As is obvious from the code, PyModule_GetDict() can fail if being passed a non-module object, and when the (unlikely) dict creation at the end fails. The documentation of the C-API function should be fixed to reflect that, i.e. it should state that NULL is returned in the case of an error.

PyObject *
PyModule_GetDict(PyObject *m)
{
    PyObject *d;
    if (!PyModule_Check(m)) {
        PyErr_BadInternalCall();
        return NULL;
    }
    d = ((PyModuleObject *)m) -> md_dict;
    if (d == NULL)
        ((PyModuleObject *)m) -> md_dict = d = PyDict_New();
    return d;
}

The path with PyDict_New() is never taken, because PyModule_New already fills md_dict.

So, can we close this issue?

I gave two reasons why this function can fail, and one turns out to be assumed-to-be-dead code. So, no, there are two issues now, one with the documentation, one with the code.

> I gave two reasons why this function can fail, and one turns out to be assumed-to-be-dead code. 

If the call to PyDict_New() is never called, the test can be replaced with an assertion.

I've changed the dead code to

    assert(d != NULL);

and added the following sentence to PyModule_GetDict documentation:

    If *module* is not a module object (or a subtype of a module object),
    :exc:`SystemError` is raised and *NULL* is returned.

you can merge the patch.

New changeset bd2a4c138b76 by Berker Peksag in branch '3.5':
Issue #12946: Document that PyModule_GetDict can fail in some cases
https://hg.python.org/cpython/rev/bd2a4c138b76

New changeset c2e74b88947d by Berker Peksag in branch 'default':
Issue #12946: Merge from 3.5
https://hg.python.org/cpython/rev/c2e74b88947d

New changeset 4b64a049f451 by Berker Peksag in branch 'default':
Issue #12946: Remove dead code in PyModule_GetDict
https://hg.python.org/cpython/rev/4b64a049f451

Thanks for the report, Stefan!