__index__ method

[asmeurer]

I just noticed that the __index__ method is missing from the array object specification. See https://www.python.org/dev/peps/pep-0357/.

__index__ should return int(self) if the input is a 0-D array with integer dtype, and raise TypeError otherwise (similar to __int__ except it doesn't work on floating-point arrays). This will allow 0-D integer arrays to be used as indices. While this is not strictly necessary for an array type to support itself as an index, it does mean that things like indexing Python lists with 0-D arrays will work correctly, and it's generally a good Python practice to do this for integer-like types.

[rgommers]

I just did a quick check that NumPy, PyTorch, JAX and TensorFlow all support this behavior. So seems reasonable. Size-1 lists don't support it, so it's not obvious that it should work. But may as well, since it's common practice.