On Fri, 14 May 2021, Rémi Bernon wrote: [...]
FWIW I was confused that the testbot pattern page reports this failure only since 05/03.
Looking at the reports before that date shows that this failure did not happen then.
I guess maybe the way the tests are run have changed around that date.
It's possible something changed on fg-deb64*, cw-gtx560* and cw-rx460* since they have unattended upgrades. However the TestBot's debiant2 VM does not and I did not change anything in that VM around that date (I updated vkd3d on 2021-03-02 and added MwmDecor on 2021-05-11).
So the only explanation I can think of is that some other test changed on that date, causing this one to start failing [1].
I could reproduce the same failure when running the tests in winetest since Wine 6.7 at least (and it was probably the case even before), although it's not marked as failing there:
I had the same issue with msxml3:domdoc. test.winehq.org showed it starting to fail on Linux on 2021-03-25 but it was still crashing even when I tried older builds. Eventually it turned out to be an old refcounting bug so probably something changed in the build toolchain that caused it to crash due to the memory corruption.
[1] When a test mysteriously starts failing on a TestBot Windows VM another possibility is that of a change in the build toolchain used for the official WineTest binary. But for Linux TestBot VMs the TestBot uses its own self-built binaries so that rules out changes in the build toolchain (unless the VM snapshot was changed in which case anything goes).