> Why do you need to specify a `PyThreadState`?

From past issues, I understood that Julien's use case is a profiler which inspects memory without calling functions nor switching to other threads. It's like "dump the state of all Python threads".

I'm fine with exposing _PyErr_GetTopmostException() as a *private* function.  Moreover, the function was already exposed in 3.8. It was not exposed because of the visibility change which was made in 3.9.