The following code causes an assertion failure:

import unicodedata
def bad_normalize(*args):
    return None

unicodedata.normalize = bad_normalize
import ast
ast.parse('\u03D5')


This is because init_normalization() (in Python/ast.c) assumes that
unicodedata.normalize() is valid, and stores it in the compiling struct.
Later, new_identifier() calls the stored function, assumes it returned a
string, and passes it to PyUnicode_InternInPlace(), which asserts it is a
string.

Oh, and passing arguments to this function is tricky. If faked unicodedata.normalize save the args tuple it will see a mutated tuple containing a hanging reference.

This is a good opportunity of using the fast call protocol.

New changeset 7dc46d8cf5854d9f4ce3271b29c21aea4872e8ad by Serhiy Storchaka (Oren Milman) in branch 'master':
bpo-31592: Fix an assertion failure in Python parser in case of a bad unicodedata.normalize(). (#3767)
https://github.com/python/cpython/commit/7dc46d8cf5854d9f4ce3271b29c21aea4872e8ad

New changeset a4dfe1c9eaca6faf206f63a178233ffea208c906 by Serhiy Storchaka (Miss Islington (bot)) in branch '3.6':
[3.6] bpo-31592: Fix an assertion failure in Python parser in case of a bad unicodedata.normalize(). (GH-3767) (#3836)
https://github.com/python/cpython/commit/a4dfe1c9eaca6faf206f63a178233ffea208c906