The function

has a rather peculiar way to handle memory. It takes a sequence of Python bytes / text and converts it to a NULL terminated array of char *. It allocates memory for the array itself but uses PyUnicode_AsUTF8 or PyBytes_AsString for the elements inside the array. Both functions do not return a copy of the string but rather a reference to an internal buffer. The internal buffer is shared between callers and must be treated as const, immutable block. In fact in Python 3.7+, the return value of PyUnicode_AsUTF8 has been changed to const char*, https://docs.python.org/3.7/c-api/unicode.html#c.PyUnicode_AsUTF8AndSize . However ldap_search_ext does not take a const char*[] but a char *attrs[]. This breaks the const promise starting with Python 3.7.

Even more peculiar is the seq argument. It is used to pass back a reference to the Python sequence that holds the PyObject *. This trick is used to keep the Python sequence and all its members alive, so the elements in the array of char * do not get deallocated by Python.

I suggest to change the function to copy the strings to allocated buffers instead of (ab)using Python objects to provide the buffer. The performance impact of additional PyMem_NEW and memcpy is negligible.