Ian Douglas Scott

> Which is generally the same as what auto does in CSS This is not true when `width` and `height` are set differently, when scaling up, at least. You can...

It seems even `width: 100%; height: min-content` expands the height this way? I might expect `min-content` to be more link `Shrink`. `object-fit: contain; width: 100%; height: 100%` does seem to...

The API isn't great here, but I don't know if anything better is really possible. Otherwise this should work fine. (While I think https://github.com/Smithay/smithay/pull/750 isn't quite right.)

Merging it seems good to me, given it's useful to support and I can't really think of a better way to do so.

Yeah, I was wondering if it was done this way for a reason like that. The current way it's implemented doesn't look bad for a lot of images, but results...

Also I think the advantage of atlasing is usually related to reducing draw calls? Currently there's one draw call per image. And the overhead is more a concern when drawing...

> I wouldn't reuse the logic of an image atlas for a text atlas. They are completely different use cases. Probably, though I don't know what would be correct there....

Performance with GPUs isn't always particularly logical. I've seen comments about instanced rendering preforming worse than a theoretically less "optimal" approach. And worse, it varies by GPU and driver.

I wanted benchmarking of some kind to answer questions about how the glow and wgpu backends compare in performance, whether or not the texture atlas performs better than separate textures...

Good point about Bevy. It makes sense that this is something they have worked on too.