Test oklab()/oklch() with lightness close to 0/1

[foolip]

This is an interop issue where Chrome's behavior is different from
Firefox and Safari's: https://crbug.com/329106317

These tests don't care what the color is, as long as it is the same.

[foolip]

I am adding these tests so that implementing the same behavior as Chrome would fail these tests, to make the discontinuity that results from special-casing lightness 100% very obvious.

[foolip]

It worked! https://wpt.fyi/results/css/css-color?label=pr_head&max-count=1&pr=45073

[foolip]

I've updated this PR to also test a color with lightness close to 0.

[foolip]

https://wpt.fyi/results/css/css-color?label=pr_head&max-count=1&pr=45073 is updated now and what I expected. The ones close to 0 aren't very interesting to look at, the problem is that the channel values aren't quite zero, but you need to know it to see it.

Here is Chrome, Firefox and Safari for oklab-l-almost-1.html, since there aren't screenshots for passing tests.

Chrome:

Firefox and Safari both look like this:

That's it. Please review :)

[mdubet]

The tests look very convincing to me that this discontinuity is very unfortunate 👍 But are they following the spec ?

The wording of the spec "displayed as black ..due to gamut mapping to the display" seems to indicate that the discontinuity should not happen because a proper gamut mapping should prevent it ? Which mapping ?

cc @svgeesus

[foolip]

AFAIK, the spec doesn't say anything normative to the effect of "no large discontinuities" so the tests are written from a place of "something this close must round to the same thing".

I deliberately avoiding depending on what the color(s) actually are, to shine a light on the interoperability issue here. Other tests could test the colors of 0 / 0.0001% / 99.9999% / 1, it's just not what I was trying to isolate out here.

[dbaron]

I'm hoping @svgeesus can look at the substance here, but I have one wording comment:

[foolip]

I've sent #45092 to update the tests I took inspiration from. If that lands, I can make these tests match that style.

[foolip]

I was impatient, so I've updated to match #45092 before it lands. It's easy to change the class names or wording if a reviewer asks for that.

[svgeesus]

The wording of the spec "displayed as black ..due to gamut mapping to the display" seems to indicate that the discontinuity should not happen because a proper gamut mapping should prevent it ? Which mapping ?

The spec used to say that L <=0 (black) and L>= 100% (white) got clamped, which effectively mandated a discontinuity between, say, 99.98% and 100.02%.

It was changed to say that this was caused by gamut mapping to the display, which has two benefits:

• no sudden discontinuity as you approach the white or black point, regardless of the chroma
• saying that un-displayable colors have to get mapped to a displayable color seems clear and obvious.

[svgeesus]

Yes, the special casing is no longer in the spec so this undesirable discontinuity used to be spec compliant but no longer is.

[foolip]

Thanks for review and the spec background @svgeesus!