Strongly agree. Master was very painful to work with last week. On Mon, Feb 18, 2013 at 10:43 AM, David Nalley <da...@gnsa.us> wrote:
> Hi folks, > > I know folks have said this elsewhere, but I'd like to reiterate it. > > I am somewhat frustrated with our reaction to broken builds. In > general it seems we don't care, and this makes it more difficult to > fix problems. Jenkins reporting a broken build (be it a broken run of > RAT, failure to compile, failure of a unit test, building docs, etc.) > should be our Andon cord [1]. We should all stop commits that aren't > fixing the broken build. To illustrate why this is a problem, > > RAT failures started occurring recently, this keeps us from testing > whether CloudStack builds, because each build is conditioned on the > successful completion of the test before it. That in turn keeps > apidocs from building, which keeps marvin from building, which keeps > documentation from building. We essentially are blind until it gets > fixed. > > That means, that like TPS, when the andon cord gets pulled we all need > to focus on the problem, and not continuing our own work and ignoring > the problem and potentially contributing to making the solution more > painful. > > My requests: > > If you see that the branch you are working on is broken - please don't > commit to it, unless your commit is going to fix it. > If you see that the branch you are working on is broken - please help > fix it. (This should become priority #1, for everyone) > Please don't make a commit the last thing you do before going offline > - make sure that your commit isn't breaking any of the tests before > you leave. If your commit breaks something and you've gone offline for > 16 hours, you've made life painful for others, so make sure there is > some buffer between your last commit and you going offline so you can > see and remedy any problems that arise. > > And just for the record - as of this moment, 4.1 and master are broken > at various stages. > > > --David > > [1] http://en.wikipedia.org/wiki/Andon_(manufacturing) >
