perf: widen diff overscan during rapid scrolling

[benvinegar]

Summary

• widen the mounted diff/file window for a short idle period after bursty scroll jumps
• share the rapid-scroll overscan heuristic in a small tested helper
• prefetch highlights using the expanded rapid-scroll halo

Validation

• bun test src/ui/lib/adaptiveScrollOverscan.test.ts src/ui/components/ui-components.test.tsx src/ui/diff/rowWindowing.test.ts
• bun test ./test/pty/scroll.test.ts
• bun run format:check
• bun run typecheck
• bun run lint
• bun run bench:large-stream

This PR description was generated by Pi using OpenAI GPT-5

[greptile-apps]

Greptile Summary

This PR adds an adaptive overscan halo for rapid scrolling in DiffPane — when large scroll jumps are detected, the mounted row window widens temporarily (up to 240 rows, lasting 160 ms of idle) so the terminal shows real rows instead of placeholder regions while React commits lag behind input events.

• A new adaptiveScrollOverscan.ts module extracts the heuristic (deltaRows >= 4 threshold, 3-viewport minimum expansion, 240-row cap) with accompanying unit tests, and the constant RAPID_SCROLL_OVERSCAN_IDLE_MS is re-used to prefetch highlights using the same expanded window.
• DiffPane integrates rapidScrollOverscanRows into three windowing/prefetch sites (visibleViewportFileIds, the mounted diff window, and highlight prefetch), with a useCallback-stable activateRapidScrollOverscan and proper timeout cleanup on unmount.

Confidence Score: 4/5

Safe to merge. The changes are well-scoped and self-contained; the worst-case outcome of any issue is a brief 160ms over-rendering at remount, which is invisible to users.

The core logic is straightforward and the timeout lifecycle is handled correctly. The one behavioral gap — prevScrollTopRef starting at 0 so a remount with a non-zero scroll position briefly fires an unneeded overscan burst — is transient and harmless. The threshold boundary at deltaRows === 4 is also untested, leaving a minor gap in confidence that the strict < 4 guard is intentional.

src/ui/components/panes/DiffPane.tsx around the initial readViewport() call; src/ui/lib/adaptiveScrollOverscan.test.ts is missing a boundary-value case.

Important Files Changed

Filename	Overview
src/ui/lib/adaptiveScrollOverscan.ts	New helper that computes a temporary overscan row count after a large scroll jump. Logic is clean, well-bounded (max 240 rows), and correctly handles negative deltas via Math.abs.
src/ui/lib/adaptiveScrollOverscan.test.ts	Tests cover slow movement, bursty mid-range, and capped large jumps. The exact threshold boundary (deltaRows === 4) is not tested, leaving the inclusive/exclusive edge implicit.
src/ui/components/panes/DiffPane.tsx	Integrates rapidScrollOverscanRows into three windowing/prefetch sites and adds timeout lifecycle cleanup. On initial mount the prevScrollTopRef baseline is 0, so a component remounting with a non-zero scroll position will briefly fire an unneeded overscan burst for 160ms.
CHANGELOG.md	Adds a user-facing note under Fixed for the rapid-scroll window expansion. Entry is clear and correctly categorized.

Flowchart

%%{init: {'theme': 'neutral'}}%%
flowchart TD
    A[Scroll event fires] --> B[readViewport via microtask]
    B --> C{nextTop changed?}
    C -- No --> G[Update scrollViewport state]
    C -- Yes --> D[computeRapidScrollOverscanRows\ndeltaRows and viewportHeight]
    D --> E{absDelta less than 4 rows?}
    E -- Yes --> F[return 0]
    E -- No --> H[min 240, max absDelta x2 vs viewportRows x3]
    F --> I[activateRapidScrollOverscan\nearly-exit: no change]
    H --> J[activateRapidScrollOverscan overscanRows]
    J --> K[setRapidScrollOverscanRows to max of current and new]
    K --> L[Reset idle timer 160ms]
    L --> M[rapidScrollOverscanRows drives windowing]
    M --> N1[visibleViewportFileIds: max 8 or rows]
    M --> N2[Mounted diff window: max 24 or height x2 or rows]
    M --> N3[Highlight prefetch: max 24 or height x3 or rows]
    L --> O[On idle timeout: reset rows to 0]
    G --> P[Render]
    I --> G

Loading

Prompt To Fix All With AI

Fix the following 2 code review issues. Work through them one at a time, proposing concise fixes.

---

### Issue 1 of 2
src/ui/components/panes/DiffPane.tsx:504-513
**False overscan burst on remount with non-zero scroll position**

`prevScrollTopRef` is always initialized to `0`. When `readViewport()` is called immediately on effect setup (line 550), any non-zero `scrollBox.scrollTop` — which can happen if `DiffPane` unmounts and remounts while the underlying scroll box retains its position — will compute `deltaRows = nextTop - 0`. A value ≥ 4 rows triggers `activateRapidScrollOverscan`, adding a 160 ms overscan burst even though no real rapid scrolling occurred. The impact is minor (a brief over-render and then reset), but could be avoided by initializing `prevScrollTopRef` lazily from `scrollBox.scrollTop` before the first scroll-delta check.

### Issue 2 of 2
src/ui/lib/adaptiveScrollOverscan.test.ts:5-8
The exact threshold boundary `deltaRows === RAPID_SCROLL_MIN_DELTA_ROWS` (4) is untested. Since the guard is `< 4` (strictly less than), `deltaRows: 4` should trigger the calculation — but there's no assertion confirming that. Adding a case at `4` would pin the inclusive boundary explicitly.

```suggestion
  test("leaves slow row-by-row movement on the default window", () => {
    expect(computeRapidScrollOverscanRows({ deltaRows: 1, viewportHeight: 30 })).toBe(0);
    expect(computeRapidScrollOverscanRows({ deltaRows: -3, viewportHeight: 30 })).toBe(0);
    // Exactly at the threshold (4) is the first value that should trigger expansion.
    expect(computeRapidScrollOverscanRows({ deltaRows: 4, viewportHeight: 30 })).toBe(90);
  });
```

_{Reviews (1): Last reviewed commit: "perf: widen diff overscan during rapid s..." | Re-trigger Greptile}

[greptile-apps]

False overscan burst on remount with non-zero scroll position

prevScrollTopRef is always initialized to 0. When readViewport() is called immediately on effect setup (line 550), any non-zero scrollBox.scrollTop — which can happen if DiffPane unmounts and remounts while the underlying scroll box retains its position — will compute deltaRows = nextTop - 0. A value ≥ 4 rows triggers activateRapidScrollOverscan, adding a 160 ms overscan burst even though no real rapid scrolling occurred. The impact is minor (a brief over-render and then reset), but could be avoided by initializing prevScrollTopRef lazily from scrollBox.scrollTop before the first scroll-delta check.

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/ui/components/panes/DiffPane.tsx
Line: 504-513

Comment:
**False overscan burst on remount with non-zero scroll position**

`prevScrollTopRef` is always initialized to `0`. When `readViewport()` is called immediately on effect setup (line 550), any non-zero `scrollBox.scrollTop` — which can happen if `DiffPane` unmounts and remounts while the underlying scroll box retains its position — will compute `deltaRows = nextTop - 0`. A value ≥ 4 rows triggers `activateRapidScrollOverscan`, adding a 160 ms overscan burst even though no real rapid scrolling occurred. The impact is minor (a brief over-render and then reset), but could be avoided by initializing `prevScrollTopRef` lazily from `scrollBox.scrollTop` before the first scroll-delta check.

How can I resolve this? If you propose a fix, please make it concise.

[benvinegar]

Fixed by making the first viewport read a baseline snapshot. A retained non-zero scrollTop on remount no longer counts as scroll input or activates rapid overscan.

This comment was generated by Pi using OpenAI GPT-5

[greptile-apps]

The exact threshold boundary deltaRows === RAPID_SCROLL_MIN_DELTA_ROWS (4) is untested. Since the guard is < 4 (strictly less than), deltaRows: 4 should trigger the calculation — but there's no assertion confirming that. Adding a case at 4 would pin the inclusive boundary explicitly.

Suggested change

	test("leaves slow row-by-row movement on the default window", () => {
	expect(computeRapidScrollOverscanRows({ deltaRows: 1, viewportHeight: 30 })).toBe(0);
	expect(computeRapidScrollOverscanRows({ deltaRows: -3, viewportHeight: 30 })).toBe(0);
	});
	test("leaves slow row-by-row movement on the default window", () => {
	expect(computeRapidScrollOverscanRows({ deltaRows: 1, viewportHeight: 30 })).toBe(0);
	expect(computeRapidScrollOverscanRows({ deltaRows: -3, viewportHeight: 30 })).toBe(0);
	// Exactly at the threshold (4) is the first value that should trigger expansion.
	expect(computeRapidScrollOverscanRows({ deltaRows: 4, viewportHeight: 30 })).toBe(90);
	});

Prompt To Fix With AI

This is a comment left during a code review.
Path: src/ui/lib/adaptiveScrollOverscan.test.ts
Line: 5-8

Comment:
The exact threshold boundary `deltaRows === RAPID_SCROLL_MIN_DELTA_ROWS` (4) is untested. Since the guard is `< 4` (strictly less than), `deltaRows: 4` should trigger the calculation — but there's no assertion confirming that. Adding a case at `4` would pin the inclusive boundary explicitly.

```suggestion
  test("leaves slow row-by-row movement on the default window", () => {
    expect(computeRapidScrollOverscanRows({ deltaRows: 1, viewportHeight: 30 })).toBe(0);
    expect(computeRapidScrollOverscanRows({ deltaRows: -3, viewportHeight: 30 })).toBe(0);
    // Exactly at the threshold (4) is the first value that should trigger expansion.
    expect(computeRapidScrollOverscanRows({ deltaRows: 4, viewportHeight: 30 })).toBe(90);
  });
```

How can I resolve this? If you propose a fix, please make it concise.

Note: If this suggestion doesn't match your team's coding style, reply to this and let me know. I'll remember it for next time!

[benvinegar]

Added a boundary test for deltaRows === 4 so the inclusive burst threshold is explicit.

This comment was generated by Pi using OpenAI GPT-5