🤖 perf: skip composer height measurement for empty drafts

[ammar-agent]

Summary

Eliminates the forced-reflow storm that made chat/workspace switching slow: when the composer draft is empty (the common case), useAutoResizeTextarea and focusMessageInput no longer measure scrollHeight inside the commit phase / pre-paint rAF, letting CSS size the empty textarea instead.

Background

A CDP CPU profile of a live session (30s, repeated chat switches, ~49k-node transcript) showed switches blocking the renderer main thread for 0.6–3.4s each. The hottest application frame by far (2.18s self time, 14% of busy time; up to 991ms inside a single switch) was useAutoResizeTextarea's layout effect, called from commitLayoutEffects during the sync render triggered by the switch click.

The cost is not JS: the effect writes height = "auto" then reads scrollHeight, which forces a synchronous layout of the entire freshly-mounted (fully dirty) document before first paint. focusMessageInput in ChatInput repeats the same dance in a pre-paint rAF on every switch.

Implementation

• useAutoResizeTextarea: fast path for value === "" — clear any stale inline height (matching the existing send-path convention of style.height = "" + "let CSS min-height take over"), record state, and skip measurement entirely. No scrollHeight read, no reflow. The first keystroke afterwards takes the existing grow-only path (measures once on a clean tree, no auto reset).
• focusMessageInput (ChatInput): skip the auto/scrollHeight resize dance when the composer is empty; it fires right after a workspace switch while the new transcript is still dirty.
• VimTextArea / AskUserQuestionToolCall textareas: rows={1} so the natural (un-measured) empty height stays one line — textareas default to rows=2. With the consumers' existing min-h-* clamps, the rendered empty height is unchanged (verified live: empty composer renders at exactly its CSS min-height, same as before).

Non-empty drafts keep the existing measure-on-mount behavior.

Validation

• Captured the profile against the running dev instance via CDP; per-switch blocking spans attributed 29–45% of their time to this exact reflow path.
• Verified via HMR on the same live instance: empty composer has no inline height, rows=1, and offsetHeight exactly equals its CSS min-height (pixel-identical to the previous measured state).
• Hook unit tests updated/extended: empty mount performs zero scrollHeight reads and zero style writes; clearing to empty resets the inline height; first keystroke after the unmeasured empty state grows without the auto reset.

Risks

Low-moderate, scoped to composer sizing. The empty state is now CSS-governed; if a consumer styled the empty composer expecting an inline px height (none found — both consumers have min-h clamps that already dominated), its empty height could shift by a couple px. Typing, draft-restore, deletion/shrink, and max-height capping paths are unchanged and covered by existing tests.

---

Generated with mux • Model: anthropic:claude-fable-5 • Thinking: high • Cost: $2.11

[ammar-agent]

@codex review

[chatgpt-codex-connector]

Codex Review: Didn't find any major issues. You're on a roll.

Reviewed commit: 3626f5cf4d

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".

[ammar-agent]

Chromatic note for baseline acceptance (UI Tests: 1 change)

I rebuilt the Storybook for main and this branch locally and DOM-diffed textarea geometry across all 110 affected stories:

• 109 stories: inline height 62px → none, rendered height identical (CSS min-height clamp dominated before and after) — pixel-invisible.
• 1 real visual change: app-phoneviewports--i-phone-16-e-sidebar-with-sections. On main, the empty draft composer was stuck at 360px tall (exactly 50vh) — a bogus pre-layout scrollHeight measurement persisted as inline height, pushing the model selector/MCP section below the fold. On this branch it renders at its intended 112px min-h-28 clamp.

The flagged change is a strict improvement (fixes a pre-existing empty-composer inflation bug), safe to accept as baseline.