Re: RGB channels are set to zero if alpha is 0.

Really? Majority of PNG files are premultiplied and they render visually
alright?!  Hard to believe...

On Mon, Aug 28, 2017 at 10:31 AM, Bill Spitzak <[email protected]> wrote:

> src/cairo-png.c has an unpremultiply_data function that is explicitly
> setting the output to zero if the alpha is zero, to avoid
> divide-by-zero. (It outputs a garbage value of (c/a)%256 if the alpha
> is non-zero but less than any of the color channels). Note that some
> work I have seen with trying to support unpremultiplied it is far more
> common to output ((!c || c>a) ? c : c/a).
>
> IMHO it is probably better for Cairo to write and read the channels
> unchanged. Despite the spec, the majority of png files actually are
> premultipled, since the majority of software just writes their
> channels unchanged to the png library, and almost all compositing
> graphics (including Cairo) more naturally works with premultiplied
> data. Nuke at one time tried to guess whether it was premultipled, but
> this had to be abandoned as the guess would screw up people using
> emission color like you and it quickly became clear that almost all
> images are premultiplied.
>
> Removing this conversion from Cairo would probably be a good idea. It
> would make Cairo match other software, it would remove surprises for
> uses such as you, it would be faster, and write+read of png data would
> be lossless. Use the fact that it is the "toy" api if you want an
> excuse for ignoring the png spec.
>
>
> On Sun, Aug 27, 2017 at 11:55 AM, Guillermo Rodriguez
> <[email protected]> wrote:
> >
> > El domingo, 27 de agosto de 2017, Bram Stolk <[email protected]>
> escribió:
> >>
> >> I must say that PNG then makes for a very poor choice of writing to by
> >> Cairo.
> >
> >
> > Note that PNG support in Cairo is explicitly described as a "toy API",
> > intended to be used for testing and quick demos, but not much more.
> >
> > Guillermo
> >
> >
> >
> > --
> > cairo mailing list
> > [email protected]
> > https://lists.cairographics.org/mailman/listinfo/cairo
> --
> cairo mailing list
> [email protected]
> https://lists.cairographics.org/mailman/listinfo/cairo
>



-- 
behdad
http://behdad.org/

-- 
cairo mailing list
[email protected]
https://lists.cairographics.org/mailman/listinfo/cairo