Using 'align-self' seems hacky

[tabatkins]

The place-self properties align an element in its container. Start/end put it against the start/end edge, etc. None of the alignment properties align a single element wrt sibling elements unless they're doing a global alignment of all the elements in a container/line (the space-* values).

As such, I don't think align-self is the right place for "align this one element within its siblings". It's squeezing in a different concept in a way that I think will bite us later.

[dbaron]

I guess it seemed the opposite to me -- it seem like this was a case where align-self didn't do anything and probably wasn't going to, except for this case, where we can actually make it do something useful that pretty much matches what it does.

[tabatkins]

The problem here is that the property then becomes context-dependent. In some cases it aligns the element within its container. In other cases it aligns an element among its siblings, relative to the line-grid. And since it's reusing the same keywords, you have to know exactly what context the element is in to understand what the property is trying to do.

This sort of bimodal design feels attractive (as you say, it's making a property do something useful in a case where it's currently not doing anything), but it's a trap! It just makes things more difficult to understand, and constrains us in how we can evolve the property in the future. (For example, you're already saying that in the line-grid context, only three keywords do anything; the rest are meaningless. Similarly, if we ever have to add more line-grid-specific controls to this property, we'll have to define that in the container-aligning case, those keywords are meaningless.)

It also means that we can't ever cross the streams, and, for example, align a grid item to the line-grid, or a flex item to the line-grid in the cross axis (it might be possible in the main axis, as it doesn't currently respond to justify-self).