Pull request overview

Follow-up UI performance work focused on reducing per-render CPU cost during streaming by memoizing expensive parsing/formatting paths and improving tool-result lookup behavior in the chat message renderer.

Changes:

• Add bounded memoization caches for frequently re-run helpers (tool-call JSON parsing, partial JSON scanning, syntax highlighting, LaTeX preprocessing, search-results parsing, tool-result classification/line splitting).
• Replace per-tool-call linear search of tool result messages with an indexed lookup to avoid O(n^2) behavior across many tool calls.
• Add unit tests intended to prevent regressions in tool-call parsing and lookup performance.

Reviewed changes

Copilot reviewed 10 out of 10 changed files in this pull request and generated 3 comments.

	it('returns the same array reference for the same JSON string', () => {
		// parseToolCalls is not exported, but deriveAgenticSections uses it
		// internally. We verify memoization through behavior: calling
		// deriveAgenticSections twice with the same toolCalls should not
		// re-parse (which we verify by checking the returned sections

		// If the lookup were still O(n^2), this would be noticeably slow
		const start = Date.now();
		const sections = deriveAgenticSections(msg, toolMessages, [], false);
		const elapsed = Date.now() - start;

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.