Re: librump hits again...

[Antti Kantee <pooka%netbsd.org@localhost>]

On Fri May 14 2010 at 15:56:23 +0200, Michal Suchanek wrote:
> Or better yet, have the commits go to a queue and only accept ones
> that don't break builds ;-)

This is an interesting proposal, although I'm not sure it's very practical
or even that it works with cvs.  Conflict resolution is probably quite
hard, since you have two states in the repository (queued and passed).
Which do you commit to?  What happens if a queued commit does not pass
and a later commit conflicts with it?  Force the people who's commits
were correct in the first place to recommit?

IMHO the issue is better solved at the social layer, since sometimes
the brokenness appears only in UPDATE builds, or non-UPDATE builds or
some weird mk.conf knobs or whatever, and neither developers nor an
automatic tester can't be held fully responsible for not spotting the
corner cases.

But I'm sure tech-repository would love to talk more about this ;)