pypromice available through conda

October has been a chaotic month with many parts of my job converging. Firstly, we reached a big milestone with pypromice, the Python package for all our automated weather station (AWS) processing in the PROMICE and GC-Net network. Milestone reached: pypromice is now on conda-forge. This means that pypromice is installable through conda like so:

conda install pypromice -c conda-forge

This has been a long process that began as a suggestion in the review of pypromice through the Journal of Open Source Software (JOSS). Although it was not necessary for the review, we wanted to put pypromice on conda-forge - it was something I had never done before and was curious to see what the process was like.

An example of one of our weather stations in the PROMICE network. This is the newly installed NUK-B weather station, which we installed at the beginning of the month. NUK-B (NUK bedrock) forms a transect with our two other stations on the nearby Qamanaarsuup Sermia, NUK-U (NUK upper) and NUK-L (NUK lower)

The first step was to ensure that all pypromice’s dependencies were also available as conda packages. This was fine for all but one package - pyDataverse. We use Dataverse to store and mint all of our AWS datasets with citeable DOIs (see our data here), so pyDataverse is used in our package to access up-to-date versions of our AWS datasets through the Dataverse API. Getting pyDataverse on conda-forge was straightforward and fun to work with the developers.

After that, we started working on readying pypromice for conda-forge. The main thing was just generating the build from the already-existing pypi distribution of pypromice. In order to do so, the meta .yaml file can be generated initially using the grayskull Python package, like this:

grayskull pypi pypromice

There were some hiccups with the build as pypromice has CLI functionality, which previously was mapped in the package using a bin/ directory that was defined in the setup.py package build. However, after a first review this needed to be changed to map the CLI functionality with entry_points. This took some time, as I also took the opportunity to do some directory and file restructuring in pypromice.

Icebergs and bergy bits in Nuuk Fjord. This was the majority of the view in the helicopter back from installing NUK-B at the top of the fjord

Once accepted, we then had to wait for the conda-forge feedstock for pypromice to be generated. This feedstock automatically generates new builds from new pypromice versions, which can be reviewed and accepted through pull requests. Once merged, the feedstock updates with the new build. It’s a slick workflow that requires minimal effort on my side - just a quick check.