On Sat, Nov 12, 2011, Scott Brozell wrote:
>
> If i understand this then timely testing of the master branch
> would not be done nightly (or sooner) on commits but only after e.g. days.
No, that was not my intention. Testing of the master branch would
continue as at present. But breaking something would not be viewed as
such a terrible thing as it is now by some people. I did write that
updating master would imply a commitment to follow the cruise-control tests,
and to promptly fix any problems that it reveals.
For example, cruise control tests 5 variants of Amber and 4 of AmberTools.
My fear is that if we (implicitly) say "you have to run all these tests before
committing to master" we'll miss a lot of good commits, and people will
retreat to their private branches and not share code.
Of course, when we get a flurry of last-minute commits, as is typical late
in the development cycle, we may need to ask people to be more careful, or use
some other model.
....dac
[Aside: cruise control has real limits if something fails. I tried to see
why the parallel make was failing, but the "build log" on cruise control is
not human-readable. So, when a red light appears on the dashboard, it may be
best just to try to reproduce the problem on your local machine.]
_______________________________________________
AMBER-Developers mailing list
AMBER-Developers.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber-developers
Received on Sat Nov 12 2011 - 07:00:03 PST
