On Thu Nov 12, 2026 at 5:47 AM JST, Yury Norov wrote: >> Forbidding zero-sized pools would actually require *more* code in this >> patch to enforce that rule, and would not allow us to remove any in >> exchange afaict. > > I trust you here. Maybe declaring nbits in the allocation API as > NonZero would help - IDK... I just want to make sure we're not > bloating the implementation for unneeded and confusing functionality.
I think the current code is pretty much as simple as we can make it for our use-case. Of course we can always tune it and even update the API if we find out it doesn't work as well as we thought, or that allowing zero-sized pools create issues. But if we forbid them preemptively we would actually be adding more complexity, which makes me think the current approach looks right.
