[cairo] win32 surface create/destroy

[Theo Veenker]

Vladimir Vukicevic wrote:
> On 4/18/09 1:07 AM, Theo Veenker wrote:
>>>
>>> Also CS_OWNDC is bad - see
>>> http://blogs.msdn.com/oldnewthing/archive/2006/06/01/612970.aspx
>>
>> Good to know. I'll just drop CS_OWNDC, as it just makes things worse.
> 
> For the CS_OWNDC case, the system is modifying that DC behind cairo's 
> back -- while a cairo surface is attached to the DC, it's generally bad 
> to use GDI on that same DC unless you return things back to the same 
> state that they were in before making a cairo call.  In particular, the 
> bounds on the DC are inferred by examining the clip region attached to 
> the DC at surface creation time; I believe this clip region will change 
> in BeginPaint as the system clips to the dirty areas, but cairo won't 
> know that and will still use the clip that was set when the surface was 
> created.

You're right. I found out I had to reset the clip region of the surface
on resize to make it work with windows that set CS_OWNDC.

> 
> This is also why you should create a new surface each time when 
> interacting with a platform-native drawing system -- unless you have 
> permanent control over something like a DC, you need to take care to not 
> have cairo and native drawing clash with eachother.

The overhead for create/destroy surface was about 20..50us on my system.
So not that terrible. I anyway tried to see if it would be possible to keep
the surface around (not create/destroy on each redraw). It required dropping
BeginPaint()/EndPaint() in MW_PAINT for the reasons you mention, redraw
the part indidated by GetUpdateRect() and call ValidateRect(). In the
end its no problem at all to keep the surface.

Theo