Brandt Bucher

Also, it looks like `PyInterpreterState` has a similar leak.

Confirmed that setting `tstate->_static = false;` after the `memcpy` fixes the issue.

Fixed, but leaving open to track the above discussion.

I don't know if I'll have time to review again today (still need to prepare my talk for the release stream). Is this able to wait until next week? If...

Any chance we could revert the recent renaming of tstate.exc_state and tstate.root_cframe in https://github.com/python/cpython/pull/30590? It broke Greenlet again: https://github.com/python-greenlet/greenlet/issues/288 If it's only a name change (and the members themselves are...

> I think it is before - extended args are added in assemble_emit. But jump distances are computed in `assemble_jump_offsets`, which happens before this. If any `EXTENDED_ARG`s are removed from...

> I wonder why it only showed up on windows. I bet the startup sequence contains quite a bit of Windows-specific code. Probably just a really lucky/unlucky code path.

Here you go: ``` All benchmarks: =============== Slower (21): - nbody: 93.3 ms +- 1.6 ms -> 96.8 ms +- 1.9 ms: 1.04x slower - unpickle_list: 4.91 us +- 0.06...

@pablogsal, leak tests didn't catch this because we use the "raw" domain to allocate `PyThreadState` and `PyInterpreterState`.

> Oh interesting, can you confirm that changing the domain to memory makes the test fail? You can't. :) ``` Fatal Python error: _PyMem_DebugCalloc: Python memory allocator called without holding...