Skip to content

Stop two bridge tests depending on the runner's speed - #17

Merged
dhivadhiva merged 2 commits into
mainfrom
fix/stop-test-timing
Nov 16, 2026
Merged

dhivadhiva merged 2 commits into
mainfrom
fix/stop-test-timing

Conversation

@dhivadhiva

Copy link
Copy Markdown
Contributor

Two fixes found while releasing 1.7.0. Neither changes what the app
does for a user in any way they would notice.

  • The channel-opening deadline test required an answer within 280 ms of
    wall-clock time; a slow Windows runner took 853 ms and failed the
    v1.7.0 build. The test now uses a 3 s channel delay and a 2 s bound.
  • A command the bridge stopped was sometimes reported with the watcher
    signal's number (138) instead of the stop's (143), because some shells
    had already collected the command when the wrapper's wait was cut
    short. The wrapper now asks the shell once more for the status it
    kept, and counts a begun stop as the stop's TERM when it kept none.
    150 of 150 runs under dash and under bash now give 143.

Tests: analyze clean, full suite and the real-sshd tests pass, local CI
passes.

The test let the channel take 300 ms to open, gave the call 100 ms and
required the answer within 280 ms of wall-clock time. A slow Windows
runner took 853 ms and failed it. The channel now takes 3 s and the
bound is 2 s: still far below what a call that waited for the channel
would take, and far above a slow runner's overhead.
When the watcher's signal cut the wrapper's wait short, some shells had
already collected the command, so the wrapper found no process to wait
for again and reported the signal's own number (138) instead of the
stop's (143). Ask the shell once more for the status it kept, and when
it kept none and a stop had begun, report the stop's TERM.
@dhivadhiva
dhivadhiva merged commit 79de50e into main Nov 16, 2026
1 check passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant