Improve scroll performance for blocklists with many AIBlocks.

[vorporeal]

Description

Scrolling an AIBlock-heavy transcript caused large amounts of unrelated render work. A symbolicated Samply profile showed that TaskStore::lookup_exchange accounted for about 24% of renderer samples during the affected scroll. The lookup was repeated because two costs multiplied:

• Tooltip dismissal updated every AI block on each scroll, even when that block had no open tooltip.
• Each recording-related AI block rebuilt the conversation-wide recording span map during every render. That calculation traversed all exchanges twice.

This PR removes both multipliers:

• AI blocks and search result views now notify only when tooltip state changes. Scrolling still dismisses an open tooltip, but no-op dismissals do not dirty unrelated AI blocks.
• BlocklistAIActionModel now caches recording spans by conversation. Recording output changes, completed action results, restored results, and rewind cleanup invalidate the cache. Recording-related blocks share the cached map instead of rebuilding it independently.

Linked Issue

No linked issue. This change follows a profile-driven investigation of Warp-on-Web scrolling performance.

Testing

• I manually tested my changes locally with ./script/run.
  • Compared an AIBlock-heavy transcript before and after the change.
  • Confirmed that scrolling performance was substantially improved in the native build.
• cargo check -p warp
• ./script/format --check
• cargo clippy -p warp --all-targets --tests -- -D warnings
• Warp-on-Web manual verification. ./script/wasm/run was stopped before verification completed.

No integration test was added. This change targets redundant invalidation and derived-state recomputation during rendering, and the affected scrolling behavior was verified manually with a representative transcript.

Agent Mode

• Warp Agent Mode - This PR was created via Warp's AI Agent Mode
• Agent conversation: https://staging.warp.dev/conversation/6c08df13-9e77-44a3-9fd7-d7dd25de7270

CHANGELOG-BUG-FIX: Improved scrolling performance in transcripts with many Agent Mode blocks.

Co-Authored-By: Warp agent@warp.dev

[vorporeal]

• Improve scroll performance for blocklists with many AIBlocks. #15280 👈 (View in Graphite)
• master

This stack of pull requests is managed by Graphite. Learn more about stacking.

[warp-for-oss]

@vorporeal

I'm starting a first review of this pull request.

You can view the conversation on Warp.

I completed the review and no human review was requested for this pull request.

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz

[warp-for-oss]

Overview

This PR reduces redundant blocklist render work by suppressing no-op tooltip notifications and caching recording span maps per conversation. I did not find a code-level correctness, security, or spec-drift issue in the changed lines.

Concerns

• The change is user-facing because it changes scrolling responsiveness for AIBlock-heavy transcripts. Per repository guidance, please include screenshots or a short screen recording demonstrating it working end to end.
• The PR specifically targets Warp-on-Web scroll performance, but the testing notes say Warp-on-Web manual verification was stopped before completion. Please verify the target surface before merging.

Verdict

Found: 0 critical, 2 important, 0 suggestions

Request changes

Comment /oz-review on this pull request to retrigger a review (up to 3 times on the same pull request).

Powered by Oz