A set of merge requests were opened that would effectively drop X.Org (X11) session support for the GNOME desktop and once that code is removed making it a Wayland-only desktop environment.
Going along with Fedora 40 looking to disable the GNOME X11 session support (and also making KDE Plasma 6 Wayland-only for Fedora), upstream GNOME is evaluating the prospect of disabling and then removing their X11 session support.
Some concerns were raised already how this could impact downstream desktops like Budgie and Pantheon that haven't yet fully transitioned over to Wayland. In any event we'll see where the discussions lead but it's sure looking like 2024 will be the year that GNOME goes Wayland-only.
Why should it not be fixed by the compositor, exactly?
As far as I see it, that's a smart design choice, the issue is just that we needed a universal implementation, an x.org equivalent, and we now have that with wlroots, now that that exists, there's no downsides to that approach, as far as i'm aware.
In what way exactly does that make it a smart design choice. It sounds like compositor implementers essentially have to work around the bad design choice by including a library and even then each compositor will have to update the dependency version for wlroots each time something needs to be fixed that breaks the wlroots ABI (or for containers, static linking,… just each time).
No, it sounds like compositors will use a library so that they don't have to do a shitload of work that they'd have to do otherwise.
…this is already how x.org works. You have to implement the x.org server, or create your own implementation of X11.
The only reason you think your criticism doesn't apply to X.org is because nobody updates X.org anymore… There's no more breaking changes to be made because it's a fundamentally broken, shitty protocol.