[Biopython-dev] Travis-CI to complement BuildBot

[Peter Cock]

Dear Biopythoneers,

We now have more automated testing being done, thanks
to Travis-CI (Travis Continuous Integration):
http://travis-ci.org/#!/biopython/biopython

What this means is that whenever a commit is pushed to
https://github.com/biopython/biopython/ it should trigger
Travis to run the unit tests on their farm of Debian 64bit
virtual machines.

Travis-CI http://travis-ci.org/#!/biopython/biopython
Pros: Testing every time a commit is pushed, should
be minimal maintenance effort for us.
Cons: Only tests on 64bit Debian Linux, currently
doesn't include Jython, and only an older PyPy

BuildBot http://testing.open-bio.org/biopython/grid
Pros: Testing multiple platforms (currently Windows,
Mac, and both 32bit and 64bit Linux) including Jython.
Cons: Adding new platforms (buildslaves) takes quite
a lot of time to setup, and since they are all volunteer
machines we can only really test on a nightly basis.

i.e. I see Travis-CI and our BuildBot as complementary.

This was prompted by the success BioRuby was having
with Travis, http://travis-ci.org/#!/bioruby/bioruby and
http://lists.open-bio.org/pipermail/bioruby/2012-May/002247.html

Thank you to the BioRuby team for their assistance last
month in getting this enabled for Biopython to. I quickly
got this working on my personal repository and worked
out a reasonable configuration file balancing dependencies
and run time - numpy is a problem though:
https://github.com/travis-ci/travis-cookbooks/issues/48

This week I had a fresh attempt at hooking this up to
the official Biopython repository:

(1) Create an account on Travis via your Github account
(2) Adding a .travis.yml file to the repository
(2) Enable testing of this repository via your Travis profile,
easy for personal repositories, harder if the repository is
for an "Organization" on Github (like Biopython). See:
http://about.travis-ci.org/docs/user/how-to-setup-and-trigger-the-hook-manually/

This was done using my personal Github account (peterjc),
but in principle any of the developers with admin rights to
the Github biopython account could do this. It didn't take
effect immediately, but overnight the backlog of tests I'd
tried to trigger happened - thus the first dozen builds that
Travis tests include a fair number of repeats. Oops.

Are there any views on if and where we should ask Travis
to send email reports? e.g. Just the Biopython Developers
with commit rights? We should be able to send them to
the biopython-dev mailing list but I fear too many emails
would annoy people.

Regards,

Peter