This MR fixes this random failure: http://test.winehq.org/data/126363ea5f9056449e8bd22cc69b51bd2d7dd9aa/linux_n...
It happens mostly in d3d9, but if you try hard enough you can see it on d3d8 too.
There are more random failures in test_window_position that I have seen when running on an actual display but not in a dummy X server or Xephyr. One failure mode is that we lose focus in the initial create_device and minimize ourselves. This particular problem is specific to focus follows mouse. Another is that fvwm seems to place the window slightly offscreen, maybe shifted by the size of the window decoration. I am still looking into those issues.
Unrelated to test_windoow_position, there is at least one more random failure in test_wndproc (Test failed: Expected message 0x46 for window 0, but didn't receive it, i=0.), so expect more fvwm3 related MRs after this one...