Installation issues

[dandavies99]

Part of JOSS review and related to #22 and #20.

Installation instructions should probably be reviewed. At the moment there are a few interrelated problems:

• pip install surfinpy results in a successful installation, but python setup.py test fails due to no module named 'yaml'. Suggest that pyyaml needs adding to install_requires and on the version on PyPI.

• The alternative install instructions (pip install -r requirements.txt etc.) don't work on a mac (at least not on my machine - osx 11.5.1 with xcode 13.2.1). This is because of a gcc error (see below) that comes from trying to install pymatgen==2020.1.28. There have been similar issues around this before. My suggestion would be to pin to a more recent version of pymatgen in requirements.txt. Alternatively, set pymatgen>=2020.1.28 and numpy>=1.17.2 if possible.

gcc: error: unrecognized command-line option '-stdlib=libc++'
error: command 'gcc' failed with exit status 1

• Again, relating to pymatgen, the package is not actually compatible with python=3.5+ as it says in the README due to pymatgen==2020.1.28 requirement (see below). Suggest change to python 3.6+ and adding python_requires='>=3.6', to setup() in setup.py.

Could not find a version that satisfies the requirement pymatgen==2020.1.28

• In general, if specific versions of every dependency are pinned in requirements.txt, it might be good to add in the README that it is recommended to create a new (conda) environment for this library. Otherwise, the chances of this installing nicely in the same environment of other packages that use matplotlib, numpy, pymatgen etc. are quite low.

• The local install instructions could probably be simplified. pip install -r requirements.txt -e . will install the package in the current directory using pip so you get the advantage of pip knowing about it and it's dependencies. -e is optional and installs in "editable mode" so any changes made to the code base are applied immediately.

[awvwgk]

• pip install surfinpy results in a successful installation, but python setup.py test fails due to no module named 'yaml'. Suggest that pyyaml needs adding to install_requires and on the version on PyPI.

Interesting, didn't notice this when installing in a conda env and pip check didn't find an issue as well.

Edit: Looks like pyyaml is a dependency of older pymatgen versions.

surfinpy[2.0.0]
  ├─ pymatgen[2019.10.16]
  │  ├─ pandas[0.17.0]
  │  │  ├─ pytz already visited
  │  │  ├─ python-dateutil already visited
  │  │  ├─ python already visited
  │  │  └─ numpy already visited
  │  ├─ requests[2.10.0]
  │  │  └─ python already visited
  │  ├─ tqdm[4.10.0]
  │  │  └─ python already visited
  │  ├─ netcdf4[1.2.4]
  │  │  ├─ setuptools already visited
  │  │  ├─ python already visited
  │  │  ├─ numpy already visited
  │  │  ├─ hdf5 >>> NOT FOUND <<<
  │  │  └─ libnetcdf >>> NOT FOUND <<<
  │  ├─ sympy[1.0]
  │  │  ├─ fastcache[1.0.2]
  │  │  │  └─ python already visited
  │  │  ├─ python already visited
  │  │  └─ mpmath[1.2.1]
  │  │     └─ python already visited
  │  ├─ tabulate[0.7.5]
  │  │  └─ python already visited
  │  ├─ pybtex[0.21]
  │  │  ├─ setuptools already visited
  │  │  ├─ python already visited
  │  │  ├─ latexcodec[2.0.1]
  │  │  │  ├─ python already visited
  │  │  │  └─ six already visited
  │  │  └─ pyyaml[6.0]
  │  │     ├─ libgcc-ng already visited
  │  │     ├─ python_abi already visited
  │  │     ├─ python already visited
  │  │     └─ yaml[0.2.5]
  │  │        └─ libgcc-ng already visited

[dandavies99]

Yeah, that makes sense. Over in surfaxe we switched to ruaml (e.g. from ruamel.yaml import YAML) the same as pymatgen, seeing as it was a dependency anyway.

[jstse]

Removed the requirements for specific versions as we do not required specific versions. Hopefully, this will correct the installation issues you was having.

Updated the local instalation instructions as requested to use pip install -r requirements.txt -e .

See Pull request for more information #27