[Numpy-discussion] Bug? NumPy types circumvent some overloaded methods in ndarray subclasses

[Wes McKinney]

This strikes me as a bug-- haven't checked NumPy 1.6 yet but this
happens in 1.5.1. Here's a toy example:

class MyNdarray(np.ndarray):

    def __new__(cls, data):
        subarr = np.array(data, dtype=np.float64).view(cls)
        return subarr

    def __radd__(self, other):
        print 'hello'
        return (self.view(np.ndarray) + other).view(MyNdarray)

    def __add__(self, other):
        print 'hello'
        return (self.view(np.ndarray) + other).view(MyNdarray)

In [25]: foo = MyNdarray(np.arange(10.))

In [26]: foo + 3
hello
Out[26]: MyNdarray([  3.,   4.,   5.,   6.,   7.,   8.,   9.,  10.,  11.,  12.])

In [27]: foo + np.float64(3)
hello
Out[27]: MyNdarray([  3.,   4.,   5.,   6.,   7.,   8.,   9.,  10.,  11.,  12.])

In [28]: 3 + foo
hello
Out[28]: MyNdarray([  3.,   4.,   5.,   6.,   7.,   8.,   9.,  10.,  11.,  12.])

In [29]: np.float64(3) + foo
Out[29]: MyNdarray([  3.,   4.,   5.,   6.,   7.,   8.,   9.,  10.,  11.,  12.])

I'm not sure if there's an easy workaround here. I thought maybe it
was an array priority issue but after some tinkering it doesn't look
like it. Any ideas?

- Wes