env_climada.yml improvements?

[mathause]

I found out that python is not listed in env_climada.yml - you could add add it along with its pin. (That could also slightly simplify the advanced installation instructions).

Further peewee is required in setup.py but not in env_climada.yml

[mathause]

overpy, deprecation, and bayesian-optimization also seem to be missing

[emanuel-schmid]

@mathause: Thanks for reaching out!

The purpose of env_climada.yml file is to allow users to install climada in the most simple way possible, and not to be a record of the environment itself.
Accordingly we try to avoid redundancy and keep the file minimal.

• python is a requirement of many listed packages
• peewee, depreccation, overpy and bayesian-optimization are (or used to be) available from pypi only. Listing them in the pip section would just make maintenance more complex.

[mathause]

python is a requirement of many listed packages

Correct, but they don't pin python to <3.12. So I had to delete the environment and start over which was annoying. (Admittedly I did not follow your installation instructions where this is described.)

peewee, deprecation, overpy and bayesian-optimization are (or used to be) available from pypi only. Listing them in the pip section would just make maintenance more complex.

I found all of them on conda-forge (and they are all available for more than five years).

The purpose of env_climada.yml file is to allow users to install climada in the most simple way possible, and not to be a record of the environment itself. Accordingly we try to avoid redundancy and keep the file minimal.

I assume you have to change the file anyway when you change your pins (and I totally agree that it is super annoying that we need to keep track of this in several locations!) But of course up to you how to handle this - feel free to close without further comment.

[emanuel-schmid]

Correct, but they don't pin python to <3.12. So I had to delete the environment and start over which was annoying. (Admittedly I did not follow your installation instructions where this is described.)

Sorry for that. That's a valid point though.

I found all of them on conda-forge (and they are all available for more than five years).

Yes, totally true for deprecation, peewee and probably overpy. Our policy to take whatever is possible from conda-forge has been broken for these. We might update that at some point.
As for bayesian-optimization that's different story: at at the time it was introduced (around July 2024) nothing from conda-forge was compatible with the other packages under the then given restrictions. Finding a valid dependency configuration that works (i.e., can be installed and runs) for Windows, OSX and Linux is rather difficult. I won't touch this until I have to.