Hi Dmitry
There is a test case in Wine (dlls/user/tests/win.c,test_SetMenu) which shows that your patch is wrong.
This test case also fails on Win9x, as you can see in the code of the test case:
ok(DestroyMenu(hMenu), "DestroyMenu error %ld\n", GetLastError()); ok(!IsMenu(hMenu), "menu handle should be not valid after DestroyMenu\n"); ret = GetMenu(parent); /* This test fails on Win9x */ if (!is_win9x) ok(ret == hMenu, "unexpected menu id %p\n", ret);
It doesn't make sense to me to check that the menu handle is invalid AND it is still assigned to the window. Windows 2000 only passes this test because it keeps using the menu. It destroys the menu when it's not needed anymore. WINE is not so clever (yet), therefore we have to do it the same way as Windows 9x.
I think we should modify this testcase. After all, the behavior of DestroyMenu in this situation is undocumented.
I've searched for other patches on this subject. I was very surprised to see that Andreas Mohr submitted a patch for the same problem about two years ago:
http://www.winehq.com/hypermail/wine-patches/2002/08/0124.html http://www.winehq.com/hypermail/wine-patches/2002/09/0039.html
He describes the problem very well. It seems that his patch was not checked in, or it has been modified since. My patch doesn't have the problem that he describes in the second post.
It was said many times already that the real fix for that bug is to move all user object management into wineserver and use the same handle allocation scheme as the one used for windows.
Sorry, I'm new to WINE development and don't know this discussion. Maybe you can provide a link?
However, I don't think that this will fix the problem. I'm sure that my patch won't break existing applications, and it's very important for many Delphi applications, because Delphi often calls DestroyMenu on active menus.
Regards
Michael