Repository navigation
Flaky on Windows arm64: review prompt shim tests time out after 10 seconds #367
Description
What happened
On main (commit 706d63b), a rerun of "Native desktop - Windows arm64" in CI run 43523947929 failed in the step "Run review prompt argument tests (Windows)".
Four tests in apps/server/tests/windows_terminal_shims.rs failed, all at line 172:
review_prompt_shim_passes_ampersandreview_prompt_shim_passes_backslash_before_quotereview_prompt_shim_passes_caretreview_prompt_shim_passes_bang
Each one reported called Result::unwrap() on an Err value: Elapsed(()). That is the 10-second tokio::time::timeout that waits for the marker in the terminal output. Result line: 6 passed; 4 failed.
Why this looks flaky
The same commit passed this step in the first attempt of the same run, and the same code passed it on the pull request run 43496452314. The failing attempt had a full Rust cache hit.
Measured so far: 1 failure in 3 runs of this step on Windows arm64.
Not known
- Whether the terminal never printed the marker or printed it late. The test unwraps the timeout before it can print the collected output.
- Whether Windows x64 is affected.
Suggested first step
Make the timeout failure print the collected output and the exit state, so the next failure shows which case it is.
Activity
test(terminal): start the shim timeout window only once the shim prints
mubeda commentedon Nov 14, 2026
Status after pull request 370 (2026-11-14): not reproduced, so this stays open.
ci: widen the temporary #367 reproduction to 80 cold runs of the revi…
test(terminal): give Windows shim tests 60 seconds for a stalled firs…
ci: run the temporary #367 reproduction 240 times against the 60-seco…
test(terminal): give Windows shim tests 60 seconds for a stalled firs…
docs(plan): record the #367 reproduction and its verification
mubeda commentedon Nov 14, 2026
Reproduced and fixed in pull request 370 (2026-11-14).
output="\u{1b}[?9001h\u{1b}[?1004h", exit=None: the pseudoconsole was up and the shim had neither printed nor exited. That run's batch took 19.8 s; the other 79 took 3.8 to 6.7 s and a warm rerun about 1 s. So the runner stalled process start-up for about 20 seconds.Not known: what stalls the runner. No stalled run occurred among the 240, so the fix was not seen turning a stalled run into a pass.