Re: build test farm [was Re: [Chicken-users] Cmake broken again: paths are not quoted]

[Brandon J. Van Every]

Toby Butzon wrote:

On Thu, Jul 20, 2006 at 04:00:48PM -0400, John Cowan wrote:
  

Sourceforge.net does provide a "compile farm" that provides the ability
to build on multiple platforms, though Felix would have to move Chicken
there to take advantage of it.  It doesn't do Windows, though.
    

Hmm... maybe it would be possible to construct our own, with volunteered
resources. The normal process would be to pull from darcs, try the build,
and post apparent success or failure (maybe to a page on the swiki?). If
it fails, a log could be posted, too. (A similar alternate process would
exist for checking release tarballs, but this would run only when a new
tarball has been released, so as not to waste cycles rebuilding something
that's already been tested.) I would envision this being a chicken
script ("compilefarm.egg"?) and it'd be triggered by cron/task
scheduler, on the volunteer's terms (so volunteers are in full control
of how much it encroaches upon their system).
  

I think it's worth asking William Hoffman of Kitware whether they can provide CMake + CTest + Dashboard testing resources.  If they don't provide the testing machines themselves, I'm sure they can at least give us pointers on how to set up such a thing, and what the consequences / impacts are.  I do not believe CMake currently supports Darcs, that's an issue.  But perhaps we can scare someone up to deal with that issue, for CMake's benefit.

I don't think the program to do it would be all that complicated; the
question is, could we find volunteers to give up some cpu cycles on
various platforms? (I for one have at least a Windows box and a Linux
2.6 box I leave always running that would be offered.)
  

It would be worth knowing if the Dashboard can function in a distributed or peer-to-peer manner.

The product would be a page with a table on it, showing each volunteered
machine, architecture and OS, build schedule, when the last build ran, and
success/failure. (And, if failure, a link to the log of what happened.)
  

I feel that this sort of reportage reinvents the Dashboard, and thus is a waste of open source development resources.  Unless someone out there is such a Chicken web guru that they deliver proof-of-concept in a few days work.

Maybe this would also provide some more incentive for a test suite, and
at a minimum, we could start with one for (use srfi-1) and whatever other
problems might be diagnosed after "successfully" building.

I'd like to hear what others think about this.
  

I think test suites need no additional motive.  Even without a nightly build, test suites are quite valuable to development.  Bigloo has an excellent test suite, for instance.  The question is, will someone do the work of providing and integrating a test suite.


Cheers,
Brandon Van Every