Mark Shannon

Tracking just `PyStackRef_FromPyObjectNew` will solve your immediate problem, but it doesn't help with a broader deferred reference counting implementation nor with top-of-stack caching. I think what I proposed handles all...

> @markshannon did you intend for these to be private? My thinking was that it is better to keep any function private unless we need it to be public. It...

The "code like" object is entirely opaque to the monitoring machinery. As far as it is concerned, a location is a pair of an object and an integer index. Debuggers,...

In case my earlier comments were a bit cryptic: * The `active` field should only be modified by the VM, so tests will need to call `set_events` before and after...

> What do you mean by track? Stash on the thread state or in the frame stack, perhaps. The idea being that the "code-like" object can be displayed in tracebacks...

> So, state_array and a corresponding version should be allocated and used together; they must be valid from the call PyMonitoring_EnterScope to the end of the corresponding PyMonitoring_ExitScope. Typically they...

@encukou are you OK with this being merged?

What's the root cause of https://github.com/python/cpython/issues/120321? Until you know that, how do you know this fixes the problem? This feels like slapping a lock on something and hoping it works.

What about all the other places that `tstate->exc_info` is updated when we push or pop a generator frame? I suspect that this doesn't crash only because specialization is turned off...

`gen_send` isn't the problem. It contains an instance of the problematic code, but there are instances in bytecodes.c in `_SEND` and _FOR_ITER_GEN`. Pushing and popping generator frames needs to be...