W3C home > Mailing lists > Public > www-style@w3.org > July 2010

Re: [css3-background] Where we are with Blur value discussion

From: Aryeh Gregor <Simetrical+w3c@gmail.com>
Date: Fri, 16 Jul 2010 18:20:38 -0400
Message-ID: <AANLkTilcJlYEiwPSvxMZ_w5Oamrc_Xe4NywK_UC-YCyx@mail.gmail.com>
To: Brad Kemper <brad.kemper@gmail.com>
Cc: "Tab Atkins Jr." <jackalmage@gmail.com>, "L. David Baron" <dbaron@dbaron.org>, Simon Fraser <smfr@me.com>, Brendan Kenny <bckenny@gmail.com>, www-style list <www-style@w3.org>
On Thu, Jul 15, 2010 at 8:45 PM, Brad Kemper <brad.kemper@gmail.com> wrote:
> I mean that you seem to be arguing against the sort of language we have in the draft, because you want the blurs to be pixel perfect. Yet the part you are objecting to is the part that attempts to have measurable results that match the author's input length, and replace it with something where the only way to measure the result is to compare it to results in other standards (which are similarly vague in their wording).

I still don't get what you're saying.  Both canvas and SVG are
absolutely precise: a given input maps to a Gaussian blur with sigma
equal to a specific well-defined value.  I'm not suggesting CSS refer
to either canvas or SVG, I'm suggesting it use the same precise type
of definition that canvas and SVG already do: a blur of X pixels
should map to a Gaussian blur with a well-defined value of sigma.
This gives an extremely measurable result.

> I think the most important step towards having pixel perfect results is to first say precisely what the results should be, the most important metric of which is 'how many pixels make up the width of the blur in the final result?".

This is well-defined if we pick an exact well-established blur type,
e.g., Gaussian.

> As far as how smooth the blend is, or how the shades are distributed within that space, I think that's less important, as long as the blend perceptibly fills the space it is supposed to, and is not so heavily weight to one end or the other that it noticeably changes the whole character of the blur. If an implementor wants to improve performance by, say, having 2 identical opacity pixels adjacent to each other in a large blur, rather than having a tenth of a percent of difference between them, I think that would probably be fine, as long as the blur width (or blur extending amount) matched what the author asked for.

That's not so easy to automatically test, and results will vary more.
I don't see the problem with setting a reference in terms of Gaussian
blur, and then for those who can't implement it exactly, permit some
level of deviation (either explicitly or just assumed).
Received on Friday, 16 July 2010 22:21:11 GMT

This archive was generated by hypermail 2.3.1 : Tuesday, 26 March 2013 17:20:29 GMT