Here is a minimal example how to reproduce this issue, extracted from
UTF32LETest.


valgrind --db-attach=yes --suppressions=Misc/valgrind-python.supp ./python uninitialized.py


It seems that in Lib/codecs.py the equivalent of "\x00".decode('utf-32-le')
is called (line 477).

This leads to:

PyUnicodeUCS2_DecodeUTF32Stateful (s=0x5b0fc0c "", size=1, errors=0x5add144 "strict", byteorder=0x7fefff39c, 
    consumed=0x0)


So we have a string of size 1, but s[2] and s[3] will be accessed
in the function.