[css-align] place-items shorthand has wrong grammar

[MatsPalmgren]

https://drafts.csswg.org/css-align/#place-items-property
The grammar allows auto in the first (or single) value.
It shouldn't allow it because align-items doesn't have an auto value:
https://drafts.csswg.org/css-align/#propdef-align-items

So, I think the correct grammar should be:

<x> = [ normal | stretch | <baseline-position> | <self-position> ]
place-items:  <x> [ auto | <x> ]?

IOW, auto should only be allowed as a second value.

[fantasai]

I'm assuming nobody implements the legacy value. I think we should just get rid of it until we have a clear indication that that's how we want to solve <div align>; that would also get rid of auto here.

[MatsPalmgren]

I'm assuming nobody implements the legacy value.

Firefox implements all values for justify-items, including legacy and the special rules about that.

A quick check in Chrome devtools shows that they support legacy too.

[fantasai]

Ok, wow. I remember dbaron raising concerns about that. Is that replacing the text-align values for implementing <div align> now?

[MatsPalmgren]

No, Gecko only supports CSS Align in Grid and Flexbox layout currently.
We do not support any of the CSS Align properties in Block/Table/Multicol layout at all.
We intend to implement layout support for CSS Align on a "per module" basis, i.e. support
all of it for Block, then support all of it for Table etc. Doing it all at once is just too much
with Mozilla's limited resources. We could of course delay Grid and Flexbox support
until we've implemented it everywhere, but that would delay Grid for a year or more
and I don't think anyone wants that.

Note the word layout above though. The CSS Align properties are fully implemented
and enabled in the style system and that affects all box types, so they will show up as
computed values on blocks etc, it's just that they have no effect on the layout, except
for Grid/Flexbox.

(I think Chrome is taking a similar approach but I don't know the details.)

[MatsPalmgren]

The reason behind taking this all-or-nothing-per-module approach is to make it
easier for authors to know where these properties are supported. For now they
are only supported in Grid/Flexbox layout, and not at all on anything else. I think
it would be confusing for authors if the properties are supported in varying
degrees in different modules by different UAs. That would lead to web-compat
problems eventually. So I'm hoping that other UA vendors take a similar approach.
Perhaps we should coordinate what to support next? (I think Block layout is the one
we've been discussing for Gecko.)

[fantasai]

I agree with your approach, and coordination sounds good.

Wrt legacy, I think we need to decide if the approach in the draft is actually one we want to take for handling <div align> etc, since that's the only reason it exists. The trickiest part of that is the style layer, though, so if you've already got that sorted maybe it's okay.

[javifernandez]

Hi,

On 11/26/2016 05:17 PM, Mats Palmgren wrote: I'm assuming nobody implements the |legacy| value. Firefox implements all values for |justify-items|, including |legacy| and the special rules about that. I quick check in Chrome devtools shows that they support |legacy| too.

Both Chrome/Blink and Safari/WebKit implements legacy keyword.

[javifernandez]

Hi,

On 11/26/2016 10:27 PM, Mats Palmgren wrote: Note the word /layout/ above though. The CSS Align properties are fully implemented and enabled in the style system and that affects all box types, so they will show up as computed values on blocks etc, it's just that they have no effect on the layout, except for Grid/Flexbox. (I think Chrome is taking a similar approach but I don't know the details.)

Our opinion as implementors of the spec in Chrome was to enable the CSS Alignment feature, replacing the old syntax implemented by FF. The Chrome community agreed and prioritized web compatibility, so they recommend to follow what FF eventually does on this regard. We are not sure what WebKit will do, though, but we expect to follow the same approach. Other than that, the 'legacy' implementation introduced several complexities in the style resolution logic; in the same line than the effort we had to do to satisfy the 'auto' value resolution for the self-alignment properties. Now that have have got rid of those 'auto' values resolution, it's the perfect moment to take a decision on this legacy keyword issue.