Deploying changes to `color(rec2020 ...)`

[svgeesus]

TLDR; Do we just make this backwards-incompatible change to remain in sync with Color 4, or continue to provide the old implementation as well (with some different name)?

An issue was raised on CSS Color 4 to say that we were using the wrong transfer function for the BT.2020 color space. Instead of the inverse of the OETF in BT.2020, we should be using the EOTF in BT.1886, because we say that rec2020 colorspace is display-referred.

• Define rec2020 color space to use 2.4 gamma w3c/csswg-drafts#12574

I have prototyped this change in a fork of color.js and the changes are clearly visible

Last week, CSS WG resolved to make the change. CSS Color 4 has not been updated yet but will be soon. WPT will also need to be updated, which will make all browsers fail two of those five tests (tests on component values 0 or 1 are unaffected by a change in transfer function).

I can see four ways to make this change (on a scale from cautious to move-fast-break-things):

1. Announce an upcoming breaking change in the next color.js beta release notes, see if anyone complains, meanwhile do nothing for a while
2. Add a new --rec2020-proposed color space, leaving the old one as-is to not break anything
3. Change the existing rec2020 but add a new color space called rec2020-oetf or something, with the existing implementation
4. Just change the existing rec2020, because apart from tests and demos there is little real-world usage

[facelessuser]

I would prefer 3 or 4. Change it as a breaking change and note it in the changelog, and optionally, the old version could be added as its own space if desired. I'm fairly certain most people targeting this library are expecting it to align with CSS, so a change to do so doesn't seem too surprising to me.

[facelessuser]

Yeah, EOTF in BT.1886 definitely has a different response.

[svgeesus]

My PR uses option 3; fix rec2020 for spec parity, but save the old implementation (with a new name, as a custom colorspace, and note that it is scene referred). Anyone who was actually using it for that (to estimate scene light) has a path forward. Most users will want what CSS says, so that is the default.