Pin matplotlib to <3.3 and add compilers

[bouweandela]

Cartopy 0.17 does not work with matplotlib > 3.2 SciTools/cartopy#1615 and pinning Cartopy to > 0.17 seems to result in an environment that is very difficult to solve (see my attempts in #1893).

Therefore I think it would be best to pin matplotlib to <3.3 for now and try to upgrade again in a month or so.

I also added the compilers package to make sure the conda compilers are always used. This solves an issue I was having in #1893 because cartopy failed to compile when trying to install a newer version of it with pip on top of environment.yml and it also makes sure all R packages are compiling with the correct compilers, the r-base package does have some compiler dependencies, but not all (possibly related to #1816).

This should fix #1894.

[Peter9192]

• Environment can be resolved quite easily
• Tried to run example python recipe and climwip PR with this env, both work
• I did get some warnings and clobber messages like these:

SafetyError: The package for iris located at /home/peter9192/miniconda3/pkgs/iris-2.4.0-py38_0
appears to be corrupted. The path 'lib/python3.8/site-packages/iris/plot.py'
has an incorrect size.

and

ClobberError: This transaction has incompatible packages due to a shared path.
  packages: conda-forge/linux-64::c-compiler-1.1.1-h516909a_0, conda-forge/linux-64::fortran-compiler-1.1.1-he991be0_0
  path: 'bin/gcc'

not sure if that can be ignored or not.

[bouweandela]

ClobberError: This transaction has incompatible packages due to a shared path.
packages: conda-forge/linux-64::c-compiler-1.1.1-h516909a_0, conda-forge/linux-64::fortran-compiler-1.1.1-he991be0_0
path: 'bin/gcc'

This error is to be expected because the GCC compiler is likely a single package and conda tries to take it apart into separate packages, so I think this should be fine.

SafetyError: The package for iris located at /home/peter9192/miniconda3/pkgs/iris-2.4.0-py38_0
appears to be corrupted. The path 'lib/python3.8/site-packages/iris/plot.py'
has an incorrect size.

I'm not sure about this warning, it doesn't appear for me when I run conda env create -n test -f environment.yml and I do not see it on CircleCI either, so it might be something related to your local machine. You can clean the cache of downloaded packages by runnign conda clean -a, maybe that helps?

[Peter9192]

Okay trying again with a clean conda cache

[Peter9192]

Okay trying again with a clean conda cache

Unfortunately this doesn't help. I get this for r-base, iris/plot and iris/quickplot. But indeed, the issue seems to be related to my machine then, I'll look into it a bit more later.