[SciPy-dev] Scipy workflow (and not tools).

David Cournapeau cournape@gmail....
Wed Feb 25 11:02:43 CST 2009


On Thu, Feb 26, 2009 at 1:49 AM, Travis E. Oliphant
<oliphant@enthought.com> wrote:

>
> I do see a real need to fix the SVN-Trac workflow bottleneck as well as
> anything that helps the release process.    It's actually at the release
> process where I would institute any formal review process.   I'm also in
> favor of having a regular (i.e. every 3-6 months) release process.   The
> difficulty there again is man-power.

One thing which may help here is to have a turn-around for the release
manager: a different person every time. This person would have the
last world of what goes in/what does not, with almost strictly
enforced deadlines. In particular, we should really enforce code
freeze - although I can understand the point that reviews may make
things harder, I don't think it is possible at all to make good
release without enforcing very strict timelines. There has to be no
new code for some time before the release, time which is more than
just one day or two. C/Fortran code would be the first to be freezed,
then python, then docstring. The exact time can be tweaked after
experiments, of course. But if we get this right, I believe that
having freeze periods can make the time from patch to inclusion
actually faster.

Having a different person means it is not always the same person,
obviously, and it may also keep people "honest", in the sense that a
release manager will also be a coder later under a different release
manager.

cheers,

David


More information about the Scipy-dev mailing list