Re: [css3-images] Resolving on gradient issues

On Wed, Aug 3, 2011 at 4:32 PM, Alan Gresley <> wrote:
> On 3/08/2011 12:00 PM, Tab Atkins Jr. wrote:
>> 4. Switched the linear gradient keywords to upward, etc.
> I'm looking at 2nd of August edit of the ED. Indication of gradients
> following a path to and from a corners seem to missing. Is this being left
> out until CSS4-Images or is something like 'upward leftward' possible?

You cannot currently make a linear gradient go explicitly between
corners.  This is being reserved for Images 4, so I can give it the
time and attention it deserves without holding up the rest of the

> Since you would like to move the spec to LC and would like vender prefixes
> dropped from gradients, can the LC reflect the current implementation of
> WebKit and Gecko for gradient with alpha transparency to be done in
> non-premultiplied. The reason are.

Opera and IE both render gradient in premultiplied space as the spec
requires.  This is sufficient for the issue to not hold up CR.

(Simon Fraser anticipates that it will take a while before Webkit, or
at least Webkit on Mac, renders gradients in premultiplied space,
because it would require patching the functionality into Core
Graphics, the graphics library used by Webkit on Mac.  I don't know
how quickly it may appear in Webkit on systems that use Skia, or on

> 1. The keyword 'transparent' is already used in the wild (it being
> understood as transparent-black by some authors) and I believe that a new
> keyword is needed for something that reflects 'transparent-color' for
> gradients using premultiplied space (Brian has indirectly mentioned this
> recently). All this requires is a removal of the non-normative part in green
> text mentioning "The definition and implications of "premultiplied" color
> spaces......."

No, it would require more than just removing that note.  It's
non-normative, so by definition removing it won't change conformance
requirements at all.

> 2. I believe a syntax change or possibly prgba() can be used as a trigger
> gradients to premultiplied space where there is alpha.

I don't believe we should select the color-space based on the syntax
of the colors used, because a color should be independent of the
syntax used for it.  Instead, I believe this should be done either by
defining SVG's color-interpolation property to apply more widely to
all CSS values, or by giving gradients a specialized switch to control
the color-space they transition in.  I'll address this in a later
level of Image Values.

(I also think it would be useful to transition in hsl space, for
example.  If we make hsl() the trigger for this, what happens if you
mix hsl() and prgba() colors?  Any answer is arbitrary.)


Received on Wednesday, 3 August 2011 23:51:31 UTC