How (if at all) can binary operator methods of numpy arrays be overridden by the right operand?

问题

I have a custom class which defines custom, commutative, behaviour for the binary + operator:

class foo(object):
    def __add__(self, other):
        ...
    __radd__ = __add__

It is designed to deal with scalars and with numpy arrays in its own way---a way that is intentionally different from the way numpy adds arrays.

If I start with the following:

f = foo()
a = numpy.array([1, 2, 3])
c = 4

then all is well if I do f + a, f + c or c + f, in the sense that I get the custom behaviour I want in all cases. However a + f gets me a different result (and in some situations an exception) because the numpy array method a.__add__ takes over. I understand that __radd__ is "only called if the left operand does not support the corresponding operation" per the Python docs. I also read somewhere that this is overridden if a.__add__ returns NotImplemented.

So is there any way of telling numpy "hey, I got this, you should return NotImplemented when asked to add a foo"? In Matlab this would be accomplished by superiorto. Is there any equivalent in numpy? (NB: I don't want to have to subclass numpy.ndarray---I want to be able to deal with numpy arrays as they come.)

[In an ideal world I would like to be able to cope with native Python objects in the same way, such that [1,2,3] + foo() gave the same result as foo() + [1,2,3]---but I'm guessing it's got to be impossible to disable list.__add__ selectively based on the class of the right operand. Right?]

来源：https://stackoverflow.com/questions/47600049/how-if-at-all-can-binary-operator-methods-of-numpy-arrays-be-overridden-by-the