[Numpy-discussion] Latest Array-Interface PEP
Travis Oliphant
oliphant at ee.byu.edu
Thu Jan 4 18:19:14 EST 2007
>> PyArrayView_FromObject(obj)
>>
>
> yes, that's what I'm looking for -- please do add that to the C-API
>
>
>>> By the way,, how compatible is this with the existing buffer protocol?
>>>
>>>
>> It's basically orthogonal. In other-words, if you defined the array
>> view protocol you would not need the buffer protocol at all. But you
>> could easily define both.
>>
>
> OK, so if one of these were passed into something expecting the buffer
> protocol, then it wouldn't work, but you could make an object conform to
> both protocols at once -- like numpy does now, I suppose -- very nice.
>
> Another question -- is this new approach in response to feedback from
> Guido and or other Python devs? This sure seems like a good way to go --
> though it seems from the last discussion I followed at python-dev, most
> of the devs just didn't get how useful this would be!
>
The new approach is a response to the devs and to Sasha who had some
relevant comments. Yes, I agree that many devs don't "get" how useful
it would be because they have not written scientific or
graphics-intensive applications. However, Guido is the one who
encouraged me at SciPy 2006 to push this and so I think he is generally
favorable to the idea. The Python devs will definitely push back. The
strongest opposition seems to be from people that don't 'get' it and so
don't want "dead interfaces" in Python.
They would need to be convinced of how often such an interface would
actually get used. I've tried to do that in the rationale, but many
people actually posting to python-dev to support the basic idea (you
don't have to support the specific implementation --- most are going to
be uncomfortable knowing enough to make a stand). However, there is a
great need for people to stand up and say: "We need something like this
in Python..."
-Travis
More information about the NumPy-Discussion
mailing list