Re: Testing for drawing fixes r30523

[Quentin Mathé]

Hi Doug,

Le 2 juin 2010 à 21:36, Doug Simons a écrit :

> This change has broken some things on Windows. Here is a screenshot  
> from our application showing the problem, with an image displayed in  
> a tableView (the table here contains only one row):
>
> <Untitled Image.png>
>
> When I roll back this change, it displays as it should:
>
> <Untitled Image 2.png>
>
> It's hard to tell from this particular image, but the image itself  
> is not flipped, just displaced. Hopefully that's enough for you to  
> tell what's going on. If you need additional information please let  
> me know.

I think the backend is trying to compensate/revert an extra flipping  
previously done in Gui.
The code where the problem is located is -[Win32GState  
compositeGState:fromRect:toPoint:op:fraction:] pasted below.

The implementation looks similar to the old Cairo implementation, so  
the best way is probably to rewrite it in a similar way. That means I  
have to set up Windows somewhere since it's hard to get the  
implementation right. I don't have time today, but tomorrow I'll try  
to get a working GNUstep install on Windows to work on it.

By the way, does the scrolling still work on Windows?

> if (viewIsFlipped && (self
>  != source))
>     {
>       destPoint.y -= sourceRect.size.height;
>     }

You can try to remove this part since the backend shouldn't need it  
now. But I I don't think it's enough to fix it.

>   destRect.origin = destPoint;
>   destRect.size = sourceRect.size;
>   [ctm boundingRectFor: destRect result: &destRect];

The transform should be applied to destPoint only here.

>   rectTo = GSWindowRectToMS(
> self
> , destRect);
>   x = rectTo.left;
>   y = rectTo.bottom - sourceRect.size.height;
>
>   {
>     NSRect newRect;
>
>     [source->ctm boundingRectFor: sourceRect result: &newRect];
>     rectFrom = GSWindowRectToMS(source, newRect);
>     y += (sourceRect.size.height - newRect.size.height);
> // adjust location for scaled source

The line before the last one looks dubious to me. To get it working  
properly in all cases (rotation, scaling, flipping etc.), I had to  
rewrite the Cairo backend to use a more complex check.

Thanks for feedback,
Quentin.