[Numpy-discussion] let's use patch review
Thu May 15 22:39:36 CDT 2008
2008/5/15 Francesc Alted <firstname.lastname@example.org>:
> I don't need to say that this procedure was not used for small or
> trivial changes (that were fixed directly), but only when the issue was
> important enough to deserve the attention of the mate.
I think here's the rub: when I hear "patch review system" it sounds to
me like an obstacle course for getting code into the software. Maybe
it's justified, but I think at the moment there are many many things
that are just awaiting a little bit of attention from someone who
knows the code. A patch review system overapplied would multiply that
number by rather a lot.
How about a purely-optional patch review system? I've submitted
patches I wanted reviewed before they went in the trunk. As it was, I
didn't have SVN access, so I just posted them to trac or emailed them
to somebody, who then pondered and committed them. But a patch review
system - provided people were promptly reviewing patches - would have
fit the bill nicely.
How frequently does numpy receive patches that warrant review? The
zillion little doc fixes don't, even moderate-sized patches from
experienced developers probably don't warrant review. So in the last
while, what are changes that needed review, and what happened to them?
* financial code - email to the list, discussion, eventual inclusion
* matrix change - bug filed, substantial discussion, confusion about
consensus committed, further dicussion, consensus committed, users
complain, patch backed out and issue placed on hold
* MA change - developed independently, discussed on mailing list, committed
* histogram change - filed as bug, discussed on mailing list, committed
* median change - discussed on mailing list, committed
* .npy file format - discussed and implemented at a sprint, committed
Did I miss any major ones? Of course, svn log will give you a list of
minor fixes in the last few months.
It seems to me like the review process at the moment is just "discuss
it on the mailing list". Tools to facilitate that would be valuable;
it would be handy to be able to point to a particular version of the
code somewhere on the Web (rather than just in patches attached to
email), for example.
More information about the Numpy-discussion