The width attribute

[yoavweiss]

During discussions about Client-Hints, where it makes sense to include the width attribute in the outgoing Width hint, since it can be useful to retrofit responsive images to legacy sites, @foolip suggested that it may make sense to include the same logic in srcset, and I think it makes a lot of sense.

Basically, the same logic applies. If sizes is missing and width is there and contains a non-% value, we should use the width value as the source-size-value (or add some "if sizes is not there, use width logic to the consumer of "parse a sizes attribute".

@zcorpan - WDYT?

[foolip]

My main concern here is that it would be nice if the computed source-size is the same as what is sent in the Width header. (Which is exactly the suggestion, of course.)

[joemcgill]

If I'm understanding the suggestion properly, i.e., use the width value as the default source-size-value instead of 100vw if the value is present (and a non-% value), then I'm mostly in agreement, and this is basically the logic we use to create a "smart default" for the sizes attribute in the WordPress plugin (see: ResponsiveImagesCG/wp-tevko-responsive-images#34).

The one suggestion I would propose, is to only use the width value when it is less than 100vw (if that's even possible). Otherwise, where someone has CSS for images set to max-width: 100% but has failed to set a sizes attribute, they could end up with a much larger download than needed.

[yoavweiss]

Thanks @joemcgill :)

The change I'm proposing only relates to the width HTML attribute, so a max-width: 100% setting won't have any impact.

[joemcgill]

Right, I understood, but was probably unclear in my description. I just meant that if an inline width attribute is used as a default when sizes is omitted (in the case of extending this logic to srcset calculations), then it may be preferential for the consumer to only use the width value when it's smaller than the viewport width.

[igrigorik]

bump ... any blockers to resolve this? This is an important feature for Client Hints.

[tabatkins]

I'm down with this. It makes sense to me.

[zcorpan]

OK with me, but we should ignore it when it uses percentages (since it refers to containing block, and it's invalid anyway). Also I'm not sure I like @joemcgill's idea to have the viewport width be an upper bound, that seems unnecessary magic that could get confusing results, especially for Client Hints. You can still achieve the effect by using the sizes attribute, but then it's explicit and intentional.

[joemcgill]

seems unnecessary magic that could get confusing results

I agree. Probably best to keep things straightforward.