[mtlcl410]

What this changes

image_pixels_follow_the_frame_while_a_corner_is_dragged (crates/ui-egui/tests/editing.rs, from #1677) fails intermittently on test (windows-latest) with:

the picture is drawn in the frame [[704.0 245.2] - [828.8 322.0]], not only at []

Two recent examples:

• M7: the paragraph editor draws on its page only (#3061) #3070, run 39047581578;
• chore: cargo-deny ignores RUSTSEC-2026-0370/0371/0372 (zvariant, zbus_names, zvariant_derive; via eframe's AT-SPI) #3071, which changes only deny.toml.

So it isn't the PRs.

Cause. The drag preview draws the picture from the page raster, which arrives asynchronously. The test drags four frames after opening the document and never waits for that raster. On a slow runner it isn't there yet, so the preview draws no picture at all, hence the empty list.

Fix (test only). Before selecting and dragging, wait until render_pending() is false (up to 2 s), as other tests in this file do. Then assert that the raster arrived. No product code changes.

How you verified it

• Reproduced the failure. I added a 300 ms sleep to every page render (a local change, not committed) to stand in for a slow runner. main's test then fails with the CI message (not only at []), and with this change it passes.
• Without the delay, it passes as before.
• cargo fmt --all -- --check, cargo clippy -p pdfcraft-ui-egui --tests -- -D warnings

🤖 Generated with Claude Code

https://claude.ai/code/session_01RUNQSPVXx3dDBabRZHAnjz

[skylartaylor]

Thanks for tracking this down and reproducing the timing issue. Sorry I missed your earlier PR before opening #3129. I compared both with current main, and #2812 has already landed a wait before each drag that checks for the visible page raster as well as completed rendering. That covers the extra settling in mine too, so I’m closing #3129 rather than adding redundant waits. I appreciate the work you put into this.

[FredApps]

Read the whole diff. It adds a bounded wait (up to 400 steps of 5 ms, one step per pass) for the page raster before the drag, then asserts it arrived, so a slow runner now fails with a clear message instead of an empty drag preview. That addresses the cause you describe and touches only the test. CI is green at this head.