Repository navigation
Linux e2e flake with unreleased ansi/vterm: command_prompt_cancels_and_opens_a_window #8
Description
session::command_prompt_cancels_and_opens_a_window (tests/atrium/session.rs:464) fails on Linux when atrium builds against the unreleased nativelite-ansi / nativelite-vterm (the path deps on main). It does not fail against the published versions.
Evidence (WSL Ubuntu, 2026-09-30)
| build | ansi / vterm | result |
|---|---|---|
working tree, main @ b826f79 |
path deps (ansi 11 commits past v0.4.0, vterm 10 past v0.5.3) | 3 of 12 failed, test run alone |
| published atrium 0.38.0 | crates.io ansi 0.4.0 / vterm 0.5.3 | 36 of 36 passed; full suite green |
Windows is not affected: the gate passes there with the path deps.
What the failure shows
The keys do reach the new window. The screen has sub- on row 3 and a fresh $ below it. But the test waits for the raw passthrough bytes sub-\r, and atrium drew the line through a repaint instead (ESC[3;1Hsub-ESC[4;1H$). This is the same class of test assumption afda305 fixed: a byte match that assumes the output arrives unrendered.
Open question
Which unreleased change makes the repaint path win the race: scrollback / history, write_ascii fast paths, autowrap marks, checkpoints? Bisect across the ansi and vterm commits before either is released and atrium bumps to it. Then decide whether it is only the test (match on rendered screen content, as afda305 did) or a real change in when atrium repaints a new window.
Blocks: atrium's next ansi/vterm dependency bump.
Activity