Another push for sched_ext

> Witness the metric ton of toy schedulers written for it, that's all effort not put into improving the existing code.

I would expect most of the work there to be not so much about "code" but much more about experimenting with and benchmarking a very wide range of workloads. It sounds like the real question here is not really about "improving existing code" but rather "Can a unique scheduler do everything well enough?" If feels like a very difficult question but it's probably even harder to answer with less room to experiment...

> While shipping out-of-tree code is often discouraged, it can also serve to demonstrate interest in a feature and flush out any early problems that result from its inclusion. If things go well, this practice can strengthen the argument for merging the code into the mainline, albeit with the ever-present possibility of changes that create pain for the early adopters.

Absolutely: trial and error. Sometimes that requires actually shipping it to gather measurements and experience on a very large scale.

I heard some guys even send rockets in space nowadays with the expectation that they will fail.