feat(desktop): state diagnostics — render + store churn counters

[OutThisLife]

Instrumentation for diagnosing renderer state churn, built ahead of the streaming/multi-tab performance work. It turns "I think the sidebar re-renders while an agent is typing" into a number.

Two dev-only counters, plus a perf scenario that drives them:

window.__RENDER_COUNTS__   what re-rendered, attributed to props / state / parent
window.__ATOM_CHURN__      which store published it, and whether it mattered

The wasted column in each is the fix list — components that re-rendered with neither changed props nor changed hook state, and stores that published a value deep-equal to the last one. @nanostores/react bails out on reference equality only, so the latter re-renders every subscriber for nothing; that's the "preserve reference identity on no-ops" rule in apps/desktop/AGENTS.md, now measurable.

Both are inert until start(), so the idle cost is one no-op branch per commit and per notify.

First result

Against 5 concurrent streaming tabs (render-churn --spawn --tiles 5 --tokens 240), reproducible across runs:

sidebar_renders          6
sidebar_wasted           0
wasted_renders      10,432
total_renders       78,385
commits              1,566
wasted_notifies          0

The sidebar hypothesis is refuted — 6 renders across the whole run, all on genuine busy/needsInput edges, none wasted. The stableArray guards on $workingSessionIds / $attentionSessionIds (store/session-states.ts:236-259) are holding. $sessionStates notified 1,200 times across 10 listeners with zero wasted notifies, so the publishing side is honest too.

The real cost is in components re-rendering on a parent commit without their own inputs changing: TooltipContent 2,048 wasted of 2,304, StatusbarItemView 1,446 of 2,174, Block and Ct 880 of 880 each. That's the next PR, not this one.

Why bippy

React 19.2 removed injectProfilingHooks from react-dom (grep -c → 0), so the entire mark* profiling family is dead on this stack and onCommitFiberRoot is the only channel left.

<Profiler> cannot answer the question: React invokes onRender for every Profiler in a committed tree including subtrees that bailed out, and a bailed-out subtree still reports nonzero actualDuration, so there's no safe threshold. Counting those callbacks would have "proven" the sidebar hypothesis true. This uses didFiberRender instead.

react-scan ships the right idea in its react-scan/lite subpath, but lite's only imports are bippy and bippy/source, while the package pulls ~217 transitive deps and floats react-grab/react-doctor on latest (non-reproducible installs; its main entry currently breaks Vite — upstream #448, #467). Taking bippy directly: MIT, zero dependencies.

Import order

main.tsx imports the entry statically, above react-dom. react-dom captures the devtools hook at module init, not at createRoot:

order	_renderers	commits
bippy first	1	1
react-dom first	0	0

A dynamic import behind import.meta.env.DEV resolves a microtask too late. Production exclusion is therefore a build-time alias to a no-op module rather than tree-shaking, since a static side-effect import can't be eliminated. Verified against the built bundle: bippy, __RENDER_COUNTS__, __ATOM_CHURN__, traverseRenderedFibers, didFiberRender all → 0 occurrences.

[github-actions]

૮ >ﻌ< ა ci review

_{ran on 37ac8ae}

ℹ️ Info

Desktop E2E visual evidence · View test artifacts · View job

1 visual diff.

_{inline evidence upload failed.}

Failed to upload diff-665a0833239e-onboarding-overlay-diff.png with gh image (exit code 1): Error uploading /home/runner/work/_temp/e2e-evidence/diff-665a0833239e-onboarding-overlay-diff.png: step 0 (get upload token): uploadToken not found on repo page — do you have write access to NousResearch/hermes-agent? (or, if NousResearch enforces SAML SSO, authorize at https://github.com/orgs/NousResearch/sso)