[css-values-5] How should random() handle extremely large ranges / extremely tiny steps?

[weinig]

The definition of the CSS random() function (https://drafts.csswg.org/css-values-5/#randomness) could use clarification on the expected math for computing a uniform distribution for ranges where for finite (but very large) max - min overflows and equals infinity. A naive implementation (or one that is using say, the C++ standard library's std::uniform_real_distribution) returns infinity for all cases here, but additional work could make this work if that is desired.

A similar problem arises for sufficiently tiny step values.

For all other math functions defined by calc, there have been clear IEEE-754 semantics to follow, or explicit algorithms (as with progress()). Given the new territory here, it would be useful to be specific in the requirements.

[tabatkins]

I'm very intentionally not specifying a PRNG for this purpose, similar to how JS doesn't.

I should indeed specify the behavior if the range becomes infinite, or the step becomes 0, tho.

[tabatkins]

Okay, gave it a behavior for if the span is infinite (becomes NaN, same as if either of the range endpoints were infinite). I don't see a particular reason to try and be smarter about this.

I had a definition for a 0 step (it technically falls out of the definition already) but on review that behavior sucked; it wasn't the limit behavior as the step decreased to 0. So I ripped that out and defined instead that 0/negative/positive-but-sufficiently-small steps are just cause the step to be ignored, so it can generate any value in the range.

(Negative could be invalid, but I'm treating it like I defined the range to be [0, inf], which causes a clamp if the value is out of range.)