[PATCH 2/2] quartz/tests: Test renderer's parent WM_DESTROY notification.

[Gabriel Ivăncescu]

On 09/01/2020 17:46, Gabriel Ivăncescu wrote:
> On 09/01/2020 17:38, Nikolay Sivov wrote:
>>
>>
>> On Thu, Jan 9, 2020 at 6:23 PM Gabriel Ivăncescu 
>> <gabrielopcode at gmail.com <mailto:gabrielopcode at gmail.com>> wrote:
>>
>>     On 09/01/2020 16:46, Zebediah Figura wrote:
>>      > On 1/9/20 7:13 AM, Gabriel Ivăncescu wrote:
>>      >> Signed-off-by: Gabriel Ivăncescu <gabrielopcode at gmail.com
>>     <mailto:gabrielopcode at gmail.com>>
>>      >> ---
>>      >>   dlls/quartz/tests/filtergraph.c | 35
>>     ++++++++++++++++++++++++++++++++-
>>      >>   1 file changed, 34 insertions(+), 1 deletion(-)
>>      >>
>>      >> diff --git a/dlls/quartz/tests/filtergraph.c
>>      >> b/dlls/quartz/tests/filtergraph.c
>>      >> index ca45031..726c54a 100644
>>      >> --- a/dlls/quartz/tests/filtergraph.c
>>      >> +++ b/dlls/quartz/tests/filtergraph.c
>>      >> @@ -4232,16 +4232,38 @@ static HWND 
>> get_renderer_hwnd(IFilterGraph2
>>      >> *graph)
>>      >>       return hwnd;
>>      >>   }
>>      >> +static LRESULT CALLBACK parent_wndproc(HWND hwnd, UINT msg, 
>> WPARAM
>>      >> wParam, LPARAM lParam)
>>      >> +{
>>      >> +    switch (msg)
>>      >> +    {
>>      >> +        case WM_PARENTNOTIFY:
>>      >> +            if (LOWORD(wParam) == WM_DESTROY)
>>      >> +                ok(0, "Received WM_PARENTNOTIFY with
>>     WM_DESTROY.\n");
>>      >> +        break;
>>      >> +    }
>>      >> +    return DefWindowProcA(hwnd, msg, wParam, lParam);
>>      >> +}
>>      >> +
>>      >>   static void test_window_threading(void)
>>      >>   {
>>      >>       WCHAR *filename = load_resource(avifile);
>>      >>       IFilterGraph2 *graph = create_graph();
>>      >> +    WNDCLASSA cls = { 0 };
>>      >> +    HWND hwnd, parent;
>>      >>       HRESULT hr;
>>      >>       DWORD tid;
>>      >>       ULONG ref;
>>      >> -    HWND hwnd;
>>      >>       BOOL ret;
>>      >> +    cls.lpfnWndProc   = parent_wndproc;
>>      >> +    cls.hInstance     = GetModuleHandleA(NULL);
>>      >> +    cls.hCursor       = LoadCursorA(0, (const char*)IDC_ARROW);
>>      >> +    cls.lpszClassName = "TestParent";
>>      >> +    RegisterClassA(&cls);
>>      >> +
>>      >> +    parent = CreateWindowExA(0, "TestParent", NULL,
>>      >> WS_OVERLAPPEDWINDOW, 50, 50, 150, 150, NULL, NULL,
>>     cls.hInstance, NULL);
>>      >> +    ok(parent != NULL, "Failed to create parent window.\n");
>>      >> +
>>      >>       hr = IFilterGraph2_RenderFile(graph, filename, NULL);
>>      >>       if (FAILED(hr))
>>      >>       {
>>      >> @@ -4256,6 +4278,10 @@ static void test_window_threading(void)
>>      >>       {
>>      >>           tid = GetWindowThreadProcessId(hwnd, NULL);
>>      >>           ok(tid != GetCurrentThreadId(), "Window should have 
>> been
>>      >> created on a separate thread.\n");
>>      >> +        ok(!(GetWindowLongW(hwnd, GWL_EXSTYLE) &
>>      >> WS_EX_NOPARENTNOTIFY), "Window has WS_EX_NOPARENTNOTIFY.\n");
>>      >> +
>>      >> +        SetParent(hwnd, parent);
>>      >> +        SetWindowLongW(hwnd, GWL_STYLE, GetWindowLongW(hwnd,
>>      >> GWL_STYLE) | WS_CHILD);
>>      >
>>      > Does the application actually do this, or does it use
>>      > IVideoWindow::put_Owner()? If it is the former, I'd like to see a
>>      > comment to that effect.
>>      >
>>
>>     I actually don't know what the app does when it sets the parent. I
>>     wrote
>>     the test to see if Windows notifies the parent, and it seems that it
>>     doesn't (I also set the WS_CHILD just in case), despite not having 
>> the
>>     WS_EX_NOPARENTNOTIFY style in the first place. Wine would fail this
>>     test
>>     without the first patch, even without put_Owner.
>>
>>     I'll investigate what the app does. However, while this patchset 
>> fixes
>>     that app, I thought a more generic solution would be preferable, 
>> since
>>     that's how Windows seems to work (unless you have a better idea).
>>
>>
>> I don't understand how this window is set, just from this test alone. 
>> But, is it possible to subclass application window, and simply not 
>> forward certain messages?
>>
>> You can probably even test if it was subclassed when message returns.
>>
>>
> 
> You mean subclass the parent? I'll look into that, but it seems a bit 
> overkill. (that said, if Windows does it...)
> 
> FWIW, the issue is that DestroyWindow notifies the parent if the child 
> doesn't have that style.

Scratch that, the parent doesn't even process messages, that's why it 
deadlocks in the first place. Subclass wouldn't really help at all...