Decoding of UTF-16 C strings is broken

[mhsmith]

In the context of a Java interface (jchar is uint16_t):

cdef const jchar *jchar_str = ...
unicode_str = (<char*>jchar_str)[:str_len * 2].decode("UTF-16")

Cython-generated code:

 __pyx_t_3 = __Pyx_decode_c_string(((char *)__pyx_v_jchar_str), 0, __pyx_t_5, NULL, NULL, PyUnicode_DecodeUTF16); if (unlikely(!__pyx_t_3)) __PYX_ERR(5, 99, __pyx_L1_error)

But this is not correct, because the last argument of __Pyx_decode_c_string has the signature PyObject* (*decode_func)(const char *s, Py_ssize_t size, const char *errors), (which are the three parameters of PyUnicode_DecodeUTF8), whereas PyUnicode_DecodeUTF16 has an extra parameter, int *byteorder.

This results in a compiler warning passing argument 6 of '__Pyx_decode_c_string' from incompatible pointer type, and I don't expect the code will work correctly.