Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost> writes: > joerg@ wrote: > >> > "Please don't commit untested and broken fix." >> > "Please file a PR instead if you can't test it." >> >> I agree with Christos that leaving the build broken is worse. The >> alternative band aids are much more involved, too. That shouldn't stop >> the second point from that list. > > If you claim leaving the build broken is worse than commiting untested code, > you should ask to update our commit guideline first: > > http://www.netbsd.org/developers/commit-guidelines.html Perhaps we should. The problems are: A broken build is evidence that the prior commits were not tested, and thus need fixing. Objecting to fixing the build without testing should be a far lower priority than objecting to commits that break the build. A broken build, or a build that succeeds but fails to work is a serious problem because it prevents bisecting to find bugs. I've seen this in current/i386 pretty often, less so recently. During a ~week that the build was broken many commits happened, and some of those were trouble, and it was a mess to sort out. So arguably no commits should be allowed at all during a time when the build is broken or there are sudden significant new test failures, other than fixing the build.
Attachment:
pgpy_8VZMnyZ6.pgp
Description: PGP signature