Gammapy project setup

This page gives an overview of the technical infrastructure we have set up to develop and maintain Gammapy.

If you just want to make contribution to the Gammapy code or documentation, you don’t need to know about most of the things mentioned on this page!

But for Gammapy maintainers it’s helpful to have a reference that explains what we have and how things work.

gammapy repository

This section explains the content of the main repository for Gammapy:

Package and docs

The two main folders of interest for developers are the gammapy folder and the docs folder. In gammapy you find the Gammapy package, i.e. all code, but also tests are included there in sub-folders called tests. The docs folder contains the documentation pages in restructured text (RST) format. The Sphinx documentation generator is used to convert those RST files to the HTML documentation.

Tutorials

The tutorials folder contains Jupyter notebooks that are part of the user documentation for Gammapy. They are copied to a docs/notebooks folder during the process of documentation building and converted to the Sphinx-formatted HTML files that you find in the Tutorial notebooks section. Raw Jupyter notebooks files and .py scripts versions are placed in the docs/_static/notebooks folder generated during the documentation building process.

We do have automated testing for notebooks set up (just check that they run and don’t raise an exception) in Travis CI (see below) which runs python -m gammapy.utils.tutorials_test and looks at the tutorials/notebooks.yaml file for which notebooks to test or not to test. It is also possible to perform tests locally on notebooks with the gammapy jupyter command. This command provides functionalities for testing, code formatting, stripping output cells and execution. See gammapy jupyter -h for more info on this.

The gammapy download command allows to download notebooks published as tutorials as well as the related datasets needed to execute them. For stable releases, the list of tutorials to download, their locations and datasets used are declared in YAML files placed in the download/tutorials folder of the gammapy-webpage Github repository. The same happens for conda working environments of stable releases declared in download/install folder of that repository. The datasets are not versioned and are similarly declared in the download/data folder.

Build

The setup.py and Makefile contain code to build and install Gammapy, as well as to run the tests and build the documentation, see How to contribute to Gammapy?.

The environment-dev.yml file contains the conda environment specification that allows one to quickly set up a conda environment for Gammapy development, see Get set up.

The astropy_helpers folder is a git submodule pointing to https://github.com/astropy/astropy-helpers It is used from setup.py (also using ah_bootstrap.py) and provides helpers related to Python build, installation and packaging, including a robust way to build C and Cython code from setup.py, as well as pytest extensions for testing and Sphinx extensions for the documentation build. If you look into those Python files, you will find that they are highly complex, and full of workarounds for old versions of Python, setuptools, Sphinx etc. Note that this is not code that we develop and maintain in Gammapy. Gammapy was started from https://github.com/astropy/package-template and there are besides the astropy_helpers folder a number of files (like ah_bootstrap.py, but also ez_setup.py or gamampy/_astropy_init.py) that are needed, but should be ignored. The Astropy team has set up a bot that from time to time makes pull requests to update the affiliated packages (including Gammapy) as new versions of astropy_helpers and the extra files are released.

The Dockerfile and binder.py files are used for Binder, see below.

Version

One more thing worth pointing out is how versioning for Gammapy works. Getting a correct version number in all cases (stable or dev version, installed package or in-place build in the source folder, …) is surprisingly complex for Python packages. For Gammapy, the version is computed at build time, by setup.py calling into the get_git_devstr helper function, and writing it to the auto-generated file gammapy/version.py. This file is then part of the Gammapy package, and is imported via gammapy/_astropy_init.py from gammapy/__init__.py. This means that one can simply do this and always get the right version for Gammapy:

>>> import gammapy
>>> gammapy.__version__
>>> gammapy.__githash__

Cython

We also have some Cython code in Gammapy, at the time of this writing less than 1% in these two files:

  • gammapy/detect/_test_statistics_cython.pyx
  • gammapy/maps/_sparse.pyx

and again as part of the Astropy package template there is the gammapy/_compiler.c file to help setup.py figure out information about the C compiler at build time. These are the files that are compiled by Cython and your C compiler when you build the Gammapy package, as explained in How to contribute to Gammapy?.

Other

There are two more folders in the gammapy repo: examples and dev. We started with the examples folder with the idea to have Gammapy usage examples there and have them be part of the user documentation. But this is not the case at the moment, rather examples is a collection of scripts that have mostly been used by developers to develop and debug Gammapy code. Most can probably just be deleted, some should be moved to user documentation (not clear where, could move all content to notebooks) or automated tests. The idea for the dev folder was to just have a place for scripts and checks and notes by Gammapy developers. Like for examples, it’s mostly outdated cruft and should probably be cleaned out.

The files .travis.yml, appveyor.yml, readthedocs.yml and environment-rtd.yml are the configuration files for the continuous integration (CI) and documentation build / hosting cloud services we use. They are described in sections further down on this page.

Finally, there are some folders that are generated and filled by various build steps:

  • build contains the Gammapy package if you run python setup.py build. If you run python setup.py install, first the build is run and files placed there, and after that files are copied from the build folder to your site-packages.
  • docs/_build contains the generated documentation, especially docs/_build/html the HTML version.
  • htmlcov and .coverage is where the test coverage report is stored.
  • v is a folder Pytest uses for caching information about failing tests across test runs. This is what makes it possible to execute tests e.g. with the --lf option and just run the tests that “last failed”.
  • dist contains the Gammapy distribution if you run python setup.py sdist

gammapy-extra repository

For Gammapy we have a second repository for example data files, Jupyter notebooks and a few other things:

Example data

The datasets and test_dataset folders contain example datasets that are used by the Gammapy documentation and tests. Note that here is a lot of old cruft, because Gammapy was developed since 2013 in parallel with the development of data formats for gamma-ray astronomy (see below).

Many old files in those folders can just be deleted; in some cases where documentation or tests access the old files, they should be changed to access newer files or generate test datasets from scratch. Doing this “cleanup” and improvement of curated example datasets will be an ongoing task in Gammapy for the coming years, that has to proceed in parallel with code, test and documentation improvements.

Other

  • The figures folder contains images that we show in the documentation (or in presentations or publications), for cases where the analysis and image takes a while to compute (i.e. something we don’t want to do all the time during the Gammapy documentation build). In each case, there should be a Python script to generate the image.
  • The experiments and checks folders contain Python scripts and notebooks with, well, experiments and checks by Gammapy developers. Some are still work in progress and of interest, most could probably be deleted.
  • The logo folder contains the Gammapy logo and banner in a few different variants.
  • The posters and presentations folders contain a few Gammapy posters and presentations, for cases where the poster or presentation isn’t available somewhere else on the web. It’s hugely incomplete and probably not very useful as-is, and we should discuss if this is useful at all, and if yes, how we want to maintain it.

Other repositories

Performance benchmarks for Gammapy:

Data from tutorials sometimes accesses files here:

Information from meetings is here:

Gammapy webpages

There are two webpages for Gammapy: gammapy.org and docs.gammapy.org.

In addition we have Binder set up to allow users to try Gammapy in the browser.

gammapy.org

https://gammapy.org/ is a small landing page for the Gammapy project. The page shown there a static webpage served via Github pages.

To update it, edit the HTML and CSS files this repo: https://github.com/gammapy/gammapy-webpage and then make a pull request against the default branch for that repo, called gh-pages. Once it’s merged, the webpage at https://gammapy.org/ usually updates within less than a minute.

docs.gammapy.org

https://docs.gammapy.org/ contains most of the documentation for Gammapy, including information about Gammapy, the changelog, tutorials, …

TODO: describe how to update.

Gammapy Binder

We have set up https://mybinder.org/ for Gammapy, which allows users to execute the tutorial Jupyter notebooks in the web browser, without having to install software or download data to their local machine. This can be useful for people to get started, and for tutorials. Every HTML-fixed version of the tutorial notebooks that you can find in the Tutorial notebooks section has a link to Binder that allows you to execute the tutorial in the myBinder cloud infrastructure.

myBinder provides versioned virtual environments coupled with every Github commit of the gammapy Github repository. The Binder docker image is created using the Dockerfile and binder.py files. The Dockerfile makes the Docker image used by Binder running some linux commands to install base-packages and copy the tutorials and datasets neeeded. It executes binder.py to conda install Gammapy dependencies listed in the environment YAML file placed in the download/install folder of the gammapy-webpage Github repository

Continuous integration

We also have a Jenkins server set up at MPIK (at https://www.mpi-hd.mpg.de/gamma-jenkins ) that is running on Ubuntu. We could use it to e.g. run more extensive CI builds such as e.g. making nightly or weekly test releases and running an extensive set of “science verification” tests that might involve larger datasets or be slow. It could also be used for performance tests, to check for regressions in CPU or memory usage. If anyone is interested in setting this up for Gammapy, please get in touch.

Releases

At this time, making a Gammapy release is a sequence of steps to execute in the command line and on some webpages, that is fully documented in this checklist: How to make a Gammapy release. It is difficult to automate this procedure more, but it is already pretty straightforward and quick to do. If all goes well, making a release takes about 1 hour of human time and one or two days of real time, with the building of the conda binary packages being the slowest step, something we wait for before announcing a new release to users (because many use conda and will try to update as soon as they get the announcement email).

Data formats

Data formats should be defined here, and then linked to from the Gammapy docs: