@tabatkins @fantasai

The Flexbox automatic minimum size rules were drafted with the assumption that only replaced elements had an aspect ratio. The Sizing rules, which make the content-based size a very strong constraint on the auto min size, are correct for non-replaced elements; but the Flexbox rules, which make the "content-based" (aka replaced element natural size) a weaker constraint, are correct for replaced elements.

Probably what we need to fix here is to distinguish content-based sizes of non-replaced elements which should be a stronger constraint than a transferred size vs content-based sizes of replaced elements (those derived from the natural size of the replaced content), which should be a weaker constraint than a transferred size.

So this needs some tweaks to merge the two into a consistent set of rules.

Agenda+ to confirm that we want to make non-replaced elements ignore any transferred preferred sizes for the purpose of automatic minimum sizing.

Remember, this is the minimum size below which a flex item, by default, cannot shrink. We don't want to be too constraining, so that when it's sensible to shrink it can shrink.

Current spec:

• if specified size exists: min(specified, content-based)
• else: min(transferred, content-based)

Proposed spec:

• for replaced:
  • if specified size exists: min(specified size, natural size)
  • else: min(transferred preferred, natural size)
• for non-replaced:
  • if specified size exists: min(specified size, content size)
  • else: content size

In all cases, we apply an opposite-axis minimum transferred through the aspect ratio if any: min(specified size, max size, opposite-min * aspect-ratio)

Alternately we can ignore the transferred min of a non-replaced element.

[All of these are clamped by the appropriate maximum sizes.]

The various inputs into this are

• definite min/preferred/max sizes in affected axis
• definite min/preferred/max sizes transferred from opposite axis (if there's an aspect ratio)
• content-based size in affected axis (if non-replaced)
• natural size in affected axis (if replaced)

(cc @aethanyc since this affects tests they wrote)

I think that is an outcome that makes sense; it is certainly consistent with block layout in the sense that aspect ratio is "ignored" if it would cause content to overflow.

flex-aspect-ratio-026.html was written with the assumption that the flexbox's automatic minimum size rules has a higher priority over the aspect-ratio's. I didn't aware for non-replaced element, applying the flexbox's rule min(transferred, content-based) somewhat contradict to the outcome that the aspect-ratio's automatic content-based min size is trying to produce.

The proposed change -- non-replaced elements ignore any transferred preferred sizes for the purpose of automatic minimum sizing -- looks reasonable to me.

(cc @dholbert)

(Also a reminder that grid probably needs a similar change)

Echoing @aethanyc :

The proposed change -- non-replaced elements ignore any transferred preferred sizes for the purpose of automatic minimum sizing -- looks reasonable to me.

Yup, seems reasonable to me, too.

OK, we've updated the Flexbox and Grid specs to exclude non-replaced elements from using the transferred size suggestion:

• https://drafts.csswg.org/css-flexbox/#content-based-minimum-size
• https://drafts.csswg.org/css-grid/#content-based-minimum-size

Note: We also redrafted Flexbox’s definition to be a bit to be easier to read and to match Grid more closely in f86bc38

Keeping Agenda+ to confirm with the CSSWG.

The CSS Working Group just discussed [css-flexbox] [css-sizing-4] Interaction of flexbox minimum height and aspect-ratio minimum height.

The CSS Working Group just discussed [css-flexbox] [css-sizing-4] Interaction of flexbox minimum height and aspect-ratio minimum height, and agreed to the following:

• RESOLVED: non-replaced elements with an aspect ratio honor the intrinsic size in flexbox for min-height:auto

I'm a little skeptical by the grid behaviour now - and it doesn't seem correct to me.

E.g. consider this testcase:
https://www.software.hixie.ch/utilities/js/live-dom-viewer/?saved=9297

is it really correct that we ignore the transferred size suggestion for non-replaced elements?

@fantasai @tabatkins

@bfgeek I think it is. The goal of the automatic minimum size is to prevent things from becoming so small that their contents are unreadable (overflowing or shrunk to zero). The preferred size still accounts for the aspect ratio, so if the column is allowed to grow wider the item will be square as intended, but when there isn't enough space in the column (and the item's own contents are small enough to fit), it won't overflow the column. That's a good behavior for the automatic minimum size, imho.

If the author wants the aspect ratio to force overflow, they can set the width or min-width to max-content instead of auto.