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

On Tue, Jul 27, 2010 at 2:29 PM, Brad Kemper <brad.kemper@gmail.com> wrote:
> On Jul 27, 2010, at 9:44 AM, "Tab Atkins Jr." <jackalmage@gmail.com> wrote:
>
>> We can automate it, we just have to build the automation manually -
>> screenshot and then run pixel comparisons.
>
> So what would you use for for the screen shot? The approximate Gaussian or the true Gaussian?

You'd compare it against a true gaussian.  You don't actually have to
do a screenshot *comparison* - just take the one screenshot, and
compare each pixel's value against the predicted value given by the
gaussian distribution.

> Would this mean that if a better algorithm is used that better approximates natural shadow blurs in half the time of a Gaussian then it isn't allowed? Or that a simple stepped blend with appropriate corner rounding (optimized for a number of steps that were good enough for the size of the e-ink halftone dots it used, say) wouldn't be allowed if it was on hardware optimized to render it that way?

Gaussians are sufficiently close to physical shadows that I don't
expect this to be a problem - that's precisely why gaussians are used
for this.

When the magical day comes that raytracing with accurate diffraction
models is doable on common consumer hardware, if such a thing would be
nonconformant by our definition, we can amend it at that point.  Right
now I don't recommend trying to optimize for that case.

(My previous complaint against Aryeh's suggestion along these lines
was arguing against requiring a gaussian precisely.  The gaussian
approximation I'm advocating will probably allow physically accurate
shadows.)

~TJ