[css-values] Behavior of NaN in pow()/hypot()

[tabatkins]

While fixing #4158, I stumbled across the fact that Math.hypot(Infinity, NaN) yields Infinity, and Math.pow(NaN, 0) yields 1 (both in spec and impl). These are the only two cases in the math functions I could find where NaN wasn't "infectious".

Do we want to carry this behavior along to CSS? Should we specify that NaN is specially handled in these functions? Or perhaps officially undefine it, allowing either "infectious" or "JS" behavior as valid?

[Loirooriol]

IMO the infectious behavior makes sense if when replacing each NaN argument with a variable (over non-NaN numbers), the resulting function is not constant.

But here hypot(+∞, x) must be +∞ no matter what x is, and pow(x, 0) must be 1. So extending this for NaN doesn't seem unreasonable. Though max(+∞, NaN) = NaN, so it's not much consistent.

[tabatkins]

Right, that's clearly the logic being used here; replacing NaN with any number wont' change the result.

But yeah, max() and min() are fully infectious both in CSS and JS, so they're not consistent with the reasoning, sigh.

I'm thinking I'll just stick with infectious NaN, and call it out as an intentional divergence.

[Crissov]

Internal consistency seems more important than external compatibility indeed, so, in CSS, NaN needs to be infectious either always or never (if treatable).