[Numpy-discussion] Should arr.diagonal() return a copy or a view? (1.7 compatibility issue)

Frédéric Bastien nouiz at nouiz.org
Wed May 23 13:40:13 EDT 2012

On Wed, May 23, 2012 at 1:29 PM, Travis Oliphant <travis at continuum.io> wrote:
> Then are you suggesting that we need to back out the changes to the casting
> rules as well, because this will also cause code to stop working.   This is
> part of my point.   We are not being consistently cautious.

The casting change have already been release, or is there other change
that break the user interface in 1.7? I probably won't change again
the interface that was released except if there is good reason. I
don't remember the detail enough to suggest any direction on this.

I remember the past change on dtype that broke Theano. I consider this
an "error", but we all do them from time to time. I don't blame
anybody for this. We just changed Theano after understanding the numpy

I prefer that we try to be consistently cautious then to tell we
didn't always do it in the past so we won't try now. Also, I would
find better to voluntarily break this "consistently cautious"
guideline if it is well documented and advertised if needed then to
don't try to be "consistently cautious".


More information about the NumPy-Discussion mailing list