How to add cartopy?

[RobertPincus]

For making maps in many cases we will want to add the cartopy Python module but simply adding the module to requirements.txt leads to unresolvable dependencies (https://github.com/RobertPincus/how_to_eurec4a/commit/acac4cc2486c9d152415dfbdb2955ee4322a472b). I can't get my local module stack to work with conda - can someone (@d70-t ?) show me how to modify requirements.txt?

(I love that autocorrect wants to change cartopy to "carroty")

[d70-t]

Yes, I was fearing that at some point we really want to have cartopy... I just closed #11 because I don't have a really good solution on how to do that.

The problem is that one can not "just" install cartopy via pip. As per their docs one has to install quite some dependencies before installing via pip. This makes the whole book more complicated to run on someones computer. I basically see two options:

• install dependencies via apt etc... on the continuous integrations machines and the binder environment before running the things -> users / authors would need to install cartopy dependencies manually on their machines prior to be able to compile the book
• pull in conda as a dependency and install cartopy via conda -> users / authors would need to have conda running on their machines

I personally never got used to conda but others may not be used to installing things manually... Do you have more ideas or preferences?

[RobertPincus]

Well, I use conda to manage environments in other CI situations (https://github.com/earth-system-radiation/rte-rrtmgp/blob/main/.github/workflows/continuous-integration.yml#L112).

Though using conda will also take some fussing - when I tried to set up a conda environment by just installing 'jupyter and the modules in requirements I got derailed because somehow one jupyter module was being called from the system and not the venv. I could maybe get this worked out.

Or we could make a container? https://github.com/earth-system-radiation/rte-rrtmgp/blob/main/.github/workflows/containerized-ci.yml#L10, https://github.com/earth-system-radiation/containers

[d70-t]

I am actually not worried too much of getting it to work within CI or binder. There's e.g. hackathon_env which properly sets up a binder environment (or docker image using repo2docker) which includes cartopy.

I am more worried of how to expain and maintain that explanation to someone who follows running locally > via git.

[d70-t]

Maybe we could also do a setup like it is described here on stackoverflow, i.e.: make a conda environment which includes the requirements.txt. That way we could keep a single source of information, but give a choice to users if they want to install via conda or via pip?

[RobertPincus]

It's great if we have a way for users of conda to run the book. But then how do we support users of pip?

[d70-t]

Well, if you want to run cartopy via pip you only have the option to install its dependencies manually... But chances are high that people already did that on their machines as cartopy is quite a common thing. And if cartopy should already be installed on someones machine, pip shouldn't complain during the installation.

That said, we would still need to explain the installation of cartopy in the pip case, as for some, pip install -r requirements.txt will fail.

[RobertPincus]

I made a conda environment.yml file that works locally (installs everything from requirements.txt plus cartopy in a conda environment, https://github.com/RobertPincus/how_to_eurec4a/tree/add-p3-examples) but I don't have the CI right yet

[RobertPincus]

Indeed I'm quite confused, @d70-t . Documentation suggests that the environment I create will be activate in later steps but conda info shows no environment active (https://github.com/RobertPincus/how_to_eurec4a/runs/2449453339?check_suite_focus=true) and both conda activate how_to_eureca and the equivalent source activate command fail.

Otherwise I would be ready to add examples from the P-3...

[d70-t]

Hmm, I probably can look into it tomorrow, but I have to admit, that I never really used conda, so maybe I'll have a hard time on this.

[RobertPincus]

I will also keep working... but it doesn't seem to be a problem with the conda installation but rather that the virtual environment isn't being invoked

[RobertPincus]

@d70-t Got it... it helps to follow directions