[css-position-3] Auto margin resolution isn't symmetrical.

[bfgeek]

https://drafts.csswg.org/css-position-3/#abspos-margins

Otherwise, the remaining space is calculated as the size of its inset-modified containing block in the relevant axis minus the box’s used size in the relevant axis, and this remaining space is divided among any auto margins in the relevant axis. However, if the remaining space is negative and both margins are auto, the start margin resolves to zero and the end margin receives the remaining space.

There is a sutle different between the horizontal and vertical (or inline, and block? - needs research) algorithms:
https://drafts.csswg.org/css-position-3/#abs-non-replaced-width
https://drafts.csswg.org/css-position-3/#abs-non-replaced-height

See this testcase:
https://www.software.hixie.ch/utilities/js/live-dom-viewer/?saved=8318

In the vertical (or block, but which block?) direction negative auto margins get distributed evenly between the top and bottom, where-as horizontal (or inline) the inline-start margin is zero.

[bfgeek]

cc/ @fantasai

Unsure what all the UAs do for different writing modes.

[Loirooriol]

In the vertical (or block, but which block?)

Chromium seems to use the wm of the abspos in order to decide between the algorithms, and Firefox the wm of the containing block. But in both browsers, the direction that decides which size becomes zero is the one of the containing block. https://www.software.hixie.ch/utilities/js/live-dom-viewer/saved/8322

[fantasai]

@Loirooriol For this kind of thing, we generally use the writing mode of the containing block, so I think Firefox is more correct in that respect.

[fantasai]

Edits checked in to make the behavior match CSS2, translating to vertical writing modes per containing block.

[tabatkins]

And specifically, we aligned with Firefox's behavior (using containing block for both the axis and direction), because using the WM of one element but the direction of another, like Chrome does, is almost always wrong.

[bfgeek]

Are there additional tests for this now?