ESS2019 installation guide updates

[mkuron]

Lessons learned today:

• We require MPI 3 because we depend on const-correctness in a few places. That means that OpenMPI 1.6.5 and lower are not supported anymore.
• Installing the ROCm driver breaks access to /dev/kfd, causing hwloc initialization during mpiexec to hang. Rebooting helps.
• Add matplotlib, ipython and jupyter to the Mac install guide.
• Homebrew now defaults to Python 3, requires manually enabling cython, and it's unclear whether the hdf5 package still supports MPI (HDF5 formula no longer supports mpich Homebrew/homebrew-core#26974)
• Anaconda (~/anaconda[23]) and python.org packages (/Library/Python and /usr/local/bin) are also sources of conflict

Please tag for the 4.1.1 release

[codecov]

Codecov Report

Merging #3236 into python will decrease coverage by <1%.
The diff coverage is n/a.

@@          Coverage Diff           @@
##           python   #3236   +/-   ##
======================================
- Coverage      85%     85%   -1%     
======================================
  Files         528     528           
  Lines       25805   25805           
======================================
- Hits        22150   22149    -1     
- Misses       3655    3656    +1

Impacted Files	Coverage Δ
src/core/particle_data.cpp	95% <0%> (-1%)	⬇️

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update 9dc1e06...f191fb5. Read the comment docs.

[jonaslandsgesell]

If the full path of the build directory contains a space then the pypresso/ipypresso scripts do not work. When setting the variable PYTHONPATH the path to the espressomd module needs to be set in quotes in the scripts. I can describe this offline.

[mkuron]

@jonaslandsgesell, that‘s fixed in #3238 already.

[RudolfWeeber]

I think we should put all python dependencies into ```/requirements.txt``` and let the users install them via ```pip install -r requirements.txt```. That would mean to add the following packages to ```/requirements.txt```:

I think mixing distro-provided and pip-provided packages is a source for all kind of trouble during the installation. For distros, where it is possible, we should use distro packges alone.

[RudolfWeeber]

we explicitly tell the user to install pip

We probably shouldn't, unless there is a need to install something from pip.

[RudolfWeeber]

For what it's worth, even on Linux (where many packages are available from the system) I almost always use a venv with the packages for espresso as I find it simpler (also, no changes because of system updates). For execution only (no matplotlib etc), I install always install numpy, cython, and h5py in the venv.

I think that is only simpler for experts. For beginners its confusing to have to think about different environments for different python projects.

[RudolfWeeber]

can you come up with an example where a package from the package manager has a python package dependency if itself is not a python package?

Numpy is the classic. Using a custom numpy installed via pip for Espresso reasons and distro-provided numpy-using packages. When everithing is distro-provided, there is a good chance its compatible with each other and across different python apps.

[jngrad]

bors r=KaiSzuttor

[bors]

Canceled (will resume)

[jngrad]

bors r-
(failed CI)

[bors]

Canceled

[jngrad]

bors r=KaiSzuttor

[bors]

Build succeeded

• ICP GitLab CI