The msctf:inputprocessor test is not reliable

[Francois Gouget]

This test is not reliable on my Debian desktop.
http://test.winehq.org/data/3b40c03a8f0e3dcbb236d59125b115b2501086f8/linux_f...

inputprocessor.c:2117: Test failed: OnSetFocus expected current focus 0x12b070 got 0x12b0d0
inputprocessor.c:538: Test failed: Unexpected ThreadMgrEventSink_OnSetFocus sink
inputprocessor.c:549: Test failed: Sink reports wrong previous focus

If I run it in a loop it fails about once every twelve times:

i=0;while ../../../tools/runtest inputprocessor.c && [ $i -lt 100 ];do i=`expr $i + 1`; echo $i; done

Introducing a single Sleep(50) as shown below makes it totally 
reliable here.

ShowWindow(wnd3,SW_SHOWNORMAL);
     SetFocus(wnd3);
+    Sleep(50);
     processPendingMessages();

test_CurrentFocus = FOCUS_SAVE;
     test_PrevFocus = FOCUS_SAVE;
     test_OnSetFocus = SINK_SAVE;
     SetFocus(wnd1);
     processPendingMessages();
     sink_check_saved(&test_OnSetFocus,dm2,dm1,"OnSetFocus");
     /* ^^^ Line that fails! */

That seems pretty hacky though. I suspect this is all because 
processPendingMessages() will exit if there's no pending message right 
after we call SetFocus() without giving enough time for the asynchronous 
X event to arrive.

But I'm surprised that a single Sleep() fixes the issue, especially 
where it's placed. Also I don't know anything about msctf so I'm unsure 
how ThreadMgrEventSink_OnSetFocus() figures in all this.

So should I just submit a patch to add that Sleep()?

Or would it be better to change processPendingMessages() to look more 
like user32/tests/msg.c's flush_events() function although it looks like 
it's overkill for now?

[Alexandre Julliard]

Francois Gouget  writes:
> Or would it be better to change processPendingMessages() to look more 
> like user32/tests/msg.c's flush_events() function although it looks like 
> it's overkill for now?

That's preferable, yes.