[Ksummit-2013-discuss] Keeping the barbarians at the gate

[John W. Linville]

On Thu, Jul 18, 2013 at 07:16:34PM -0400, Paul Gortmaker wrote:

> The above kind of crap may be the most reprehensible, but it probably
> is the easiest to detect.  The other, more costly kind of things we
> want to detect and keep out of tree is the stuff that presents well at
> 1st glance to casual reviewers, but won't hold up to scrutiny by the
> experts in the area.  It would probably be unfair to call these crap,
> because usually at this level, people have invested considerable
> time/effort in creating the change, and have a semi plausible reason
> for why the change was made, even if it turns out to be a misguided
> one.
> 
> The latter filter responsibility is harder to delegate/distribute, so
> maybe the above suggests a tiered level of review to assist the
> maintainers?  The experts sit at the gate, but there is also a moat
> full of alligators, and a pack of wild dogs on the drawbridge as 1st
> pass filters?  How to get more dogs and alligators reviewing patches?

Tiered review really is the key here.  I've been lucky in the wireless
space to have a good group of contributors that are usually willing
to do this sort of review.  It certainly makes a huge difference both
for quality and for my own sanity.

I'm not entirely sure how to coax more of this out of folks.  I suspect
that many maintainers have trouble "holding back", letting others
develop an attitude of responsibility towards the overall subsystem
rather than just their own driver and such.  In my view, sometimes a
maintainer has to balance between a willful intransigence towards a
patch series on one hand, and on the other saying something to create
the impression that you will merge something crazy.  Somewhere in
between, responsible folks will act to keep you out of trouble! :-)

Generally speaking, I find new drivers to be the most difficult to
get widely reviewed.  Fortunately, a new driver is usually limited
to causing problems for its users.  New cross-driver features tend to
get broader review since they directly effect multiple driver teams.
Maybe there is a "bell curve" of sorts, with fewer reviewers both at
the driver end (where fewer people care) and at the core kernel end
(where fewer people understand the nuances)?  In the latter case,
maybe there is some way to "push" more discussion to wider (but still
appropriate) audiences?  Some form of "why you care about this" email?
A series of LWN articles?  Or...?  Dunno...

Well, hopefully there is something useful in the ramblings above...

John
-- 
John W. Linville		Someday the world will need a hero, and you
linville at tuxdriver.com			might be all we have.  Be ready.