developer install currently broken on macOS

[kosack]

Using a clean install of miniconda (or anaconda), and following the install instructions fails for both python 3.7 and 3.8 with an error of Found conflicts! Looking for incompatible packages.. Seems there is a missing package or too-strict dependency somewhere for macOS, since this works fine for Linux.

We used to test installs on macOS in Travis - why did that stop?

steps to reproduce:

PYTHON_VERSION=3.7 sed -i -e "s/- python=.*/- python=$PYTHON_VERSION/g" environment.yml
conda env create -n ctadev --file environment.yml

[maxnoe]

At some point the build did not work and we removed it.

[kosack]

At some point the build did not work and we removed it.

since many of the devs (myself included) use macOS, we should think about fixing that.

[kosack]

Seems simply adding conda-forge to the channel list fixes it. So I guess using - conda-forge::<PACKAGE> tries to get a single package from conda-forge, but not its dependencies, leading to no solution during dependency resolution

# A conda environment with all useful package for ctapipe developers
name: cta-dev
channels:
  - default
  - conda-forge
  - cta-observatory
...

[maxnoe]

@kosack that seems to be it.

By the way, I learned today that astropy imports their conda-forge packages to the astropy namespace, so that people don't have to solve vs. the full conda forge.

We could do the same.

[maxnoe]

Wenn we add the conda-forge channel, now suddenly everything comes from forge. Do we want that?

That for example means no mkl

[kosack]

By the way, I learned today that astropy imports their conda-forge packages to the astropy namespace, so that people don't have to solve vs. the full conda forge.

That's an option, since requiring everything from conda-forge is maybe not the best idea. It means some more maintenance for us though, since I guess they have to move packages to their channel.

There must be some way to keep the basic packages in defaults like numpy MKL, and still use a few packages from conda-forge. Not sure why the deps are so restrictive now.

[maxnoe]

For now you can probably simply add conda-forge::corsikaio, does that solve you problem?

[maxnoe]

I think this was fixed? @kosack can you confirm?