install as many dependencies as possible with conda, then pip: a small example please ? about climetlab HOT 3 CLOSED

You do not have to install any dependencies with conda. pip install climetlab should work out of the box.

Depending on the other packages that you have, depending on their versions, depending on your system, depending when/how you will update them, you may want to install some dependencies with conda to solve dependency issues. "Some dependencies" means here "some of the packages in the setup.py"

from climetlab.

Hi @floriankrb,
did you try import eccodes or import xarray after pip-only install ?

ecmwflibs/__init__.py:142: UserWarning: ecmwflibs universal: cannot find a library called eccodes
  warnings.warn(f"ecmwflibs universal: cannot find a library called {name}")

The problem is that pip install eccodes installs .py with no error, but not libeccodes.dylib;
for that, you have to either get it from conda, or build it from source
and setenv ECCODES_DIR /opt/miniconda3 or the like.

The Right Thing to do would be to build libeccodes.dylib under eccodes/.dylibs, like numpy; see
https://stackoverflow.com/questions/47380150/dealing-with-dylibs-when-distributing-python-packages
https://github.com/matthew-brett/delocate

What do you think ?

from climetlab.

These are the installations supported directly by climetlab https://github.com/ecmwf/climetlab/actions/runs/2550905617
they are tested for each release under standard environments with github actions.

I do not know which platform you are using. Guessing from the fact that you got ecmwflibs universal, I am assuming this is not vanilla linux, nor macos, nor windows, or that you have special tweaks. In such cases, yes, you can install eccodes differently (conda, build from source, other), and make sure ecmwflibs finds it.

from climetlab.