On Sat Aug 26 08:19:37 2023 +0000, Alexandre Julliard wrote:
Yeah. I don't know why we didn't do this right from the beginning. Or
apply any of the other accumulated knowledge from the TestBot, for that matter. We can certainly tweak things, but I see no evidence that the Testbot runs consistently have better results than the Gitlab ones. And honestly, if the tests can only succeed in a carefully tuned environment, that's not much better than when they only succeeded on my box. We need to accept that Wine runs in various environments with various upstream bugs, and deal with that. If upstream bugs have a high impact we need to work around them in the code; if we decide that the impact is low enough not to bother, then it should be OK to ignore the failures in the tests. There may be some unusual configurations that we decide not to support, and maybe fvwm focus is one of them, but that should be the last resort. We need to stop using upstream bugs as an excuse to not fix the tests.
For a radically different direction: the opposite of running the tests 'in a carefully tuned environment' would be to run them in either GNOME or KDE (see below). This is not to say that we should only fix bugs for these DEs, but prioritizing them could make sense. But then most Wine developers probably already use them so maybe we can consider GNOME and KDE to already be covered and dedicate the CIs to testing other environments.
Also to complement Alexandre's comment and something not necessarily related to this MR, fixing the tests can be different from fixing Wine (bad uses of flaky, adding workarounds in the tests). The goal is to fix Wine of course so that's something to keep in mind when fixing the tests.
Market share information is hard to come by but based on Debian's popularity contest [1] and GamingOnLinux stats [2] we have:
| | [1] | [2] | |----------|-----|-----| | GNOME | 41% | 30% | | XFce | 21% | 8% | | KDE | 17% | 38% | | Cinnamon | 9% | 8% |
1. https://popcon.debian.org/by_inst Stats based on the task-xxx-desktop packages. The fvwm1 + fvwm + fvwm3 packages would be around 1%. 2. https://www.gamingonlinux.com/users/statistics/