[Numpy-discussion] let's use patch review

[Stéfan van der Walt]

2008/5/16 Anne Archibald <peridot.faceted@gmail.com>:
> 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.

Those moderately-sized patches are the ones that need review,
especially.  Review provides useful information on a couple of levels:

a) Motivation -- why do we want/need this patch
b) Functionality -- does it do what the developer intended it to
c) Implementation -- is it written according to current best practices

Level (a) is normally discussed on the mailing list, if needed.  Level
(b) is covered by unit tests, *if* those were written.  Then, level
(c) is where the main advantage lies: we can learn from one another
how to develop better code.

I am somewhat split in two on this one.  I love the idea of patch
review; it undoubtedly raises the quality of the codebase.  That said,
it comes at a cost in developer time, and I'm not sure we have that
luxury (we don't have a Michael Abshoff, unfortunately).  Making it
optional might be a good compromise, although the person who wrote a
patch isn't the best one to judge whether it should be reviewed (of
course, we all think our code is good!).

Regards
Stéfan