First change is to avoid creating and mapping windows concurrently, this is completely racy and can cause random results later on depending on which window ends up being focused and depending on the X11 events. In addition, FVWM focus tracking suffers from various race conditions already, which only makes it worse.
Second change is to avoid failing the tests because of spurious window resize messages on the dummy window. I'm not sure what the window is for exactly but I believe it's not meaningful in the message sequence as it never appears there. As it is maximized, the window is sometimes resized by some WMs when the display mode changes, which cause the spurious WM_SIZE messages.
-- v3: d3d8/tests: Use static class for the dummy window. d3d8/tests: Avoid creating visible windows concurrently. d3d9/tests: Use static class for the dummy window. d3d9/tests: Avoid creating visible windows concurrently.