[css-color-5] Testable but untested sections

[svgeesus]

• Mixing colors: the color-mix() function

Has 2 simple tests, both marked tentative and a very comprehensive parsing test

• Selecting the most contrasting color: the color-contrast() function

Has some tests. Spec examples would be a start.

• Relative color syntax

Now has some tests

• Custom color spaces

• Specifying a color profile: the @color-profile at-rule

  • one simple test

• CSS and print: using calibrated CMYK and other printed color spaces

  • This section is primarily examples and explanation; it does not contain normative statements

• Device-dependent CMYK Colors: the device-cmyk() function

Hard to test reliably, but at least an @supports test

• Resolving Values

• Serialization

Has some tests, using computed value

[weinig]

I have a few for parsing and serialization and color-mixing I would be happy to convert to wpt and donate if you would find them helpful (some are being updated for the recent changes right now, so may be a bit out of date):

https://github.com/WebKit/WebKit/blob/main/LayoutTests/fast/css/parsing-color-contrast.html
https://github.com/WebKit/WebKit/blob/main/LayoutTests/fast/css/parsing-color-mix.html
https://github.com/WebKit/WebKit/blob/main/LayoutTests/fast/css/parsing-color-function.html
https://github.com/WebKit/WebKit/blob/main/LayoutTests/fast/css/parsing-relative-color-syntax.html
https://github.com/WebKit/WebKit/blob/main/LayoutTests/fast/css/parsing-lab-colors.html

My biggest concern is with floating point precision differences between browsers in some of them (since precision of component storage is only defined with a minimum bound among other issues).

Let me know and I will make some time to get out a PR.

[svgeesus]

Yes, these would be super helpful please do.

My biggest concern is with floating point precision differences between browsers in some of them (since precision of component storage is only defined with a minimum bound

Maybe round to the minimum bound then compare observed - expected < epsilon?

among other issues).

Other issues?

[svgeesus]

Please do add the rel=help pointing to the actual spec sections, though.

• writing the test file
• CSS metadata

So for example on the color-mix() test add

<link rel="help" href="https://drafts.csswg.org/css-color-5/#color-mix"/>

[weinig]

among other issues).

Other issues?

I meant things like the precision of the color conversions themselves, for issues like how many digits of significance are used for the matrices, since the exact algorithms are not defined and floating point operations can introduce error with small differences. That said, I think there is far more benefit in having the tests now, and figuring out how to solve those than not.

[svgeesus]

Oh, I see. Yes, the general problem with floats and precision, especially cumulative roundoff error.

Agree it is far better to have good tests now and improve them over time, than perfect tests never.

[weinig]

Put up a few PRs:

• color-mix() - Add tests for CSS Color 5 color-mix() web-platform-tests/wpt#31902
• color-contrast() - Add tests for CSS Color 5 color-contrast() web-platform-tests/wpt#31903
• relative colors - Add tests for CSS Color 5 relative color syntax web-platform-tests/wpt#31904

Also a few CSS Color 4 ones:

• reference tests for new colors - Update tests for new color spaces in CSS Color 4 web-platform-tests/wpt#31901
• parsing and serialization for new colors - Update basic color parsing and serialization tests to cover the CSS Color 4 color grammar web-platform-tests/wpt#31905

[svgeesus]

These are all merged now except that @tabatkins asked for changes on RCS and I don't understand why. He comments about not producing integer values, yet they are not required by spec because serializing says that decimal values are allowed in rgb()

[tabatkins]

It appears you understand my comments exactly backwards. ^_^ Sam's tests were testing serialization of %s like 10% or 25%, which produce non-integer values when converted to numbers, while their expected results were integers (which is allowed but not required, as you note). So I was asking them to either allow multiple correct values, both rounded and unrounded, or ensure that all %s tested were multiples of 20%. (They did the latter, so we're good to go now; I merged the tests.)

[svgeesus]

Aha, got it.

[svgeesus]

I have a PR for an initial @color-profile test - web-platform-tests/wpt#32157 uses a red-green swapped ICC profile. Now merged

[svgeesus]

Closing since the spec is now fairly well tested, with the exception of device-cmyk() which seems hard to produce a reftest for, as uncalibrated color can't be directly compared.