On Apr 21, 2008, at 12:32 PM, David Roundy wrote: > On Mon, Apr 21, 2008 at 11:07:54AM -0600, zooko wrote: >> >> Ideally we will soon no longer rely on this channel as strongly, >> since we will have robots to do that job, thus taking some of the >> load off of human users. > > No, the problems usually occur on rare systems or unusual > configurations, > and those are the systems on which we probably will never have > buildbots.
This objection makes sense -- we might want to support a certain system or configuration but not have available to us a buildslave running that configuration. > And it also doesn't make sense to use buildbots for those > situations, since > we don't want to spend the effort to support a configuration that > has no > users. Although it would certainly be nice to know more quickly of > simple > regressions that cause the build to break on a certain compiler > (e.g. ghc > 6.4). This part seems to contradict the above -- if we want to support a system, and if we are capable of running a buildslave on that system, then we should. If we do not wish to run a buildslave on that system even though we can, then apparently we aren't too serious about "supporting" that system. Let me put it this way: if I'm investigating using an open source project, and I want a predictor of how likely it is that the project will work with my system or my configuration, then whether the project has a live buildslave for my system is a better predictor than whether the project's web page says that my system is "supported". ;-) In any case, the issue is moot for now, until we have buildslaves with all-green tests at http://allmydata.org/buildbot-darcs which we can point to as examples of known-to-be-working platforms. Regards, Zooko _______________________________________________ darcs-users mailing list [email protected] http://lists.osuosl.org/mailman/listinfo/darcs-users
