[css-sizing][quirks] Should `stretch` be affected by percentage height calculation quirk?

[Loirooriol]

https://quirks.spec.whatwg.org/#the-percentage-height-calculation-quirk

In quirks mode, for the purpose of calculating the height of an element element, if [...] the specified value for the height property of element is a <percentage> [...] the containing block of element must be calculated using the following algorithm

It only mentions <percentage>, but this was written before CSS Sizing introduced stretch.

Arguably, as per https://drafts.csswg.org/css-sizing-4/#stretch-fit-sizing

• Otherwise, if used in an axis where percentage sizes can resolve to a definite value
  Behaves as 100%, except it sizes the margin box regardless of the value of box-sizing.

if 100% would have the quirk, stretch should also have it?

Browsers don't agree: Blink and WebKit apply the quirk for stretch / -webkit-fill-available, while Firefox doesn't for -moz-available.

<!-- quirks -->
<div style="height: -moz-available; height: -webkit-fill-available; height: stretch; background: cyan">foo</div>

[bfgeek]

The testcase you have isn't actually the percentage height quirk rather:
https://quirks.spec.whatwg.org/#the-body-element-fills-the-html-element-quirk

Basically the element fills the ICB, and then the element stretches to that definite height.

(That quirk is as-if html & body have height:stretch by default - kinda).

A better test would be:

<!-- quirks -->
<div>
  <div style="height: -moz-available; height: -webkit-fill-available; height: stretch; background: cyan">foo</div>
</div>

And blink doesn't stretch.

[Loirooriol]

OK, so Gecko and Blink don't use the quirk for stretching, and WebKit uses it even on standards mode.

IMO it seems simpler if everything uses the same reference value, to avoid having to track both the actual containing block height and the quirky containing block height. If stretch uses the quirk, then I think I can just pretend that the containing block height is what the quirk needs, and I don't need any extra piece of data.

[bfgeek]

IMO it seems simpler if everything uses the same reference value, to avoid having to track both the actual containing block height and the quirky containing block height

You need to track this anyway, e.g. in standards mode you need to track the available-size, and the percentage-resolution-size - they are mostly they same, but can differ (in abspos, floats, etc).

I'm strongly against introducing the quirk for stretch.

[Loirooriol]

I still haven't implemented the quirk in Servo, so I may be missing something. But I don't follow your examples.

For abspos, the quirk doesn't apply, so their containing block is the same as in no-quirks mode. And in no-quirks mode, Servo just uses the block size of the containing block (which is always a definite amount), we don't track the available-size and percentage-resolution-size as different things.

For floats, percentage heights have the quirk. I'm not sure when floats would use the available-size instead of the percentage-resolution-size if they differ.

[bfgeek]

I still haven't implemented the quirk in Servo, so I may be missing something. But I don't follow your examples.

Ah we store the percentage-size & available-size directly - and Servo (from your description) uses the containing-block directly.

I was basically referring to things that have a different available-size, and percentage-size.

[zcorpan]

I think we generally shouldn't make quirks apply to new things. If we want stretch to match what WebKit does for height: -webkit-fill-available, that could be done for all rendering modes.