[css-sizing] Aspect-ratio min-size needs to check for "auto", not non-zero

[tabatkins]

In https://drafts.csswg.org/css-sizing-3/#intrinsic:

Otherwise if the box has a definite non-zero min-width or min-height, use that size and calculate the other dimension using the aspect ratio; if both dimensions have a definite minimum, choose the one that results in the larger overall size.

This should probably be checking for non-"auto", not non-zero. (Open ranges are bad.)

[fantasai]

Checked for definite generally, since min-content and percentages are non-auto but still problematic. Lmk if that seems correct.

CC @Loirooriol for sanity-checking :)

[Loirooriol]

@fantasai I think that could clarify that the value to be checked for definiteness is the computed value, not the used value, since in most cases an auto minimum size resolves to a definite 0.

[fantasai]

It's currently specced as if the box has a [=definite=] non-''min-width/auto'' value, thanks to ce03a20 ...

[tabatkins]

@Loirooriol We went ahead and simplified it further, to explicitly check for a computed <length>. Since this isn't matching impls or anything, just trying to make a slightly smarter default than 300x150, we figure that's probably okay. This work for you?

[Loirooriol]

Checking for a computed <length> seems a bit strange to me, because 100% is a <percentage>, but calc(100% + 100px) is a <length> when used in min-width if I get https://drafts.csswg.org/css-values-4/#determine-the-type-of-a-calculation right?

[tabatkins]

Not quite! Following the definitions, the type of 100% + 100px is obtained by adding the types of 100% and 100px; following that link to TypedOM, we find that the resulting type is "length, with a percent hint set to length".

Then, a calc() matches a production if its type matches that production; following the link again to TypedOM, a type with a percent hint can not match <length> (because it contains a non-resolved percentage that we're inferring to be compatible with <length>).

[tabatkins]

@Loirooriol So, with that explanation, does the spec seem okay?

[Loirooriol]

Yes, then it looks good to me