[css-sizing] Intrinsic size of replaced elements incorrect

[fantasai]

Currently Sizing 3 defines intrinsic sizes of replaced elements explicitly.
This definition isn't correct, and it should be deleted in favor of falling through to the CSS2.1 reference used for non-replaced elements.

[fantasai]

Agenda+ to point out that the implication here is that, when an aspect ratio is present, min-content and max-content are dependent on sizing constraints in the opposite axis. (This might be surprising/disturbing to implementers but is probably as expected for authors.)

[dbaron]

I really don't want min-content and max-content to be anything other than intrinsic sizes. And, in particular, I want them never to depend on the element's own specified sizes, precisely so that they can be used as keywords for the element's specified sizes.

Doing so could cause problems down the road if we ever add the ability for calc() expressions to reference concepts like "this element's min-content size in the other dimension" or "the grandparent element's size". (The latter is harder than the former, but we might be able to come up with a set of ways where it does make sense or a way of resolving the constraints, and having things like "intrinsic" sizes that are almost but not quite always intrinsic would make that harder.)

[fantasai]

The problem with that is then we don't have a keyword for "shrinkwrap this thing as if it was a float with (infinite/minimal space)", which is what you generally want in e.g. grid track sizing.

[dbaron]

I just discussed this a bit with @fantasai.

To clarify what I want instead: I'd like the width and the height to both be considered when computing an element's min-content contribution to its parent. Right now (assuming we're talking about English layout, i.e., min-content as a width), it's clear that for width, an element's min-content width doesn't consider its width property (this allows things like width: min-content). What we're disagreeing about is whether this is the same for how height is considered, on elements that have an intrinsic ratio. I'd like the height to be considered at the same stage as width, whereas @fantasai wants it to be considered earlier, so that it's incorporated into what the min-content value means.

I don't think the grid track sizing thing is an issue, since that should be depending on the min-content contribution, which is affected by the aspect ratio with either proposal.

@fantasai's approach has the advantage (for the spec authors, and maybe implementors) that defining what width: auto means is much simpler. With my approach, the meaning of width: auto is more complicated since it has special rules for replaced elements.

There isn't a distinct author feature in @fantasai's proposal, I don't think; auto will still work the way she wants min-content and max-content to behave. So with what I'm proposing there are more behaviors exposed to authors, although they seem unlikely to be useful (although I believe they are underlying primitives).

I still think my proposal is more future-proof for features like the ones I described above in #794 (comment)

[fantasai]

Propose to close as fixed, after f2f discussions with dbaron in SF.