You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
E2E: a Mermaid reading-line case intermittently fails its setup check on Linux #393
apps/desktop/e2e/specs/mermaid-diagrams.e2e.ts, "keeps a reader just above the end in place as the end and the bottom region grow", failed its setup check nearEnd once in about three full default-suite runs in the Linux container lab (Ubuntu 22.04, WebKitGTK, Xvfb). It passed alone and in the next full run.
Evidence
In the failing run the reader did not move: the gap to the end stayed 56 px and both growth lists were empty. The product behaviour under test held.
nearEnd is endGap <= clientHeight * 0.1. With a gap of 56 it can only be false when the timeline's scroller is under 560 px tall; in a passing run it is 643 px.
The window was exactly the expected size, so something inside the app took at least 83 px above or below the timeline: state left by an earlier spec in the shared session, a transient banner, or a layout difference. None was shown, because the failing run's artifacts were overwritten by the next run.
The case's own setDesktopUiWindowSize(1400, 900) has no effect: openFixtureWorkspace resizes to 1000x720 afterwards. This holds for every Mermaid case and did not cause the failure.
What would decide it
Record the scroller's clientHeight, its bounding rectangle and the heights of the chat column's children in the case's timing file and in the labelled expectation, and keep each full run's artifact folder.
Closing as not planned: cause not found. The product kept the reader in place; only the test's premise about the timeline's height failed, once, and the evidence of why was lost. Changing the test to fit the height would remove the check, so nothing was changed. Reopen with the measurements named in the body.
Activity
mubeda commentedon Nov 19, 2026
Closing as not planned: cause not found. The product kept the reader in place; only the test's premise about the timeline's height failed, once, and the evidence of why was lost. Changing the test to fit the height would remove the check, so nothing was changed. Reopen with the measurements named in the body.