Continuous Integration¶
To contribute to the code base of HydPy, you need a copy of its sources (a separate fork, see section Version control). The existence of multiple copies of the same code bears the danger of integration problems due to incompatible changes. An obvious problem is when two developers introduce different changes to the same code lines. However, even untuned changes in separate code sections can break functionalities. As a means to reduce such risks, different working copies should be synchronised “continuously”. Taking the code changes of colleagues into account as often as possible decreases the likelihood of simultaneous changes to the same code sections and keeps the complexity of possible conflicts to a minimum.
The online development of HydPy relies, besides GitHub, on Travis CI and AppVeyor. Both are a hosted, distributed continuous integration services. They are configured to check and eventually release new HydPy versions in different ways. Each time someone suggests changes to the hydpy repository, either via a direct commit or a pull request (see section Version control), GitHub sends a message to the servers of Travis CI and AppVeyor. They then download the new source code, examine it, and log their results to the Travis log-page and the AppVeyor log-page, respectively.
travis.yml defines the following Travis CI workflow:
Download and install all HydPy dependencies on the Debian based Linux operating system Ubuntu separately for each currently supported CPython version.
Call setup.py to copy HydPy to the site-packages folder and build some additional files (e.g. Dynamic Link Library files). For reasons of measuring code coverage, setup.py also triggers all “conventional” unit tests and doctests.
Prepare a test coverage report (see section Tests & documentation).
Generate the online documentation (for the latest stable Python version).
Publish the online documentation on GitHub pages in case the build process succeeded, all tests passed, and no formatting errors or wrong links corrupted the documentation. Additionally, the evaluated branch must be the master branch.
The workflow defined by appveyour.yml is as follows:
As on Travis CI, download and install all HydPy dependencies for each supported CPython version, but on Windows Server instead of Linux.
As on Travis CI, call setup.py to install HydPy, but do not measure code coverage.
Create binary distributions for the respective Python versions based on the extension files generated in the last step.
Remove the already installed files, reinstall HydPy based on the new binary distributions, and apply all tests.
Make all artefacts available (currently, besides the binary distributions, the automatically generated XML schema files).
Upload the binary distributions to the Python Package Index for tagged commits in case all tests passed.
Together, both continuous integration services capture a wide range of possible problems, not only addressing the implemented hydrological models, but also the release and documentation cycles. If applied regularly, chances are good to trace unexpected errors to specific code changes. Additionally, Travis CI and AppVeyor help you to decide if contributions by colleagues are mature enough to be merged in your code.
