[Python] array equality defaults to identity

[asfimport]

I'm not sure this is deliberate, but it doesn't look very desirable to me:

>>> pa.array([1,2,3], type=pa.int32()) == pa.array([1,2,3], type=pa.int32())
False

Reporter: Antoine Pitrou / @pitrou
Assignee: Wes McKinney / @wesm

PRs and other links:

• GitHub Pull Request #1729

_{Note: This issue was originally created as ARROW-2150. Please see the migration documentation for further details.}

[asfimport]

Uwe Korn / @xhochy:
No, this at the moment only lack of implementation. We have function in C++ that actually can take care of the comparisons (they are as simple as calling Array::Equal()) but they are probably not yet invoked in Python.

[asfimport]

Antoine Pitrou / @pitrou:
So the intended behaviour is to return a bool value, not an array of bools as with Numpy arrays?

[asfimport]

Wes McKinney / @wesm:
I think we should make this raise for now, and later perform vector comparison like NumPy (once we have a kernel available to do this)

[asfimport]

Antoine Pitrou / @pitrou:
A more general question: what is the ultimate aim for the Array API? Do we want to reproduce the Numpy API, or just provide a thin storage layer?

[asfimport]

Wes McKinney / @wesm:
For the time being, I believe it should be a thin storage layer, and that the user API for analytics and computations should evolve separately.

Once we have a more sizable collection of kernel functions to work with we should spend some time thinking about what the API should look like.

[asfimport]

Antoine Pitrou / @pitrou:

For the time being, I believe it should be a thin storage layer, and that the user API for analytics and computations should evolve separately.

Such as letting the user create a Dask array from pyarrow buffers?

[asfimport]

Wes McKinney / @wesm:
Issue resolved by pull request 1729
#1729