Fix stray characters in TUI transcript from wide-grapheme cursor shift by warp-dev-github-integration[bot] · Pull Request #13591 · warpdotdev/warp

warp-dev-github-integration · 2026-07-10T23:13:52Z

Description

Fixes stray/stale characters that appear in the TUI transcript and persist while scrolling a long transcript (reported with floating g glyphs that were not part of the agent output, e.g. rendered mid-word as inline g task).

Root cause. The transcript is not fully repainted every frame — TuiFrameRenderer::draw (crates/warpui_core/src/runtime/renderer.rs) only clears + repaints on the first frame or on a resize; scrolling within a fixed viewport is an incremental cell diff. The proximate defect is in the ratatui-crossterm backend: CrosstermBackend::draw suppresses the cursor MoveTo for any cell whose column is exactly one past the previously written cell, assuming every printed cell advances the cursor by a single column. A width-2 grapheme advances it by two. Buffer::diff normally skips a wide grapheme's trailing half, which masks the bug — except for emoji carrying a VS16 variation selector (U+FE0F, e.g. ⚠️ ⌨️ ✅ ▶️), where the diff deliberately emits a trailing-clear cell at wide_x + 1. That cell lands in the false-contiguity window, so the backend skips the needed MoveTo and prints it — and everything after it on the line — one column too far right. Because scrolling only redraws the per-frame diff, the misplaced glyph is never repaired and sticks around. This is why the bug is flaky and content-dependent: it only triggers when the transcript contains VS16 emoji, which are common in agent output.

Fix. Emit the per-cell diff in batches that never span a wide grapheme: after a width-2 grapheme, a fresh backend.draw() call resets the backend's cursor tracking so the following cell gets a correct MoveTo. The change is contained to the renderer, preserves the synchronized-update wrapper and ratatui's style handling, and leaves the width-1 contiguous-run optimization intact.

Linked Issue

Reported via the factory-client bug-triage Slack thread (linked below). No GitHub issue.

Testing

Added regression test runtime::renderer::tests::wide_grapheme_does_not_shift_following_cells, which replays the renderer's byte stream onto a screen buffer and asserts the glyph after a VS16 emoji stays on its intended column. Confirmed it fails before the fix (glyph lands on column 3, next glyph pushed off-screen — the reported symptom) and passes after.
cargo test -p warpui_core --features tui --lib runtime::renderer::tests — all green (8 passed), including existing wide-grapheme / diff-optimization tests.
cargo fmt -p warpui_core --check and cargo clippy -p warpui_core --features tui --all-targets -- -D warnings — clean.

Screenshots / Videos

Not included: warp_tui only supports interactive device-auth and cannot be launched non-interactively in the cloud runner, so the flaky live symptom cannot be driven here. The deterministic headless renderer test (which reproduces the exact column shift at the byte-stream level) is the automated proxy. Final manual sign-off: on a real terminal, a long transcript containing VS16 emoji, scrolled repeatedly, shows no stray/persistent glyphs.

Agent Mode

Warp Agent Mode - This PR was created via Warp's AI Agent Mode

CHANGELOG-BUG-FIX: Fixed stray characters that could appear and linger in the TUI transcript while scrolling.

Reported and triaged from Slack: factory-client thread.

Conversation: https://staging.warp.dev/conversation/844f00e2-a398-4a38-87f3-9607585375db
Run: https://oz.staging.warp.dev/runs/019f4e23-2a19-715d-ae9d-b0f1164be78e

This PR was generated with Oz.

oz-for-oss · 2026-07-10T23:14:04Z

@warp-dev-github-integration[bot]

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

oz-for-oss

Overview

This PR changes the TUI frame renderer to split ratatui/crossterm draw batches after wide graphemes and adds a headless regression test for the cursor-shift artifact.

Concerns

The renderer batching code does not compile because baseline.diff(buffer) returns a BufferDiff iterator in ratatui 0.30, but the new code calls .len(), indexes it, and slices it.
This is a user-facing rendering change, but the PR description explicitly says no screenshot or recording is attached. For this user-facing change, please include screenshots or a screen recording demonstrating it working end to end.

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

oz-for-oss · 2026-07-10T23:18:24Z

+        // stray glyphs behind that later partial diffs never repair. Ending the
+        // current `draw` batch after each wide grapheme resets the backend's
+        // cursor tracking, so the following cell gets a fresh, correct `MoveTo`.
+        let diff = baseline.diff(buffer);


⚠️ [IMPORTANT] baseline.diff(buffer) returns a BufferDiff iterator, so the later .len(), indexing, and slicing won't compile; collect it before batching.

Suggested change

let diff = baseline.diff(buffer);

let diff: Vec<_> = baseline.diff(buffer).collect();

Thanks — this one is actually a false positive. Buffer::diff returns a Vec<(u16, u16, &Cell)> (the lazy BufferDiff iterator is the separate diff_iter method: pub fn diff<'a>(&self, other: &'a Self) -> Vec<(u16, u16, &'a Cell)> in ratatui-core 0.1.1), so .len(), indexing, and slicing all compile — cargo test -p warpui_core --features tui and cargo clippy ... -D warnings are both green. The suggested .collect() would in fact not compile, since diff() already returns a Vec. I've added an explicit let diff: Vec<_> = baseline.diff(buffer); annotation (plus a comment) at the call site to make the type obvious and prevent this confusion.

warp-dev-github-integration · 2026-07-10T23:27:39Z

On the visual-proof point: warp_tui only supports interactive device-auth, so the live TUI can't be launched non-interactively in the cloud runner. This defect is also specific to VS16 emoji (U+FE0F) rendered through the diff-based partial redraw, so a live GUI screenshot would be terminal-font/width dependent and could actually mislead (a terminal that renders the emoji as width-1 would look "wrong" even with the fix). So instead, here is a deterministic, font-independent headless before/after render of the exact defect at the byte-stream level — the layer where the bug lives:

frame 1:  "ABCDE"
frame 2 (intended):   col: 0 1 2 3 4  =  [emoji spans 0..1]  g  Y  Z

BEFORE (buggy):   "⌨️ gY"   → g shifted to column 3, Z pushed off-screen
AFTER  (fixed):   "⌨️gYZ"   → g back on column 2, Y and Z correct

This is exactly what the new regression test wide_grapheme_does_not_shift_following_cells asserts: it drives the real TuiFrameRenderer, replays its emitted byte stream onto a screen buffer, and checks the glyph column. It fails before the fix (g at column 3) and passes after (g at column 2). Final human sign-off remains an eyeball on a real terminal with an emoji-heavy transcript, but the headless test is the deterministic automated proxy for a surface that can't be driven here.

warp-dev-github-integration · 2026-07-10T23:27:40Z

/oz-review

harryalbert · 2026-07-12T22:39:37Z

Fix stray characters in TUI transcript from wide-grapheme cursor shift #13591 👈 (View in Graphite)
master

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

The crossterm backend suppresses the cursor MoveTo for a cell one column past the previously written cell, assuming every printed cell advances the cursor by a single column. Width-2 graphemes advance it by two, so the trailing-clear cell the diff emits after a VS16 emoji (and everything after it on the line) was printed one column too far right, leaving stray glyphs that scrolling's partial redraws never repaired. Emit the per-cell diff in batches that never span a wide grapheme so a fresh draw call (and thus a correct MoveTo) starts after each wide grapheme. Adds a headless regression test. Co-Authored-By: Oz <oz-agent@warp.dev>

Address review confusion between Buffer::diff (returns a Vec) and diff_iter (returns the BufferDiff iterator): make the Vec type explicit at the call site. No behavior change. Co-Authored-By: Oz <oz-agent@warp.dev>

cla-bot Bot added the cla-signed label Jul 10, 2026

warp-dev-github-integration Bot requested a review from harryalbert July 10, 2026 23:14

oz-for-oss Bot reviewed Jul 10, 2026

View reviewed changes

oz-agent and others added 6 commits July 12, 2026 19:07

Annotate diff() as Vec at the batching call site

25b51a4

Address review confusion between Buffer::diff (returns a Vec) and diff_iter (returns the BufferDiff iterator): make the Vec type explicit at the call site. No behavior change. Co-Authored-By: Oz <oz-agent@warp.dev>

fix rendering order issue

8b562e8

fix char spacing errors

a636886

fix selection geometry

62c4ee3

fix grapheme regression

1212ace

harryalbert force-pushed the factory/tui-wide-grapheme-cursor-shift branch from b270417 to 1212ace Compare July 12, 2026 23:07

harryalbert approved these changes Jul 12, 2026

View reviewed changes

harryalbert enabled auto-merge (squash) July 12, 2026 23:12

harryalbert merged commit 995e3dd into master Jul 12, 2026
27 checks passed

harryalbert deleted the factory/tui-wide-grapheme-cursor-shift branch July 12, 2026 23:23

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Uh oh!

Fix stray characters in TUI transcript from wide-grapheme cursor shift#13591

Fix stray characters in TUI transcript from wide-grapheme cursor shift#13591
harryalbert merged 6 commits into
masterfrom
factory/tui-wide-grapheme-cursor-shift

warp-dev-github-integration Bot commented Jul 10, 2026 •

edited

Loading

Uh oh!

oz-for-oss Bot commented Jul 10, 2026 •

edited

Loading

Uh oh!

oz-for-oss Bot left a comment

Uh oh!

oz-for-oss Bot Jul 10, 2026

Uh oh!

warp-dev-github-integration Bot Jul 10, 2026

Uh oh!

warp-dev-github-integration Bot commented Jul 10, 2026

Uh oh!

warp-dev-github-integration Bot commented Jul 10, 2026

Uh oh!

harryalbert commented Jul 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

	let diff = baseline.diff(buffer);
	let diff: Vec<_> = baseline.diff(buffer).collect();

Uh oh!

Conversation

warp-dev-github-integration Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Description

Linked Issue

Testing

Screenshots / Videos

Agent Mode

Uh oh!

oz-for-oss Bot commented Jul 10, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

oz-for-oss Bot left a comment

Choose a reason for hiding this comment

Overview

Concerns

Verdict

Uh oh!

oz-for-oss Bot Jul 10, 2026

Choose a reason for hiding this comment

Uh oh!

warp-dev-github-integration Bot Jul 10, 2026

Choose a reason for hiding this comment

Uh oh!

warp-dev-github-integration Bot commented Jul 10, 2026

Uh oh!

warp-dev-github-integration Bot commented Jul 10, 2026

Uh oh!

harryalbert commented Jul 12, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

warp-dev-github-integration Bot commented Jul 10, 2026 •

edited

Loading

oz-for-oss Bot commented Jul 10, 2026 •

edited

Loading