Travis-CI is fantastic for testing software (continuous integration) during open-source development. Less widely known, I think is how useful it can be for deployment.
With travis-ci, you have full access (passwordless
sudo) access to a fresh
Ubuntu virtual machine that runs automatically on every commit to your github
repository. Sure, you can use this to run your tests, but you can do a lot more
Generate this website!
This site is built with Pelican, and deployed on Amazon S3. I use travis-ci to automate the deployment -- every commit to github triggers travis to rebuild the HTML/CSS/JS content and push it to S3. This completely automates the deployment workflow. To publish a new post, I just push to github. It's Heroku-like convenience for (basically) free. You can check out the source code for this site to see how it works.
In MDTraj and OpenMM libraries, for example, our documentation is built directly from the source code using the Sphinx and Doxygen tools, which output HTML and PDF documents. Our travis-ci virtual machines, after running the tests, rebuild the documentation and deploy a versioned copy directly to Amazon S3 buckets which are configured as publicly accessible websites. No more worrying about the docs getting out of sync with the code -- they're updated automatically.
We also use travis, and its Windows cousin, Appveyor CI to build binaries
of released and development snapshots of MDTraj. We use the python conda package
manager to build binaries for a large matrix of different versions of the
dependencies, and then push the binaries directly to the binstar index
so that users can
conda install them without needing to compile.
The one caveat here is that, for MDTraj, we still target the (very ancient) RHEL5, and to avoid GLIBC incompatibles, we need to build our linux release binaries on a sufficiently old distro. The infrastructure running at Travis-CI is too new, unfortunately. But we still do use for distributing binaries of the development snapshot, which are handy for downstream integration testing downstream.