Using conda for Development Dependencies

The conda package for sunpy specifies only those dependencies that are relevant for the typical user. A power user or a developer will instead need the complete set of dependencies to use all optional features, build the documentation, and/or run the full test suite. We provide a conda environment file (sunpy-dev-env.yml) in the base directory of the GitHub repository. You can create a conda environment (here named sunpy-dev) from this file:

$ conda env create -n sunpy-dev -f sunpy-dev-env.yml
$ conda activate sunpy-dev

Alternatively, if you want to add the complete set of dependencies to an existing conda environment, you can use this file to update that environment (here named existing-env):

$ conda env update -n existing-env -f sunpy-dev-env.yml

The above calls assume that you have already downloaded sunpy-dev-env.yml, either by itself or along with the whole repository. You can alternatively supply the URL for the file, as hosted on GitHub, to either of the above calls, e.g.:

$ conda env create -n sunpy-dev -f https://raw.githubusercontent.com/sunpy/sunpy/main/sunpy-dev-env.yml

Note

This conda environment file intentionally does not specify restrictions on release versions for any dependency.

This conda environment file specifies only the dependencies for sunpy, and not sunpy itself. Depending on your needs, continue with one of the following two ways to install sunpy in this conda environment.

Normal Installation of sunpy

If you do not plan to modify the code of sunpy itself, you can simply install sunpy via conda:

$ conda install sunpy

Since the conda environment already has the complete set of dependencies for sunpy, this call should install only sunpy and no additional packages.

Editable Installation of sunpy

If you plan to modify the code of sunpy itself, you will want to perform an “editable install” of your local repository, via pip, so that Python will link to your local repository. Normally it is discouraged to have an environment that mixes package installations via conda with package installations via pip because it can lead to environment states that confuse the conda solver. That is the reason why our instructions for new developers recommends that dependencies be installed exclusively via pip. However, some of the dependencies in the complete set are difficult or even impossible to install via pip alone, yet are straightforward to install via conda.

Using the above conda environment, combined with a little care, it is possible to minimize that chance for any conda/pip conflicts. From the base directory of your local repository, install sunpy with some additional pip options:

$ pip install --no-deps --no-build-isolation -e .

The --no-deps and --no-build-isolation options ensure that pip does not itself install any dependencies. Since the conda environment is designed to already have the complete set of dependencies, the pip installation should succeed.

You now have a conda environment with an editable installation of sunpy and with (nearly) all dependencies managed by conda. As you install other packages in this environment, a package that depends on sunpy will trigger conda to install sunpy. That is fine! This conda installation of sunpy will simply mask the pip installation of sunpy. All you need to do is to remove the conda installation with the --force option so that dependencies are left undisturbed:

$ conda remove --force sunpy

Once the conda installation of sunpy is removed, the pip installation of sunpy will automatically be accessible again.

Note

For those who use mamba instead of conda, most conda commands can be translated by simply substituting “mamba” for “conda”. However, mamba remove does not support the --force option, so you do in fact have to call conda remove.

As a tip, you can follow a similar procedure to incorporate editable installations of other packages (e.g., astropy) in a conda environment. You first install the package via conda to ensure its dependencies are present, then you remove the package alone without disturbing the dependencies, and finally you perform the editable install of the package from the base directory of your local repository:

$ conda install astropy
$ conda remove --force astropy
$ pip install --no-deps --no-build-isolation -e .