There's this dynamic in the industry in which a brash young project comes out swinging against some established technique and commits early to its opposite. Then, as the project matures, its authors begin to understand why the old ways were the way they were and slowly walk back their early stridency --- usually without admitting they're doing it.

Consider Linux. Time was, metaprogramming was BAD, C was all you needed, we didn't need dynamic CPU schedulers, we didn't need multiple LSMs, and we sure as hell didn't need ABI stability. Now we have forms of all of these things (for the last, see CO-RE), because as it turns out, they're actually good.

In Python, well, turns out multiprocessing isn't all you need, and a free-threaded Python has transitioned from impossible and unwanted to exciting and imminent.

In transfer encodings, "front end" people thought that JSON was all you needed. Schemas? Binary encodings? Versioning? References? All for those loser XML weenies. Well, who's the weenie now?

And in Rust? Well, let's see. Turns out monomorphization isn't all you need. Turns out that it is, in fact, occasionally useful to unify an object and its behavior in a runtime-polymorphic way. I expect yeet_expr to go through eventually too. Others are trying to stabilize (i.e. ossify) the notionally unstable ABI, just like they did to poor C++, which is now stuck with runtime pessimization because somebody is too lazy to recompile a binary from 2010.

As surely as water will wet and fire will burn, the gods of Java with stupidity and complexity return.