In additional to C API functions PyTuple_Size() and PyList_Size() there are fast macros PyTuple_GET_SIZE() and PyList_GET_SIZE() for getting the size of a tuple or a list. But for dicts there is just PyDict_Size(). It is not just slower than a macro, but can return an error (actually this never happens in CPython core and extensions).

Proposed patch adds new private macro _PyDict_GET_SIZE() and makes the code using it instead of PyDict_Size(). I don't expect significant performance gain except perhaps few checks that a dict is empty. The main advantage to me is that not checking the result for error no longer looks as potential bug.