On Sun, Mar 23, 2014 at 8:30 PM, Sebastian Berg <sebastian@sipsolutions.net> wrote:
On So, 2014-03-23 at 07:26 -0600, Charles R Harris wrote:
>
>
>
> On Sun, Mar 23, 2014 at 6:56 AM, Ralf Gommers <ralf.gommers@gmail.com>
> wrote:
>
>
>
>         On Sun, Mar 23, 2014 at 3:07 AM, Charles R Harris
>         <charlesr.harris@gmail.com> wrote:
>                 Hi All,
>
>
>                 It is time to start looking forward to the 1.9.0
>                 release. Currently there are some 76 open PRs and they
>                 keep rolling in, which is good,
>
>
>         To make the PR list a bit more manageable, I would suggest to
>         start closing the ones which are not in a state to get merged
>         and haven't seen activity by the author for >3 months. And add
>         in the dev guide that this is normal policy and that authors
>         are free to reopen the PR when they continue working on it.
>
>
>
> I'd feel better about doing that if PR's were reviewed and dealt with
> on a regular basis, but we aren't quite there yet. That said, I'd like
> to keep the number down in the 30-40 range.
>
>
>
>
>                 but we need to decide on what is important for 1.9 and
>                 what can be put off to 1.10 because otherwise we will
>                 never finish. The datetime problems and some of the
>                 deprecations/futurewarnings that were present in 1.8
>                 need to be dealt with. The nanmedian stuff will make a
>                 nice addition to the nan functions. Apart from those,
>                 if you have a PR or fix that you think needs to be in
>                 1.9, please make it known.
>
>
>
>         The boolean subtract and ellipsis indexing deprecations
>         probably need reconsidering. I get 78 test errors right now
>         because of those if I test scipy master against numpy master.
>
>
>
>
>
> That's a lot of errors. Do you think they should be reverted
> permanently or just for 1.9?

Temporarily probably. Assuming they were a good idea to start with.
 
Good question. Just to note, I don't mind reverting/removing these. I
was somewhat aware that the double ellipsis caused a lot scipy failures,
but they seemed mostly in the tests with code like `arr[..., ...]` and I
didn't check if it might be more trouble then gain.

IIRC we had something like this before with the safe casting changes in 1.6.x. We could do the following:
1. fix the issues seen in scipy (and scikits etc.) now.
2. revert this change for 1.9.x so it doesn't cause issues with released versions.
3. re-introduce the deprecations in a year or so. In a year scipy will have 2 released versions with the fixes from (1).

Ralf