koffice/krita/image

[Boudewijn Rempt]

On Friday 02 October 2009, you wrote:

> SVN commit 1030020 by rempt:
>
> fix convolution based filters and masks
>
>  M  +6 -1      kis_mask.cc
>
>
> --- trunk/koffice/krita/image/kis_mask.cc #1030019:1030020
> @@ -111,8 +111,13 @@
>         KisPaintDeviceSP cacheDevice =
>             new KisPaintDevice(projection->colorSpace());
>
> +        // some filters only write out selected or affected pixels to dst,
> so copy
> +        KisPainter p1(cacheDevice);
> +
>  p1.setCompositeOp(cacheDevice->colorSpace()->compositeOp(COMPOSITE_COPY));
> +        p1.bitBlt(rc.topLeft(), projection, rc);
> +        p1.end();
> +
>         KisPainter gc(projection);
> -        //gc.beginTransaction("");
>         QRect updatedRect = decorateRect(projection, cacheDevice, rc);
>
>         /**
>

I'll cc the the mailing list, since this is a broader issue.

> Could you explain me how does this change help blur filter?

Not really, maybe Cyrille can -- it's something to do with the convolution 
painter.

> The point is, we can't do a bitBlt() here as needRect of the filter might
>  be much bigger than actual rc, given to this function. Actual needRect is
>  calculated in KisLayer::updateProjection and is know only there.

Maybe it's better to have it in kis_filter_mask instead? For now it's 
unfortunately needed, and I think a refactor of this thing should wait until 
2.2.

> Why does a filter needs a destination device to be prepared? Doesn't it
>  mean that the filter reads source information from it?

Not necessarily, the reason why the dst device needs to be a copy of the src 
device is that not all filters copy all pixels, even the ones they don't 
touch, to the dst device.

-- 
Boudewijn Rempt | http://www.valdyas.org