There are some pickle benchmarks in the benchmark suite. I don't know if they exercise this case (objects without __module__), but I'll run the benchmarks on this PR just to ensure there is no unintentional regression.

This does seem to make the unpickle_pure_python benchmark about 2% slower. I haven't dug in to see whether it exercises the case where __module__ is missing, but maybe we are making the common case slower to make an uncommon case faster here?

2% is so small difference, that it is most likely a random fluctuation or the compiler fluke. Indeed, the test data does not contain functions or other objects that can be be affected by this change.

On other hand, the following example (pickling a function without __module__) runs more than twice faster with this change:

import sys
sys.modules['_pickle'] = None
import pickle, timeit
def f():
    pass

f.__module__ = None
print(min(timeit.repeat('pickle.dumps(f)', globals=globals(), number=10**4)))