refactor(chat): give a row one render identity

[chrisnojima]

Problem

Whether a row paints an author header — and therefore which recycling pool it joins — was computed twice per row per render, in two modules, from a mutable Map threaded through a context.

getMessageShowUsername had a pure signature but a hidden write: shownCache.set(ordinal, showUsername). Whichever caller ran first decided what the other saw.

• list-area/index.tsx reached past the thread-context interface to useConversationThreadStore().getState() to call it for getItemType.
• wrapper/wrapper.tsx called it again inside a shallow-compared selector — different subscription semantics over the same mutating cache.
• A third caller, separator.tsx, also wrote to the shared Map.

The code documented the hazard itself:

Must be the same sticky cache the rows render with (wrapper.tsx): without it, a row that keeps its sticky header after a scroll-back load would be typed headerless here, mixing tall headered rows into the headerless pool and poisoning that pool's height average.

Change

messages/row-identity.tsx owns the derivation:

useRowIdentity(ordinal)  // → {poolKey, reserveHeader, showUsername}
useRowPoolKey()          // for getItemType
getRowIdentity(store, snapshot, ordinal)  // non-hook

All entry points go through one derivation memoized per store snapshot in a WeakMap, so the callers cannot be told different things. getMessageShowUsername keeps its pure signature and loses the write — it returns {provisional, showUsername}, and the sticky record lives module-private, keyed per conversation and reset on the window generation rather than by a hand-placed clear().

ShownUsernameCacheContext is gone from thread-context. list-area no longer imports useConversationThreadStore, useCurrentUserState or getMessageRowType.

Validation

lint:all clean — 0 bailed out, 0 whole-props deps, tsc clean both projects.
jest --runInBand — 231 suites / 2262 tests.

row-metadata.test.ts keeps the pure half. row-identity.test.tsx covers the sticky half plus the previously-unreachable failure mode: the row and the list are told the same thing whichever asks first. All six mutation-checked.

[Copilot]

🟡 Changes recommended

Reactive username updates can again make row rendering and recycling-pool identities disagree, and the ordering regression test does not exercise its claimed scenario.

Once you've addressed the issues Copilot identified, you can request another Copilot review.

Pull request overview

Centralizes chat-row header and recycling-pool identity to keep rendering decisions consistent.

Changes:

• Adds snapshot-memoized row identity derivation.
• Removes the shared mutable username-header context.
• Adds/refactors row identity tests.

File summaries

File	Description
thread-context.tsx	Removes header-cache context.
wrapper/wrapper.tsx	Uses centralized row identity.
separator.tsx	Uses shared identity derivation.
row-metadata.tsx	Makes header metadata derivation pure.
row-metadata.test.ts	Tests provisional metadata behavior.
row-identity.tsx	Implements identity caching and hooks.
row-identity.test.tsx	Tests sticky headers and pool keys.
list-area/index.tsx	Uses centralized recycling-pool keys.

Review details

• Files reviewed: 8/8 changed files
• Comments generated: 4
• Review effort level: Balanced

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.