[ntucker]

Requested by Nathaniel · project thread

Stacked on #4391.

Motivation

Before: a coding agent (or anyone in the console) could only see dispatched actions by wrapping Controller._dispatch after the page loaded, so the fetches made during page load were never visible unless the Redux DevTools extension was installed. Two stores on the default devtools name also replaced each other in __DC_CONTROLLERS__, so only one could be found, though the docs said multiple providers were fully supported.

After: in development, DevToolsManager keeps each store's recent actions from startup on globalThis.__DC_ACTIONS__, keyed like __DC_CONTROLLERS__, with or without the extension. A store whose name is taken registers as 'Data Client: My App (2)'.

__DC_ACTIONS__.get('Data Client: My App').slice(-10);

Solution

• The log is part of the dev-only middleware. It respects skipLogging like the extension does, keeps up to maxBufferLength actions (twice maxAge, 100 by default), and trims in place so the array on the global stays current. Production builds keep the pass-through middleware.
• init() picks the first free key (name, name (2), ...) and cleanup() removes that key from both maps. Where there is no window (server rendering), a same-named store replaces the previous one instead, because per-request stores are never cleaned up and unique keys would keep every request's store alive. The check is window, not document, so React Native (which has a window but no document) still gets unique keys.
• The debugging skill now uses the built-in log and returns actionLog: 'built-in'. On older versions it falls back to wrapping dispatch ('shim'), because skills aren't pinned to the user's package version. Both paths store raw actions, and the Step 3 snippets return only the useful fields.
• Docs: DevToolsManager "Programmatic store access", the coding-agents draft post, and two changesets.

Tests: packages/core/src/manager/__tests__/devtoolsManager.ts covers pre-init logging without the extension, the cap, skipLogging, duplicate names, re-init and cleanup; devtoolsManager.node.ts covers server rendering (no window: replace) and React Native (window only: unique keys). I also ran the skill's Step 1 shim and Step 3 snippets, extracted from the doc, against a real Controller with DevToolsManager and NetworkManager, in both built-in and fallback modes (temporary test, not committed).

Open questions

The Redux extension still shows two same-named stores under one name. Only the registry key is made unique, because the extension connection is opened in the constructor, before it's known which names are taken.

🤖 Generated with Claude Code

https://claude.ai/code/session_01U7fXTuhewgyp2SGAqY6LgJ

---

Generated by Claude Code

---

Note

Low Risk
Changes are dev-only middleware and documentation; production builds keep pass-through behavior. Registry key logic is covered by new unit tests for SSR, RN, polling, and duplicate names.

Overview
Dev-only programmatic debugging now exposes each store’s dispatch history on globalThis.__DC_ACTIONS__ (same keys as __DC_CONTROLLERS__), including page-load fetches, without requiring the Redux DevTools extension. The log is capped and summarized (newest 20 full payloads; polling keys limited via configurable maxPollingActions).

Multi-store registration no longer overwrites same default devtools names: browsers/React Native get suffixed keys ((2), (3)…); SSR without window replaces the prior entry to avoid leaking per-request stores.

Docs and agent skills add a symptom → “check first” table, wire the debug shim to the built-in log (actionLog: 'built-in') with a legacy dispatch-wrap fallback, and refresh Step 3 snippets (meta + action errors, safe serialization). DevToolsManager API docs and the coding-agents post document __DC_ACTIONS__.

^{Reviewed by Cursor Bugbot for commit af22339. Bugbot is set up for automated code reviews on this repo. Configure here.}

[changeset-bot]

🦋 Changeset detected

Latest commit: af22339

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

This PR includes changesets to release 7 packages

Name	Type
@data-client/core	Patch
@data-client/react	Patch
@data-client/vue	Patch
example-benchmark	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

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 98.37%. Comparing base (2a73c61) to head (af22339).
⚠️ Report is 7 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #4393   +/-   ##
=======================================
  Coverage   98.37%   98.37%           
=======================================
  Files         161      161           
  Lines        3072     3072           
  Branches      604      604           
=======================================
  Hits         3022     3022           
  Misses         10       10           
  Partials       40       40           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[chatgpt-codex-connector]

Codex Review Summary

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

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-17T19:25:26.902013Z	af22339	New commits

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.

[ntucker]

Staff engineer (Cursor agent): Re-review of 4cb6039..725d1f6 (summarize old actions, cap polling keys). One small change requested. Everything else in the delta looks right.

Requested change: the polling cap only holds on the first subscribe

In logAction, the scan removes the maxPollingActions-th newest entry for the key and then breaks. The comment says this works because "the log already holds at most maxPollingActions for this key". That's true on the first subscribe, since the SUBSCRIBE push trims the mount fetch/response itself. It stops being true once a polling key unsubscribes and later resubscribes (navigate away and back, or remount). While the key isn't polling, its entries go in uncapped: the UNSUBSCRIBE, the remount fetch and the remount setResponse. After it resubscribes, each push removes one entry and adds one, so the key's count never drops back to the cap. Older entries for that key stay until pushCapped halves the log, and each away/back cycle adds about 3 more. Memory is still bounded by maxBufferLength, but the code comment and the docs ("keeps only its last maxPollingActions actions") are wrong in that case.

Suggested fix (makes the code simpler and drops the invariant): don't break, and remove every match past the cap:

for (let i = log.length - 1, seen = 0; i >= 0; i--) {
  if ((log[i] as { key?: string }).key === key && ++seen >= this.maxPollingActions)
    log.splice(i, 1);
}

This scans the whole log (≤ maxBufferLength) once per action on a polling key, which costs nothing at poll intervals. Also please add a test for subscribe → polls → unsubscribe → fetch/response → subscribe → polls, checking the key ends at maxPollingActions.

What I validated, and how

• I ported logAction/summarize/pushCapped from 725d1f6 to a standalone Node script and ran it. With default settings, 5 cycles of mount (fetch, response, subscribe) → 10 polls → unsubscribe → 3 unrelated actions left key a at 5, 8, 11, 14, 17 entries. After resubscribing and 10 more polls it was still at 17, and they were mostly stale fetch/setresponse. With the break removed, the same run stays at 4 to 5 and ends at 4 (fetch, setresponse, fetch, setresponse). The first-subscribe case with maxPollingActions: 2 capped correctly (2 entries) in both versions.
• In the same port I checked summarization by object identity over 20,000 random actions (polling removals, subscribes/unsubscribes, pushCapped halving). Exactly the newest 20 entries were the original objects every time, all older ones were summaries, and the length never went over 100: 0 violations.
• I read gh pr checks at 725d1f6: all CircleCI jobs pass (typecheck, lint, unit tests incl. native, esmodule-types, browser build). Cursor Bugbot was still pending.

Only read in code, not run

• summarize drops the FETCH promise/resolve/reject and the payloads, and keeps error responses. Skill docs moving from meta.date to meta.fetchedAt matches what FETCH summaries keep.
• NetworkManager.skipLogging only skips duplicate FETCH, so SUBSCRIBE/UNSUBSCRIBE always reach the refcount. DevToolsManager is unshifted ahead of SubscriptionManager, so it sees them first.
• Not blocking: logAction, summarize and the polling field sit outside the NODE_ENV guard, so they ship wherever the class survives in prod, like init/handleAction already do. They're never called there.

[ntucker]

Agreed, and fixed in this PR. a569c84 drops the break, so every match past the cap is removed (the post-push review caught the same thing). It also renames the count map to pollingSubscribers and says "per key (endpoint and args)" in the docs. b3a44a2 adds the test you asked for: three subscribe → polls → unsubscribe → fetch/response cycles, then a resubscribe, and the key ends at maxPollingActions. There's also a test for entries logged before the first subscribe.

On the non-blocking note: leaving it as is. The class already ships init/handleAction the same way, and production never calls these.

---

Generated by Claude Code

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: b3a44a2f2b

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 74d740e3f0

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]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 2206128056

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]

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.

[github-actions]

Size Change: 0 B

Total Size: 112 kB

ℹ️ View Unchanged

Filename	Size	Change
examples/test-bundlesize/dist/App.js	1.45 kB	0 B
examples/test-bundlesize/dist/polyfill.js	301 B	0 B
examples/test-bundlesize/dist/rdcClient.js	10.8 kB	+3 B (+0.03%)
examples/test-bundlesize/dist/rdcEndpoint.js	8.04 kB	0 B
examples/test-bundlesize/dist/rdcNextjs.js	12.4 kB	+1 B (+0.01%)
examples/test-bundlesize/dist/rdcPipeableStream.js	9.65 kB	0 B
examples/test-bundlesize/dist/react.js	68.6 kB	0 B
examples/test-bundlesize/dist/webpack-runtime.js	698 B	0 B

_{compressed-size-action}

[github-actions]

Benchmark React

Details

Benchmark suite	Current: af22339	Previous: 7c55890	Ratio
data-client: getlist-100	149.29 ops/s (± 7.8%)	157.49 ops/s (± 8.3%)	1.05
data-client: getlist-500	47.2 ops/s (± 9.5%)	49.38 ops/s (± 7.4%)	1.05
data-client: update-entity	322.58 ops/s (± 6.9%)	357.14 ops/s (± 8.5%)	1.11
data-client: update-user	327.96 ops/s (± 9.4%)	357.14 ops/s (± 6.6%)	1.09
data-client: getlist-500-sorted	51.55 ops/s (± 8.6%)	53.63 ops/s (± 7.4%)	1.04
data-client: update-entity-sorted	303.03 ops/s (± 6.6%)	318.18 ops/s (± 8.9%)	1.05
data-client: update-entity-multi-view	303.03 ops/s (± 7.6%)	322.58 ops/s (± 6.4%)	1.06
data-client: list-detail-switch-10	9.66 ops/s (± 9.6%)	9.07 ops/s (± 9.3%)	0.94
data-client: update-user-10000	81.63 ops/s (± 9.1%)	76.05 ops/s (± 10.4%)	0.93
data-client: invalidate-and-resolve	37.81 ops/s (± 6.8%)	41.84 ops/s (± 5.3%)	1.11
data-client: unshift-item	227.27 ops/s (± 2.1%)	274.02 ops/s (± 1.7%)	1.21
data-client: delete-item	312.5 ops/s (± 8.6%)	298.57 ops/s (± 9.4%)	0.96
data-client: move-item	181.88 ops/s (± 6.6%)	185.19 ops/s (± 6.8%)	1.02

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

[github-actions]

Docs preview: https://docs-site-pr-4393-data-client.vercel.app

Built from af22339 merged into master (this build). English only: this PR doesn't touch translations.

[github-actions]

Benchmark

Details

Benchmark suite	Current: af22339	Previous: 7c55890	Ratio
normalizeLong	584 ops/sec (±2.81%)	514 ops/sec (±2.92%)	0.88
normalizeLong Values	526 ops/sec (±0.29%)	461 ops/sec (±0.36%)	0.88
normalizeLong Scalar	514 ops/sec (±2.11%)	451 ops/sec (±2.88%)	0.88
normalizeLong Scalar update	1054 ops/sec (±0.10%)	940 ops/sec (±0.85%)	0.89
denormalizeLong	364 ops/sec (±5.83%)	308 ops/sec (±6.55%)	0.85
denormalizeLong Values	326 ops/sec (±4.53%)	283 ops/sec (±4.69%)	0.87
denormalizeLong donotcache	1425 ops/sec (±0.78%)	1313 ops/sec (±0.97%)	0.92
denormalizeLong Values donotcache	980 ops/sec (±0.13%)	872 ops/sec (±1.07%)	0.89
denormalizeLong Scalar donotcache	1439 ops/sec (±0.09%)	1261 ops/sec (±0.86%)	0.88
denormalizeShort donotcache 500x	2020 ops/sec (±0.10%)	1754 ops/sec (±0.87%)	0.87
denormalizeShort 500x	916 ops/sec (±7.34%)	770 ops/sec (±6.56%)	0.84
denormalizeShort 500x withCache	9105 ops/sec (±0.11%)	8798 ops/sec (±0.42%)	0.97
queryShort 500x withCache	4840 ops/sec (±0.13%)	4590 ops/sec (±0.60%)	0.95
buildQueryKey All	78139 ops/sec (±1.01%)	72518 ops/sec (±0.57%)	0.93
query All withCache	7450 ops/sec (±3.96%)	7243 ops/sec (±6.06%)	0.97
denormalizeLong with mixin Entity	338 ops/sec (±6.52%)	280 ops/sec (±8.60%)	0.83
denormalizeLong withCache	7403 ops/sec (±0.18%)	7241 ops/sec (±0.37%)	0.98
denormalizeLong withCache (Scalar churn)	7368 ops/sec (±0.67%)	7142 ops/sec (±1.18%)	0.97
denormalizeLong Values withCache	6539 ops/sec (±1.10%)	6328 ops/sec (±1.87%)	0.97
denormalizeLong Scalar withCache	8982 ops/sec (±0.27%)	8343 ops/sec (±1.05%)	0.93
denormalizeLong Scalar update withCache	3617 ops/sec (±0.18%)	3509 ops/sec (±0.41%)	0.97
denormalizeLong All withCache	6120 ops/sec (±0.24%)	6458 ops/sec (±0.30%)	1.06
denormalizeLong Query-sorted withCache	7652 ops/sec (±4.73%)	7408 ops/sec (±5.81%)	0.97
denormalizeLongAndShort withEntityCacheOnly	2150 ops/sec (±0.21%)	1845 ops/sec (±0.64%)	0.86
denormalize bidirectional 50	7042 ops/sec (±9.81%)	6284 ops/sec (±11.56%)	0.89
denormalize bidirectional 50 donotcache	60240 ops/sec (±0.48%)	53624 ops/sec (±0.83%)	0.89
getResponse	5618 ops/sec (±2.68%)	4915 ops/sec (±5.74%)	0.87
getResponse (null)	15114528 ops/sec (±0.44%)	13533430 ops/sec (±1.34%)	0.90
getResponse (clear cache)	303 ops/sec (±7.48%)	259 ops/sec (±7.83%)	0.85
getSmallResponse	5218 ops/sec (±0.21%)	4735 ops/sec (±1.05%)	0.91
getSmallInferredResponse	4227 ops/sec (±4.82%)	4049 ops/sec (±0.96%)	0.96
getResponse Collection	4808 ops/sec (±2.75%)	5296 ops/sec (±4.78%)	1.10
get Collection	2718 ops/sec (±0.32%)	3986 ops/sec (±0.58%)	1.47
get Query-sorted	5791 ops/sec (±3.15%)	5414 ops/sec (±5.78%)	0.93
setLong	586 ops/sec (±0.20%)	519 ops/sec (±0.16%)	0.89
setLongWithMerge	325 ops/sec (±0.14%)	294 ops/sec (±1.13%)	0.90
setLongWithSimpleMerge	353 ops/sec (±0.07%)	310 ops/sec (±1.09%)	0.88
setSmallResponse 500x	1275 ops/sec (±1.00%)	1056 ops/sec (±1.11%)	0.83
setMany 50x one-per-row	285 ops/sec (±0.33%)	257 ops/sec (±1.01%)	0.90
setMany 50 batch	5417 ops/sec (±0.53%)	4599 ops/sec (±1.07%)	0.85
setMany 500x one-per-row	31.26 ops/sec (±0.42%)	29.39 ops/sec (±0.92%)	0.94
setMany 500 batch	1707 ops/sec (±0.10%)	1540 ops/sec (±0.91%)	0.90

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

[chatgpt-codex-connector]

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.

[chatgpt-codex-connector]

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.

[ntucker]

@codex review the changes since your last completed review at commit 2206128056, through the current head ad1d9e2e96b6a15d3a4abfe69fd551946e2e6240. Subsequent automatic review attempts were blocked by code review usage limits. Focus on the changes in that range and verify the fixes for previous findings, using the surrounding code as needed.

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. Breezy!

Reviewed commit: ad1d9e2e96

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".