Anaconda3 Incompatible with Old Macs

[cpiker]

Anaconda3 has decided to no longer support CPUs without AVX features for it's MacOS NumPy distribution conda/conda#9678 . Numpy 1.17.3 (or maybe .4) is the last version that doesn't require AVX CPU features.

This is not good for MacOS users. Since Macs are expensive and you can't legally run them in a virtual machine, people hang on to old Mac hardware for a long time, even Mac hardware that doesn't support AVX CPU features.

On solution is to require numpy <=1.17.3 for MacOS builds. For user running Miniconda3 this is a reasonable solution. However, for the the full up Anaconda3 distribution, this does not work. Packages (such pycdf) that are not compatible with the latest NumPy trigger a blizzard of package changes which end up failing after a long time.

At this point pycdf is only compatible with Miniconda3 on MacOS.

Volunteers with newer mac hardware are invited to step up and build the package for MacOS so that a version compatible with Anaconda3 on MacOS exists.

[ayushkarnawat]

I'm wouldn't go so far to say that Anaconda no longer want to support CPUs without AVX instructions. This is evident by the fact that using the conda-forge channel to install numpy 1.18.x (using conda install -c conda-forge numpy) works flawlessly on my MBP 2010.

Rather, I also think they made a mistake and built numpy (on the default channel i.e. conda install numpy) without flags saying to support CPUs without AVX instructions, as mentioned by cbrnr here.