Been there before, usually because you always feel like there is "one more thing to get into the release" before you consider it "done enough", and that keeps moving. At one point, you've diverged far enough from the previous release, either by time or scope, that now you feel like the next release really should come with major improvements because of other API breakages already, so now you want to fit in more, so the next release after that can skip more breakages.

Rinse and repeat over months, with volunteers, in a game engine no less, and I can easily see many projects being unable to not fall into that trap.

Anyways, I think it's less of an issue for people in practicality, most people who use LÖVE today tends to start with the HEAD source version, which also sets them up to easier contribute back upstream, when they inevitably hit something non-optimal, so maybe it works out in favor for everyone in the end anyways.

Since we've stepped from interpreted language (Lua) to compiled-to-VM language (C#), let's go all the way down to compiled, low-level language (C) with Raylib!

https://www.raylib.com/