Did something break with the selection?

[Cyrille Berger Skott]

On Friday 29 April 2011, Silvio Heinrich wrote:
> Thank you for this suggestion but I'm afraid that this is not the case :-(
> (I've could have fixed it easily if it where).
> I checked the code and I wasn't able to find anything.
> To make sure I didn't miss anything I enabled all the old composite ops 
> again and
> even did a reset of KoCompositeOpCopy2.h to the state before I modified 
> anything.
> But the problem was still present :/
> And Btw: Even now the old CompositeOp for the "Normal" blending mode is 
> used (KoCompositeOpOver.h), so this weird
> behavior of the build-up mode should only occur with the other blending 
> modes (and this isn't the case, too).
> 
> Do you have any other suggestion?
Hum I think I figured it out.

Since, the selection mask is handled by KoCompositeOpAlphaBase, and you did 
not changed that class. The only logical explanation is that something changed 
in a function that class call.

The mask is used on line 83 of KoCompositeOpAlphaBase:

srcAlpha = KoColorSpaceMaths<channels_type, quint8>::multiply(srcAlpha, *mask, 
U8_opacity);

I looked at the changes you made for a while, and could not figure out what was 
the changed behaviour </mylife ;p>. But you have fixed the behaviour of it, 
from:
        return (dst_compositetype(a)*b*c) / 
(KoColorSpaceMathsTraits<_Tdst>::unitValue * 
KoColorSpaceMathsTraits<_Tdst>::unitValue);

to:

        return (dst_compositetype(a)*b*c) / 
(dst_compositetype(KoColorSpaceMathsTraits<_Tdst>::unitValue) * 
KoColorSpaceMathsTraits<_Tdst>::unitValue);


And in the first case, we get an overflow, and if channels_type is 16 bit, we 
actually compute:

srcAlpha * mask * U8_opacity / 0xFF

which is what we want (since srcAlpha is 16bits, while both mask and 
U8_opacity are 8bits).
However, now, with the fixed behaviour, we get:
srcAlpha * mask * U8_opacity / 0xFFFF

Which is not what we want.

I think we need to change line 83 of KoCompositeOpAlphaBase to 
srcAlpha = KoColorSpaceMaths<quint8, channels_type>::multiply(*mask, srcAlpha, 
opacity);

And KoColorSpaceMaths::multiply(_T a, _Tdst b, _Tdst c) to:

inline static _Tdst multiply(_T a, _Tdst b, _Tdst c) {
    return (dst_compositetype(a)*b*c) / 
(dst_compositetype(KoColorSpaceMathsTraits<_Tdst>::unitValue) * 
KoColorSpaceMathsTraits<_T>::unitValue);
}

And it should fix it.

-- 
Cyrille Berger Skott