pyspedas installation issue on Mac M1(?) (building blosc library)

[jameswilburlewis]

A user had some issues installing pyspedas on an Apple Silicon Mac. Looking through my notes, I also had the same issue and was able to solve it by installing another package. (The issue was several layers of dependencies deep in the NetCDF4 package, there was a package called 'blosc' that didn't install cleanly). The user and I were both eventually able to finish the install -- I'm waiting to find out exactly what they did and compare it to my solution.

We will want to address this and similar issues users might encounter in our "getting started with pyspedas" documentation.

[jameswilburlewis]

The original issue might have been resolved upstream by now. However, installing netCDF4 under Python 3.13 on Mac Ventura/ARM is still an issue. (Github dropped ARM support for Ventura, so netCDF4 is only making ARM wheels for OSX 14 and 15.). See the issue I opened on their repo: Unidata/netcdf4-python#1385

conda install -c conda-forge netCDF4 works for now. Also, there is supposedly a working binary wheel here:

https://files.pythonhosted.org/packages/58/3e/5736880a607edabca4c4fc49f1ccf9a2bb2485f84478e4cd19ba11c3b803/netCDF4-1.7.2-cp313-cp313-macosx_14_0_arm64.whl

So we should give this wheel a try, and if it works, refer to it in our installation guide. (conda install -c conda-forge netCDF4 also works, but people don't necessarily want to mix pip and conda).

[plm203]

Hi James, I am having problems installing netCFD4 on a M2 x86 processor with Python 3.13. I have the default Mac OS Ventura 13.4. If i understand well there should be a wheel for my setup, as it is not ARM-based, on some recent version of netCFD4. Is this correct and if so where can such a wheel be found ? or how to get it installed with some magical command ? Alternatively if i install a newer version of MacOS should i be able to install netCFD4 via "pip install" ? Thank you very much.

PS: Since AI is gaining some maturity is there some project to develop AI tools fixing those hyper-time-consuming compatibility issues ? Like AI generating module code for newer language version - new Python releases - or at least building wheels and checking everything works, or databasing troubleshooting processes with a user-friendly interface ? It is quite insane that people expect to make robots work for complex real-world tasks while not even being able to make python packages on a newer version of Python with ample documentation. I don't have stats but i would bet the time lost to those gremlins, over the planet, is equivalent to hundreds of GUSD. Thank you again.