Re: [QUARTZ] Some fixes

[Christian Costa]

Robert Shearman wrote:
> Christian Costa wrote:
> > Robert Shearman wrote:
> > > Christian Costa wrote:
> > > > Index: pin.c
> > > >
> > > > RCS file: /home/wine/wine/dlls/quartz/pin.c,v
> > > > retrieving revision 1.9
> > > > diff -u -r1.9 pin.c
> > > > --- pin.c    16 Dec 2004 14:25:15 -0000    1.9
> > > > +++ pin.c    24 Dec 2004 13:31:52 -0000
> > > > @@ -1134,6 +1134,8 @@
> > > >     REFERENCE_TIME rtCurrent;
> > > >     ALLOCATOR_PROPERTIES allocProps;
> > > > -
> > > >
> > > > CoInitialize(NULL);
> > > > -
> > > >    SetEvent(This->hEventStateChanged);
> > > >
> > > > hr = IMemAllocator_GetProperties(This->pAlloc, &allocProps);
> > >
> > > Good catch, but it would probably be better to use 
> > > CoInitializeEx(NULL, COINIT_MULTITHREADED). This seems to be what 
> > > Microsoft uses for the pull thread.
> >
> > I was wondering about that but since I'm not very confident with this 
> > apartment stuff...
> > I will try and see if it's still work.
>
> The problem with using COINIT_APARTMENTHREADED (which is what you 
> effectively are using when you use the non-Ex version of the function) 
> is that it depends on using window messages when marshaling. The pull 
> thread does not have a message queue and does not dispatch messages, 
> therefore marshaled objects could hang when called. It is an unlikely 
> case (who would want to use marshaled objects in a performance 
> sensitive component), but you never know with these crazy Windows 
> applications.
>
> Rob

Thanks for the explanation, Rob.
I've just try and it still works fine with the change.

Bye,
Christian