[Numpy-discussion] user-defined types mixed arithmetic

[Neal Becker]

I have 2 user-defined types, and simple arithmetic working for the cases I
registered with PyUFunc_RegisterLoopForType.

I'd like to use automatic conversion to do mixed arithmetic between these 2
types.  I did PyArray_RegisterCastFunc,  and it seems this allows explicit
conversion:

>>> a
array([(0,0), (1,0), (2,0), (3,0), (4,0), (5,0), (6,0), (7,0), (8,0),
(9,0)], dtype=cmplx_int32)
>>> b
array([(0,0), (1,0), (2,0), (3,0), (4,0), (5,0), (6,0), (7,0), (8,0),
(9,0)], dtype=cmplx_int64)
>>> array (a,dtype=b.dtype)
array([(0,0), (1,0), (2,0), (3,0), (4,0), (5,0), (6,0), (7,0), (8,0),
(9,0)], dtype=cmplx_int64)
>>> 

But mixed mode arithmetic gives:

a+b
TypeError: function not supported for these types, and can't coerce safely
to supported types

I've been trying to understand this 'coerce' without much luck.  Any clues
what I need to do here?