Hi,
The 'check-gui' autopkgtest for slic3r-prusa intermittently fails on
ci.debian.net on riscv64:
https://ci.debian.net/packages/s/slic3r-prusa/testing/riscv64/
Two things in those logs stood out to me:
There were debug messages about wxTreeCtrl, which seem to have occurred in
other instances versions, but they may not actually be a problem.
358s 10:06:12 PM: Debug: window wxTreeCtrl@0x2ac9816aa0 ("treeCtrl") lost focus
even though it didn't have it
358s 10:06:12 PM: Debug: window wxTreeCtrl@0x2ac9816aa0 ("treeCtrl") lost focus
even though it didn't have it
The other remarkable thing was the overall duration of the test:
328s Waiting for 30 seconds to ensure Prusa Slicer has not crashed
358s Dumping windows list
On amd64, the whole test suite needs significantly less time:
40s Waiting for 30 seconds to ensure Prusa Slicer has not crashed
70s Dumping screenshot
This test runs as root, uses su to drop privileges, starts a window
manager and the app, waits for an arbitrary 30 seconds and takes
screenshots. This seems sufficiently high-complexity that it is going to
be difficult to make it 100% reliable, especially on slow machines like
riscv64 where 30 seconds might not be enough.
Judging from the reported duration, this may very likely be the case.
I tried running the test in a qemu container, since I currently don't have any
RISC-V hardware at my disposal for testing (I have access to the cfarm, but
they're rather impractical to use from a packager perspective). The test
reported success, although it didn't terminate properly - but this may be a
problem with qemu or my particular test container.
Would you suggest we increase the test timeout, or simply mark the GUI test as
flaky?
I think it's at least worth a try to see if a 60s (or more) timeout would fix
the issue.
Your other suggestion seem worthwhile, but I doubt they will fix the issue.
Regards,
Gregor
