On Wed, 28 Mar 2007, Robert Kern wrote:
People have been giving you reasons, over and over again. You are simply refusing to listen to them.
Exploring whether the reasoning is adequate is not the same as refusing to listen. I do not presume my view is correct.
You have a use case for arrays being the iterates. You are presuming that the only argument that can beat that is another use case for matrix objects being the iterates. This is not true; there are other principles at work.
Put slightly differently: given the surprising passion of the attacks at the suggestion that perhaps iteration over a matrix might more consistently yield arrays, I presumed there must be *many* instances in which it was obviously desirable that such iteration should yield matrices. So I asked to see some. In the context of this discussion, I found the (lack of) responses very interesting. Even in your thoughtful response it proved irrelevant rather than important for iteration over matrices to yield matrices. I understand that some people claim that a general principle of consistency is involved. I have not been able to understand this particular design decision as a matter of consistency, and I have tried to say why. However I am just a user (and supporter) of numpy, and as indicated in other posts, I make no pretense of deep insight into the design decisions. In this case, I simply wanted to open a discussion of a design decision, not win an "argument". Anyway, I understand that I am being perceived as bull-headed here, so I'll let this go. Thanks for your attempt to help me see the virtues of the current design. Cheers, Alan Isaac