Continuous Integration

[jreic]

Can we easily implement a CI for our code? I know it's easier in CERN gitlab than github, but still worth thinking about.

[jreic]

Copied from PR #10:

Implemented CI to build our code (using cvmfs and cmssw via a Docker image) after each push to the master branch.

We in principle can/should extend this to do the same for other important branches (could consider all branches, to do the same for long-term development branches), and for different releases than what I implemented it for (9_4_6_patch1).

We can also setup the CI to do other things, e.g. to download an input root file and do sanity checks of ntupling or making histograms (in other words: unit tests), if we wanted to. This would be useful to make sure that implementing something new that is intended to affect one part of the analysis doesn't inadvertently affect other areas without our knowledge. Anyway, for now: catching C++ compilation errors already will be useful.

[jreic]

Also: we could have the CI format code upon pushes too, which might be useful...

[jreic]

Other potential improvements: I currently hardcode the paths for performing the git clone in initial_setup.sh and for downloading the setup script in docker-image.yml. It would be good to make this flexible (as well as making the branch to checkout and run the CI for flexible for both).

[jreic]

Merged #10, but we should consider these other ideas for improvement too.

[jreic]

Should revisit this, since we have mainly been using branches for a while now! And unit tests would be nice.

[jreic]

I spent a little while looking into it but CVMFS didn't seem to be getting mounted successfully, so cmsset_default.sh failed. The CI seemed to think that CVMFS was mounted, but a simple ls of its directory showed nothing, so it must have been not properly connecting to CERN. Maybe even a CVMFS version issue, since the docker image is quite old. Not worth the trouble right now. Would probably be easier in a repo hosted at gitlab.cern.ch...