import functools
def f():
    def g():
        yield 1
    return g

functools.partial(f())().__qualname__ != f()().__qualname__

The qualified name gets passed in via the interpreter loop with _PyEval_EvalCodeWithName calling PyGen_NewWithQualName.  If a generator object gets called from C then the qualified name gets lost.

It seems like _PyEval_EvalCodeWithName shouldn't exist and the generator code object should be able to get back to its qualified name so subtle differences like this don't happen.

This bug is fixed in 3.9 (probably in 3.8 too)

Python 3.9.0a0 (heads/bpo-31722:fc4a044a3c, Oct  4 2019, 03:10:14)
[Clang 11.0.0 (clang-1100.0.33.8)] on darwin
Type "help", "copyright", "credits" or "license" for more information.
>>> import functools
>>> def f():
...     def g():
...         yield 1
...     return g
...
>>> functools.partial(f())().__qualname__ == f()().__qualname__
True
>>> functools.partial(f())().__qualname__
'f.<locals>.g'
>>> f()().__qualname__
'f.<locals>.g'

Looks like that this issue is fixed