Thanks; can you split out the Argument Clinic adaption and contribute that as a separate PR?

Sure, see #115242 .

Hi @erlend-aasland, I have a new thought. This change is just to isolate these two static variables:

But as they are the state of a pseudo-random generator, they can be shared between interpreters under a lock. For single interpreter usage, there is no noticeable performance decrease. And this work can be done easily.

I have no idea if this approach is acceptable, or whether we should continue with the previous work. If it does, I can do some benchmarking to see if there is a performance decrease in multiple isolated interpreters usage.

Sounds reasonable. Go ahead!

Superseded by #115301