Subject: Re: [boost] Improving review process
From: Dave Abrahams (dave_at_[hidden])
Date: 2011-01-14 09:31:49
At Fri, 14 Jan 2011 07:13:26 -0700,
Jeff Garland wrote:
> >> Note that I'm not mentioning the name of the tool just yet because I
> >> don't want to violate our 'advertising policies' on the list.
> > I think you're being overly cautious. Atlassian?
> Fair enough:
I've heard very good things about that product.
> >> If there's interest, I can make initial contact with company and
> >> get/post the details on how it would work. I was planning to
> >> propose this at BoostCon, but now that it's come up we should start
> >> the process now if folks agree.
> > Using a code review tool is an awesome idea. Many reviews are not
> > attached to code, but you can put review comments in documentation
> > just as well.
> Documentation is a bit harder unless you are annotating the document
> source directly
Yes, that's what I was suggesting.
> -- that is, I haven't see a mode in the tool to
> annotate against 'rendered html'.
That *would* be a nice feature, wouldn't it!? I don't think I've ever
seen such a tool. Wait, that's not true. See
http://djangobook.com/en/2.0/chapter01/ for an example.
> > A couple of things to consider:
> > 1. We'd still need a place for overall assessments that don't pertain
> > to specific details.
> There's an 'overall comments' section at the top of each review for
> these kinds of comments.
Yeaaaaaah... do you think putting that kind of commentary in the same
place as a code review would work for our review process, though?
Maybe it would, but I have a hard time envisioning it.
> > 2. I know this is a bold predicition, but I think we will be
> > transitioning to GitHub. It has an enormous momentum in the open
> > source world, is responsive, and will continue to make a lot more
> > sense as Boost is modularized. GitHub already supports code
> > review. I think I'd rather go with a tool that requires absolutely
> > no sysadmin on our part, is a known quantity to many already, etc.
> I haven't used the github review capabilities here -- so we'd have to
> evaluate what works best. As for the admin -- it's truly minimal --
> basically the same as giving someone sandbox access today --
> registering an email address so that comment discussions are tracked,
> etc. And the author has to upload code to the tool -- but a simple
> paragraph should be about enough to explain it.
I mean someone has to install the tool, administer the system on which
it runs, ensure that there's always enough CPU power/bandwidth, manage
upgrades, etc. Also I don't love the need to upload code. Having the
tool built into a code repo/sharing system removes steps and *should*
make things run more smoothly (no separate login, for example).
One thing I do like very much about centering the process on a code
review tool is that people can poke through making comments, etc., and
then after they've taken a good look, consider voting and adding an
overall write-up. That is, you can get into it incrementally. That's
a lot harder with the current system.
-- Dave Abrahams BoostPro Computing http://www.boostpro.com
Boost list run by bdawes at acm.org, gregod at cs.rpi.edu, cpdaniel at pacbell.net, john at johnmaddock.co.uk