user session features don’t work properly unless your DE sends session start/end information to systemd

Which features, exactly? I just tried IceWM, which has no systemd-related dependencies and vastly predates systemd, and the session appears correctly on loginctl and disappears from there a few seconds after logging out, same as logging in and out of Plasma. Seems like it works fine.

I did notice that loginctl lock-session doesn't work with IceWM, and presumably neither does anything else that involves sending D-Bus messages to the process controlling the session, but that's not the end of the world.

this breaks various systemd features in surprising ways; I found out about this when my user services wouldn’t work

I definitely have not observed this issue. I have loginctl enable-lingered myself, so my user services start during boot, before any desktop environment is loaded. I haven't tested whether user services work in IceWM without that, but as far as I know, user service managers are started and stopped by logind in response to session start/stop, and logind gets notified of session start by the PAM module pam_systemd, not by the desktop environment.

systemd forks tend to accumulate compatibility issues with the rest of userland very quickly due to breaking API and functionality changes in the interdependent systemd process ecosystem (and these breakages can very quickly propagate to downstream programs — a breakage in logind can be expected to be catastrophic for auth in general, for example).

Breaking changes affecting programs outside of the systemd suite? Can you give me some concrete examples of such breaking changes and the problems they caused? I wasn't aware there were any. I would have expected to see some serious fireworks if such a thing ever happened.

socially, both cryptocurrency projects and systemd possess notably toxic communities which severely punish forks and dissent, which is also used as a mechanism by which control over the project is maintained. the upshot to this is an additional high cost to the morale and community resources of a fork, which particularly harshly punishes forks run by individuals and small teams.

We're discussing a community hard fork that leaves IBM behind, like what happened with XFree86. What IBM says or does after that is irrelevant, I would think.