Cruise Control to mitigate failure in make bootstrap?

Cruise Control to mitigate failure in make bootstrap?

[Alan Mackenzie]

Hi, Emacs!

We seem to have continual problems with HEAD not bootstrapping.  This is
frustrating for quite a lot of people.  I suspect it is a major drain on
our productivity and general happiness.

One solution would be to use Cruise Control (see
<http://cruisecontrol.sourceforge.net/>).  It is free software (BSD
license).

Cruise Control triggers a build whenever changes are commited.  Should
the build fail, it notifies all those who have just committed of the
failure.

I have used this product in one of my day jobs, and it works well.

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: Cruise Control to mitigate failure in make bootstrap?

[Jason Rumney]

Alan Mackenzie wrote:
> I have used this product in one of my day jobs, and it works well.
>   

I suspect it works well in a day job environment, when you are working 
for the same 8 hours a day as the other developers on the project.  I 
suspect it works less well for Free Software where volunteers might have 
a haf hour to spare so they fix a bug, don't have time to do a full 
bootstrap before checking in, then might not see any mail generated 
because they are offline for the weekend.

Re: Cruise Control to mitigate failure in make bootstrap?

[aaditya sood]

Jason Rumney <jasonr <at> gnu.org> writes:

> 
> Alan Mackenzie wrote:
> > I have used this product in one of my day jobs, and it works well.
> >   
> 
> I suspect it works well in a day job environment, when you are working 
> for the same 8 hours a day as the other developers on the project.  I 
> suspect it works less well for Free Software where volunteers might have 
> a haf hour to spare so they fix a bug, don't have time to do a full 
> bootstrap before checking in, then might not see any mail generated 
> because they are offline for the weekend.

I agree. However it would let everyone else know that the trunk is broken, and
someone can take appropriate action. It also allows a trail of known good
checkpoints.

At some point if unit tests are added to emacs they could be run as part of
build, thus giving even more automatic test coverage.

regards
--
/A

Re: Cruise Control to mitigate failure in make bootstrap?

[Alan Mackenzie]

On Sat, Apr 12, 2008 at 03:02:57PM +0100, Jason Rumney wrote:
> Alan Mackenzie wrote:
> >I have used this product in one of my day jobs, and it works well.

> I suspect it works well in a day job environment, when you are working
> for the same 8 hours a day as the other developers on the project.  I
> suspect it works less well for Free Software where volunteers might
> have a haf hour to spare so they fix a bug, don't have time to do a
> full bootstrap before checking in, then might not see any mail
> generated because they are offline for the weekend.

Well, it wouldn't be worse than what we currently have.

Having seen the program (which is free software, after all) work so well,
I think it would work just as well for the Emacs project.  How long would
a make bootstrap take on a suitable server?  10 minutes?  Possibly 2
minutes?  That's not terribly long to wait to check that a change just
committed hasn't fouled things up.

I've just a look at my emacs-devel archive, and since mid January 2008,
there are 26 threads with subject line "bootstrap fails".  That's an
awful lot of inconvenience to an awful lot of people.

-- 
Alan Mackenzie (Nuremberg, Germany).

Re: Cruise Control to mitigate failure in make bootstrap?

[Bob Rogers]

   From: Jason Rumney <jasonr@gnu.org>
   Date: Sat, 12 Apr 2008 15:02:57 +0100

   Alan Mackenzie wrote:
   > I have used this product in one of my day jobs, and it works well.

   I suspect it works well in a day job environment, when you are working 
   for the same 8 hours a day as the other developers on the project.  I 
   suspect it works less well for Free Software where volunteers might have 
   a half hour to spare so they fix a bug, don't have time to do a full 
   bootstrap before checking in, then might not see any mail generated 
   because they are offline for the weekend.

Another case is where a commit breaks on a platform the committer does
not have; the smoke system alerts potential fixers in a timely way.
This seems more likely on a Free project than a proprietary one.

   FWIW, the Perl community calls this "smoke testing" [1], and has been
using it successfully for a number of subprojects [2].  I can't claim
great familiarity with it, but I think the value increases with the
coverage of the automated test suite.

					-- Bob Rogers
					   http://rgrjr.dyndns.org/

[1] http://www.perlfoundation.org/perl5/index.cgi?smoke_testing

[2] E.g. http://smoke.parrotcode.org/smoke/

Re: Cruise Control to mitigate failure in make bootstrap?

[Stefan Monnier]

> Cruise Control triggers a build whenever changes are commited.  Should
> the build fail, it notifies all those who have just committed of the
> failure.

I don't know about Cruise Control, but the way I see it working in our
case (e.g. with Romain's buildbot) is:

   bzr pull sftp://foo/trunk
   while buildot fails; do
      bzr uncommit
   done
   bzr push sftp://foo/stable-trunk

This way people how want a buildable trunk code can checkout the
stable-trunk branch.  In most cases this would be virtually identical to
the trunk.


        Stefan