osx-arm64 package on conda-forge (for apple silicon)?

[jucor]

Dear SH team

Could you please create a (set of) eo-learn builds for Apple Silicon (m1 / osx-arm64) on conda-forge, please? https://anaconda.org/conda-forge/eo-learn-geometry doesn't seem to have it, so it can't be installed. See https://anaconda.org/conda-forge/rasterio for an example that works :) Thank you!

[jucor]

(meanwhile I'll try with pip-within-conda and report here if it works -- it has some potential for conflicts or redundant dependencies between the conda and the pip versions, but that might suffice as a stopgap measure)

[jucor]

Uh oh. Yup, pip install within conda not working nicely for eo-learn on m1. It seems pip install eo-learn is trying to build wheels for some if its dependencines, in particular for numpy, and that's error-ing :/

[batic]

Hi Julien,

Unfortunately, the conda-forge eo-learn recipes are not maintained by us, so you will have to open an issue there (e.g. https://github.com/conda-forge/eo-learn-io-feedstock)

Can you use the eo-learn docker image?

[jucor]

Hi @batic ! Docker image doesn't quite cut it, sadly. Are you familiar with compilation of the feedstock? Happy to work with you on bringing it up to date -- could be a good thing for EO-learn to own its own packaging, too, distribution-wise :)

[chase-dwelle]

Had this issue too on a new machine, problems are related to scipy, which requires numba. It seems like the numba issue is nearing a resolution: https://github.com/numba/numba/issues/7822

[devisperessutti]

We recently tried to install eo-learn and managed to install it successfully with the following steps (updating conda seems the key step).

• Download Anaconda from here
• Install Anaconda (double-click on Anaconda3-2021.11-MacISX-x86_64.pkg)
• once installed, check installation with conda list
• update conda with conda update conda
• create new environment with conda create -n test-eo-learn python=3.9
• activate env conda activate test-eo-learn
• add conda-forge conda config --add channels conda-forge and conda config --set channel_priority strict
• installation of single packages like conda install eo-learn-core or conda install eo-learn-io version v0.10.1 works
• installation of eo-learn from source as
  • clone from github git clone https://github.com/sentinel-hub/eo-learn.git
  • change directory to eo-learn
  • install in development mode as python install_all.py -e
  • this ran with no errors and we could import tasks from features from latest v1.0.0

Let us know if this works for you.

[mlubej]

IIUC, we have recently gained control over the conda-forge, but not sure what was done in relation to fixing issues mentioned in this post.

In any case, closing this since there was no activity. Feel free to reopen if it's still relevant.

[zigaLuksic]

IIUC, we have recently gained control over the conda-forge, but not sure what was done in relation to fixing issues mentioned in this post.

In any case, closing this since there was no activity. Feel free to reopen if it's still relevant.

we're still not in a position to solve the conda issue, hopefully we can do more once we switch to a single-package installation