`±0.41%`)	`1.11`
`query All withCache`	`9264` ops/sec (`±5.22%`)	`10256` ops/sec (`±5.77%`)	`1.11`
`denormalizeLong with mixin Entity`	`276` ops/sec (`±5.87%`)	`335` ops/sec (`±6.11%`)	`1.21`
`denormalizeLong withCache`	`11912` ops/sec (`±3.33%`)	`12871` ops/sec (`±0.43%`)	`1.08`
`denormalizeLong withCache (Scalar churn)`	`11921` ops/sec (`±2.19%`)	`12670` ops/sec (`±1.00%`)	`1.06`
`denormalizeLong nested 25% missing withCache`	`10927` ops/sec (`±0.34%`)
`denormalizeLong nested 25% null withCache`	`11668` ops/sec (`±1.63%`)
`denormalizeLong nested 25% INVALID withCache`	`11035` ops/sec (`±2.65%`)
`denormalizeLong Values withCache`	`9169` ops/sec (`±1.87%`)	`9804` ops/sec (`±1.03%`)	`1.07`
`denormalizeLong Scalar withCache`	`12965` ops/sec (`±1.15%`)	`13747` ops/sec (`±0.99%`)	`1.06`
`denormalizeLong Scalar update withCache`	`7234` ops/sec (`±1.42%`)	`7523` ops/sec (`±0.38%`)	`1.04`
`denormalizeLong All withCache`	`11590` ops/sec (`±0.26%`)	`13111` ops/sec (`±0.23%`)	`1.13`
`denormalizeLong Query-sorted withCache`	`9966` ops/sec (`±5.64%`)	`10436` ops/sec (`±6.87%`)	`1.05`
`denormalizeLongAndShort withEntityCacheOnly`	`2708` ops/sec (`±1.53%`)	`3035` ops/sec (`±1.13%`)	`1.12`
`denormalize bidirectional 50`	`6030` ops/sec (`±10.77%`)	`7229` ops/sec (`±9.22%`)	`1.20`
`denormalize bidirectional 50 donotcache`	`66925` ops/sec (`±2.59%`)	`74415` ops/sec (`±0.62%`)	`1.11`
`getResponse`	`7383` ops/sec (`±4.85%`)	`6850` ops/sec (`±6.61%`)	`0.93`
`getResponse (null)`	`17967514` ops/sec (`±1.88%`)	`19028450` ops/sec (`±3.52%`)	`1.06`
`getResponse (clear cache)`	`264` ops/sec (`±8.19%`)	`333` ops/sec (`±7.84%`)	`1.26`
`getSmallResponse`	`5048` ops/sec (`±2.21%`)	`6392` ops/sec (`±0.50%`)	`1.27`
`getSmallInferredResponse`	`4271` ops/sec (`±1.94%`)	`4838` ops/sec (`±0.54%`)	`1.13`
`getResponse Collection`	`7172` ops/sec (`±3.78%`)	`8037` ops/sec (`±2.20%`)	`1.12`
`get Collection`	`5402` ops/sec (`±0.37%`)	`6524` ops/sec (`±0.49%`)	`1.21`
`get Query-sorted`	`8124` ops/sec (`±3.88%`)	`7759` ops/sec (`±5.02%`)	`0.96`
`setLong`	`715` ops/sec (`±2.15%`)	`737` ops/sec (`±0.35%`)	`1.03`
`setLongWithMerge`	`406` ops/sec (`±1.80%`)	`415` ops/sec (`±1.86%`)	`1.02`
`setLongWithSimpleMerge`	`435` ops/sec (`±0.85%`)	`443` ops/sec (`±1.35%`)	`1.02`
`setSmallResponse 500x`	`1549` ops/sec (`±0.38%`)	`1611` ops/sec (`±1.41%`)	`1.04`
`setMany 50x one-per-row`	`207` ops/sec (`±0.50%`)	`220` ops/sec (`±1.51%`)	`1.06`
`setMany 50 batch`	`5849` ops/sec (`±0.40%`)	`6279` ops/sec (`±0.55%`)	`1.07`
`setMany 500x one-per-row`	`20.94` ops/sec (`±0.40%`)	`22.67` ops/sec (`±0.44%`)	`1.08`
`setMany 500 batch`	`2534` ops/sec (`±2.64%`)	`2821` ops/sec (`±0.35%`)	`1.11`

This comment was automatically generated by workflow using github-action-benchmark.

- depKey lives in WeakDependencyMap, its only user - unvisitEntity handles all non-object entities in one block - WeakDependencyMap tests null and INVALID dependencies directly - concurrent-memo test helpers: updateKey, versionsOf(state) - AGENTS.md: invariant 1 without implementation names; MAX_ENTITY_DEPTH - memo-invariants rule also applies to normalizr schemas Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

…-gaps

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Entity classes are functions, and buildQueryKey records them as the first dependency. depKey gave them a stand-in in a module-level Map, which kept every queried schema alive. Functions are valid WeakMap keys. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

chatgpt-codex-connector · 2026-10-25T02:39:04Z

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

- depKey's stand-in Map already covers undefined, so UNDEF.ts goes - unvisitEntity: no value closure; isBlankPk shared by both branches - tests: rows hold the store table; one settle() for held requests - memo-invariants rule just points at AGENTS.md Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

Every 4th nested BuildTypeDescription in the project list is missing from the store, null, or INVALID, cold and withCache. Each is now a memo dependency of the project that nests it. Export INVALID from the bundle. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ntucker · 2026-10-25T04:56:48Z

@codex review

chatgpt-codex-connector · 2026-10-25T04:58:23Z

Codex Review: Didn't find any major issues. 👍

Reviewed commit: dc95623b68

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

depKey runs on every memo dependency. Looking up undefined, null and INVALID in a Map made cached reads of lists with missing, null or INVALID children up to 33% slower than master; fixed objects need no lookup. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

A missing, null or INVALID child only needs its dependency recorded; its value is not worth caching. Skip the entity cache (closure, cache entry, arrays) and push the dependency directly. Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ntucker · 2026-10-27T02:36:22Z

@codex review

chatgpt-codex-connector · 2026-10-27T02:38:02Z

Codex Review: Didn't find any major issues. Bravo.

Reviewed commit: 595ead97a0

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

Open a pull request for review
Mark a draft as ready
Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

docs(blog): Note the INVALID/null memo dependency fix

6d2dbfb

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ntucker mentioned this pull request Oct 24, 2026

fix(normalizr): Keep memo WeakMaps one level deep #4459

Open

ntucker deployed to docs-preview October 24, 2026 21:56 — with GitHub Actions

View deployment

ntucker deployed to docs-preview October 24, 2026 22:16 — with GitHub Actions

View deployment

ntucker marked this pull request as ready for review October 24, 2026 22:51

chatgpt-codex-connector Bot reviewed Oct 24, 2026

View reviewed changes

github-actions Bot reviewed Oct 24, 2026

View reviewed changes

ntucker deployed to docs-preview October 24, 2026 23:01 — with GitHub Actions

View deployment

Merge remote-tracking branch 'origin/master' into fix/memo-dependency…

0409ea2

…-gaps

ntucker deployed to docs-preview October 25, 2026 00:41 — with GitHub Actions

View deployment

test(normalizr): Cover blank nested ids in unvisitEntity

acbfb64

Co-Authored-By: Claude Opus 5.5 (1M context) <noreply@anthropic.com>

ntucker deployed to docs-preview October 25, 2026 00:56 — with GitHub Actions

View deployment

ntucker deployed to docs-preview October 25, 2026 02:40 — with GitHub Actions

View deployment

ntucker deployed to docs-preview October 25, 2026 03:29 — with GitHub Actions

View deployment

ntucker deployed to docs-preview October 25, 2026 04:57 — with GitHub Actions

View deployment

ntucker deployed to docs-preview October 25, 2026 05:18 — with GitHub Actions

View deployment

ntucker deployed to docs-preview October 26, 2026 15:29 — with GitHub Actions

View deployment

chore: Format dependOn

595ead9

ntucker deployed to docs-preview October 26, 2026 15:32 — with GitHub Actions

View deployment

Conversation

ntucker commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Motivation

Solution

Assumptions and validation

GOALS impact

Benchmarks

Uh oh!

changeset-bot Bot commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

🦋 Changeset detected

Uh oh!

ntucker commented Oct 24, 2026

Uh oh!

github-actions Bot commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

ntucker commented Oct 24, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codex Review Summary

Uh oh!

github-actions Bot commented Oct 24, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

chatgpt-codex-connector Bot left a comment

Choose a reason for hiding this comment

💡 Codex Review

Uh oh!

chatgpt-codex-connector Bot Oct 24, 2026

Choose a reason for hiding this comment

Uh oh!

github-actions Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Benchmark React

Uh oh!

github-actions Bot left a comment • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Benchmark

Uh oh!

chatgpt-codex-connector Bot commented Oct 25, 2026

Uh oh!

ntucker commented Oct 25, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 25, 2026

Uh oh!

ntucker commented Oct 27, 2026

Uh oh!

chatgpt-codex-connector Bot commented Oct 27, 2026

Uh oh!

This branch was successfully deployed

Reviewers

Uh oh!

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

ntucker commented Oct 24, 2026 •

edited

Loading

changeset-bot Bot commented Oct 24, 2026 •

edited

Loading

github-actions Bot commented Oct 24, 2026 •

edited

Loading

chatgpt-codex-connector Bot commented Oct 24, 2026 •

edited

Loading

github-actions Bot commented Oct 24, 2026 •

edited

Loading

github-actions Bot left a comment •

edited

Loading

github-actions Bot left a comment •

edited

Loading