Re: Buildbot

[John W. Eaton]

Moving this back to the maintainers list...

On 09/02/2016 01:21 PM, Nicholas Jankowski wrote:

>  Instead the idea
> is to let users run an automated build process for the different
> distributions, and report back whether that worked or not, but the build
> itself would stay local to the user. So if it works, that person would
> have a nightly build available for their platform (assuming make dist or
> equivalent is run).

Having the build available is a possible by product.  But my reason for 
setting this up is so that we are continually building on all the 
systems of interest so that bugs are noticed earlier.  In the past, 
someone might make a change, test it on one system but not notice for 
weeks (or longer) that the change breaks some other type of system or 
configuration (Windows, OS-X, 64-bit indexing, some Octave Forge 
package, or a build with some missing dependency).  By the time a 
failure is noticed, it is time consuming (even with bisecting!) to try 
to find the changeset that caused the problem and fix it.

I've already caught one problem like this and we've only had the 
buildbot thing running for a few days.

My goal is to set up builds on many more types of systems (OS X, RHEL, 
SuSE, Ubuntu, whatever).  I still need help with that.  Or money to buy 
more systems so I can do the testing myself.  But it would be better if 
we could distribute these builds to many different systems that other 
people manage.

Currently "testing" means running the test suite, and only for Octave. 
Eventually we should be building Octave Forge packages and running all 
the tests for them as well.

I agree that it would also be useful to provide access the automated 
builds so people can test things that aren't covered by the test suite 
(the GUI, primarily).  I just don't have the resources to do that at the 
moment.  Maybe we can find a place to do that job.

jwe