Installation with conda installs new compilers and kernel headers

[caioolivv]

Installing pyccl with conda from conda-forge forces the installation of new compilers and kernel headers, which breaks other libraries in the environment (in specific, NumCosmo). Is this expected behavior?

(At the moment, I can still install with pip so that's not a big problem)

[erykoff]

This appears to be because the camb conda-forge feedstock defines the fortran compiler as a run requirement which is ... odd.

[caioolivv]

Any way to go around this? I'm trying to install firecrown which requires pyccl through conda, so I'm not able to install it on the same environment as NumCosmo.

[erykoff]

Pinging @beckermr on this.

[beckermr]

Is numcosmo packaged in conda-forge?

[caioolivv]

Yes, but I'm working on a different branch of numcosmo.

[beckermr]

Got it. Thanks for the quick response.

conda + camb is doing the right thing here.

The issue is that you have a requirement on the compilers that you didn't tell conda about. You can tell conda about it by putting the requirements on the compilers into the conda-meta/pinned file in your environment's prefix. You can make that file if it does not exist.

Then conda will respect those pins when it does operations on the env.

[damonge]

I'm gonna close this one, but @caioolivv , let me know if Matt's solution doesn't work.