I'm overall in favour of doing this. Though I wonder if there will be code that relies on the fact that ints between -x to 256 are cached and everything else isn't. I'm somewhat worried some weird code out there relying on the current int interning will break. Does anyone know?

(This includes both C API code and Python code).

I'm somewhat worried some weird code out there relying on the current int interning will break.

IMHO, at a certain point we can't accommodate every way people might rely on some implementation detail. In this specific case we've been fairly clear in documentation (and elsewhere) about not relying on identity. We'll also point out the change in the release notes. Again IMHO, as long as are thoughtful about impact, we should expect that users (taking advantage of impl-specific details like this) take responsibility for adjusting their own code. "We're all consenting adults."

Don't wait on me, I'm not blocking it!