Re: ole32:fix CoWaitForMultipleHandles cause RPC hang

[Changhui Liu]

Hi Sebastian,

I tested your current work-in-progress patch:
https://github.com/wine-compholio/wine-staging/blob/master/patches/ole32-CoW...

I have two puzzles about these code:
>   index = 0xdeadbeef;
>   PostMessageA(hWnd, WM_DDE_FIRST, 0, 0);
>   PostQuitMessage(44);
>   thread = CreateThread(NULL, 0, post_message_thread, hWnd, 0, &tid);
>   ok(thread != NULL, "CreateThread failed, error %u\n", GetLastError());
>   hr = CoWaitForMultipleHandles(0, 100, 2, handles, &index);
>   ok(hr == RPC_S_CALLPENDING, "expected RPC_S_CALLPENDING, got 0x%08x\n", hr);
>   ok(index == 0 || broken(index == 0xdeadbeef) /* Win 8 */, "expected index 0, got %u\n", index);
>   success = PeekMessageA(&msg, hWnd, WM_DDE_FIRST, WM_DDE_FIRST, PM_REMOVE);
>   ok(success, "PeekMessageA failed, error %u\n", GetLastError());

The first, why use the PeekMessage to check if received a WM_DDE_FIRST message?
Is there an real application do like that way?

In fact CoWaitForMultipleHandles has a message loop, the WM_DDE_FIRST message has been
dispatched to the window's WNDPROC. So the PeekMessageA certainly failed.

I think the correct test should be this: 
1, Define a custom WNDPROC function named cowait_test_wnd_proc .
2, Define a global int variable named g_count_of_wm_dde_first.
3, Set the value of g_count_of_wm_dde_first to zero before call CoWaitForMultipleHandles.
4, Increase the value of g_count_of_wm_dde_first by one, once received a WM_DDE_FIRST message in cowait_test_wnd_proc.
5, Check if the value of g_count_of_wm_dde_first is equal to 2 after CoWaitForMultipleHandles returned.

The second  is why use todo_wine ?
        todo_wine
        ok(!success, "PeekMessageA succeeded\n");
why not just write?
        ok(!success, "PeekMessageA succeeded\n");

Thank you.

------------------
Regards.

------------------ Original ------------------
From:  "Sebastian Lackner";
Date:  Fri, Dec 12, 2014 11:57 AM
To:  "Changhui Liu"; 
Cc:  "wine-devel"; 
Subject:  Re: ole32:fix CoWaitForMultipleHandles cause RPC hang

On 12.12.2014 04:36, Changhui Liu wrote:
> Hi, I modified my patch, it has passed test on 
> https://testbot.winehq.org/JobDetails.pl?Key=10745&log_202=1#k202
>
> My test aims to reproduce the bug.

Sorry, but this patch is still not sufficient, it doesn't cover all situations how WM_QUIT messages are handled. As you can easily test with my current work-in-progress patch
https://github.com/wine-compholio/wine-staging/blob/master/patches/ole32-CoW...
your suggested fix breaks other stuff (which was working fine before), and so is most likely not correct.

I am not even sure if the problem is really in ole32, it might also be possible that the real problem is in the user32 message handling or probably somewhere else - but the only way to find that out is to write additional tests. Just testing this specific use-case will not be enough to justify a change, which might break other apps. ;)

Regards,
Sebastian