Message11501

I've been thinking about the relationships amongst thread state, interpreter state and sys module. It wasn't motivated by this problem, and I'm just trying to understand why it is the way it is in CPython first. It's a bit of Jython that a. has to be slightly different from CPython (no GIL), and b. we keep having to tweak.

I've concluded that, in any Python implementation, we need an instance of sys per interpreter -- or rather, an instance of "module administration" per interpreter. So our response to #2154 is surely correct. Is it just that module administration is terribly expensive? I probably made that a bit worse in my FS-encoding fixes, but I've been suspicious of the import subsystem since the first time I single-stepped through it.