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.
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
docs folder. In
gammapy you find the Gammapy package, i.e. all code,
but also tests are included there in sub-folders called
folder contains the documentation pages in restructured text (RST) format. The
Sphinx documentation generator is used to convert those RST files to the HTML
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
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
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.
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
The same happens for conda working environments of stable releases declared
download/install folder of that repository. The datasets are not versioned and
are similarly declared in the
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?.
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.
astropy_helpers folder is a git submodule pointing to
https://github.com/astropy/astropy-helpers It is used from
ah_bootstrap.py) and provides helpers related to Python build,
installation and packaging, including a robust way to build C and Cython code
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 few files (
gammapy/_astropy_init.py) that are needed, but rarely
need to be looked at or updated. 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
astropy_helpers and the extra files are released.
binder.py files are used for Binder, see below.
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
calling into the
get_git_devstr helper function, and writing it to the
gammapy/version.py. This file is then part of the
Gammapy package, and is imported via
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__
We also have some Cython code in Gammapy, at the time of this writing less than 1% in these two files:
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?.
There are two more folders in the
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.
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:
buildcontains 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
buildfolder to your
docs/_buildcontains the generated documentation, especially
docs/_build/htmlthe HTML version.
.coverageis where the test coverage report is stored.
vis 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
--lfoption and just run the tests that “last failed”.
distcontains the Gammapy distribution if you run
python setup.py sdist
For Gammapy we have a second repository for example data files, Jupyter notebooks and a few other things:
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.
figuresfolder 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.
checksfolders 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.
logofolder contains the Gammapy logo and banner in a few different variants.
presentationsfolders 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.
Performance benchmarks for Gammapy:
Data from tutorials sometimes accesses files here:
Information from meetings is here:
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.
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.
https://docs.gammapy.org/ contains most of the documentation for Gammapy, including information about Gammapy, the changelog, tutorials, …
TODO: describe how to update.
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
Github repository. The Binder docker image
is created using the
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
- Windows CI: https://ci.appveyor.com/project/cdeil/gammapy/branch/master
- Mac and Linux CI: https://travis-ci.org/gammapy/gammapy
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.
- Code quality: https://landscape.io/github/gammapy/gammapy/master
- Code coverage: https://coveralls.io/r/gammapy/gammapy
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 should be defined here, and then linked to from the Gammapy docs: