[PATCH 3/6] user32: Return WAIT_TIMEOUT in nulldrv_MsgWaitForMultipleObjectsEx.

[Rémi Bernon]

On 5/14/21 5:54 PM, Zebediah Figura (she/her) wrote:
> On 5/14/21 6:40 AM, Rémi Bernon wrote:
>> When there's nothing to wait for.
>>
>> Signed-off-by: Rémi Bernon <rbernon at codeweavers.com>
>> ---
>>   dlls/user32/driver.c | 1 +
>>   1 file changed, 1 insertion(+)
>>
>> diff --git a/dlls/user32/driver.c b/dlls/user32/driver.c
>> index 35e206f5e98..cb9dda6a692 100644
>> --- a/dlls/user32/driver.c
>> +++ b/dlls/user32/driver.c
>> @@ -305,6 +305,7 @@ static void CDECL nulldrv_GetDC( HDC hdc, HWND 
>> hwnd, HWND top_win, const RECT *w
>>   static DWORD CDECL nulldrv_MsgWaitForMultipleObjectsEx( DWORD count, 
>> const HANDLE *handles, DWORD timeout,
>>                                                           DWORD mask, 
>> DWORD flags )
>>   {
>> +    if (!count && !timeout) return WAIT_TIMEOUT;
>>       return WaitForMultipleObjectsEx( count, handles, flags & 
>> MWMO_WAITALL,
>>                                        timeout, flags & MWMO_ALERTABLE );
>>   }
>>
> 
> Why is this necessary? In fact, won't this give wrong results if there 
> is a queued message available?
> 

This is just the driver interface, winex11.drv returns the same thing 
when not in a X11 client thread, when there was no X11 event, or when 
events didn't get translated (which I think is not always very accurate).

Then only place where it would matter most is in wait_message, but it's 
usually called at least with the thread queue handle, so count >= 1 there.

Then it's otherwise called with count == 0 in PeekMessageW, only to try 
peeking again if the first peek failed and if the graphics driver 
indicated it received some events (if it didn't return WAIT_TIMEOUT).

This is making some test fail here with nulldrv, as it returns an error, 
and so we peek messages again although we should not.

It could maybe be made more robust and handle the error case in 
PeekMessageW, but I think it's better to make sure the graphics drivers 
behave similarly. It would be nice if this could be factored out but it 
seems difficult.

There's also another place where the return value is checked, which is 
in check_for_events, called when checking async key state, to decide 
whether to flush or not window surfaces but I don't think it matters.

For all other call sites, its return value isn't used.
-- 
Rémi Bernon <rbernon at codeweavers.com>