Aryeh Gregor wrote:
On Sun, Oct 17, 2010 at 6:58 PM, MZMcBride
<z(a)mzmcbride.com> wrote:
To put it another way, if the code backlog were
eliminated today, I think
the exact same frustrations and annoyances would exist among the developer
community because the code would sit reviewed, but un-deployed. When we say
"the biggest issue is a huge code review backlog," it's misleading. The
reality is that the biggest issue is a huge code review _and deployment_
backlog.
I don't see any reason why deployment would lag far behind review.
Review takes a lot of time, deployment is quick. We'd just need a lot
of people on hand to do damage control when stuff breaks unexpectedly
after such a huge deployment.
You don't see any reason? Simply, because it does. How did we get into the
current situation?
Broadly, I
don't see any reason why it's beneficial to engage in Great
Reveal-style code updates to the site. There are thousands of
uncontroversial revisions already reviewed that could be deployed today.
I don't think it would be wise at all to try deploying something less
than trunk. Deploying trunk means we can commit any fixes to trunk
and only trunk. Branching some earlier point and deploying that would
mean we have to maintain the branch. Plus, we'd have to have several
large deployments instead of one giant one. Better to have one giant
deployment so that there's at most one point of massive user
inconvenience, and then stick to only small ones from then on.
One idea floated today in #mediawiki was to make a new branch up to the
revision before ResourceLoader was added (grep "branch right before
that").[1] This would allow a very sizeable number of revisions to be in a
branch (from April to August, roughly) but also allow deployment and code
updates to move forward sooner. Sometimes the easiest way to eat an elephant
isn't to try in one bite. Any thoughts on creating a branch sooner than when
ResourceLoader is ready?
MZMcBride
[1]
http://toolserver.org/~mwbot/logs/%23mediawiki/20101018.txt