Skip to content

Latest commit

 

History

History
336 lines (223 loc) · 11.8 KB

installation.rst

File metadata and controls

336 lines (223 loc) · 11.8 KB
Raw

Installation

img { padding-top: 10pt; padding-bottom: 12pt; } div.topic { border-style: none; }

Dependencies

  • Required
    • Python 2.6, 2.7, or 3.3+
    • Numpy 1.9+
    • Argparse (Python 2.6 only)
  • Recommended
    • IPython (better interactive interpreter)
    • Matplotlib 1.4+ (if you want to make 2d plots using viscid.plot.vpyplot)
    • Scipy (enables nonlinear interpolation and curve fitting)
    • Numexpr (for faster math on large grids)
    • H5py (enables hdf5 reader)
  • Optional
    • Seaborn
    • Mayavi2 [1] (if you want to make 3d plots using viscid.plot.vlab)
    • PyYaml (rc file and plot options can parse using yaml)
  • Optional for developers
    • Cython 0.28+ (if you change pyx / pxd files)
    • Sphinx 1.3+
[1]Installing Mayavi can be tricky. Please :ref:`read this section <installing-mayavi>` before attempting to install it.

Installing Anaconda (optional)

The Anaconda Python Distribution makes managing dependencies and virtual environments straight forward (and almost pleasant). You can download the full distribution, but it's laden with packages you probably don't need. Also, since conda install is so easy to use, I recommend the lightweight miniconda:

if [ "$(uname -s)" == "Linux" ]; then
  wget -O miniconda.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.sh
elif [ "$(uname -s)" == "Darwin" ]; then
  curl -o miniconda.sh https://repo.anaconda.com/miniconda/Miniconda3-latest-MacOSX-x86_64.sh
fi
bash miniconda.sh -b -p $HOME/local/anaconda
rm miniconda.sh
source $HOME/local/anaconda/etc/profile.d/conda.sh
conda config --set changeps1 no
conda update -q conda
conda activate

You can now add something like this to your bashrc or profile,

export _CONDA_ROOT_PREFIX=${HOME}/local/anaconda  # <- point to anaconda
export CONDA_DEFAULT_ENV=base  # <- edit this to taste
# there is no need to edit the following lines directly
source ${_CONDA_ROOT_PREFIX}/etc/profile.d/conda.sh
export CONDA_SHLVL=1
export CONDA_EXE=${_CONDA_ROOT_PREFIX}/bin/conda
if [ "${CONDA_DEFAULT_ENV}" = "base" ]; then
  export CONDA_PREFIX="${_CONDA_ROOT_PREFIX}"
else
  export CONDA_PREFIX="${_CONDA_ROOT_PREFIX}/envs/${CONDA_DEFAULT_ENV}"
fi
export CONDA_PROMPT_MODIFIER=""
export CONDA_PYTHON_EXE="${_CONDA_ROOT_PREFIX}/bin/python"
export PATH="${CONDA_PREFIX}/bin:${PATH}"

Installing Viscid

You have a few choices for installing Viscid. Here is a quick breakdown of why you might choose one method over another.

  • :ref:`Anaconda <choice1-conda>`
    • + Installs with a single command
    • + No compiler needed
    • + Available for macOS, Linux, and Windows
    • + Automatically installs recommended dependencies
    • + Optional dependencies are easy to manage using conda
    • - You won't be able to edit Viscid's source code. You might naively edit the modules in site-packages, but this will confuse the conda package manager beyond repair.
  • :ref:`PyPI (pip) <choice2-pypi>`
    • + Installs with a single command
    • + No compiler needed
    • + Available for macOS, Linux, and Windows
    • ~ Dependencies can be managed using pipenv (or plain pip)
  • :ref:`Source <choice3-source>`
    • + Most flexible
    • ~ Dependencies can be managed using pipenv (or plain pip)
    • - Requires a C compiler for interpolation and streamline functions
    • - Requires a Fortran compiler for jrrle file support

Choice 1: Anaconda

Anaconda Version Anaconda Platforms

If you have Anaconda, then installing Viscid and all the recommended dependencies happens with one command,

conda install -c viscid-hub viscid

You can check that the install succeeded by running,

python -m viscid --check

Choice 2: PyPI (pip)

PyPI

Binary wheels are available on PyPI for Python 2.7, 3.5, 3.6, and 3.7 on MacOS, Linux, and Windows.

pip install viscid

If these wheels don't work for you, pip can also install from source (optionally requires a C compiler for interpolation and streamline support, and a Fortran compiler for Jrrle file support).

pip install --no-binary :all: viscid

Compile errors will not cause Viscid's pip install to fail, and pip hides warning messages unless you use the -v flag. To check the functionality of your install, run

python -m viscid --check

Choice 3: Source

First, you'll have to clone the Viscid git repository. This should be done in whatever directory you want to store the Viscid source code. I use ~/src myself.

git clone https://github.com/viscid-hub/Viscid.git

# Optional: set qt5 as the default matplotlib backend
mkdir -p ~/.config/matplotlib
echo "backend: Qt5Agg" >> ~/.config/matplotlib/matplotlibrc

# Optional: copy the default viscidrc file
cp Viscid/resources/viscidrc ~/.viscidrc

If you are using Anaconda to manage your dependencies, you can use the default Viscid environment to automatically install all Viscid's dependencies,

conda env create -f Viscid/resources/viscid37mayavi.yml

# if you need mayavi, but don't have OpenGL 3.2, you
# will have to use python2.7 (Viscid/resources/viscid27.yml)

# this activation must be done for each new command prompt
conda activate viscid37mayavi  # or viscid27, etc.

:ref:`Read this <conda_bashrc_blurb>` if you need help editing your bashrc or profile to set the default Anaconda environment.

Viscid should work with native compiler toolchains, but in case you don't have access to native compilers, the Anaconda toolchains should work too. Check here to find the appropriate packages for your platform. On Windows, you should use the MSVC compiler and Anaconda's m2w64-gcc-fortran. Read this for more information about acquiring the correct Microsoft compiler. If you are using CPython 3.5 or 3.6, you probably need the MSVC 14 compiler available here. You do not need to install visual studio to get the build tools.

Now you have a choice about how you want to use Viscid. If you intend to edit Viscid's source code, then I recommend building it inplace. Otherwise, it probably makes more sense to simply install Viscid into your python distribution.

Choice 3a: installed

cd Viscid
make install
# the above is synonymous with `python setup.py install --user`

To see if the install succeeded, try

# kick the tires
python -m viscid --check
# run the full test suite
make instcheck

To pull updates from github in the future, use

git pull
python setup.py install --user

Choice 3b: inplace

cd Viscid
make inplace
# the above line is synonymous with `python setup.py build_ext -i`

# To set environment variables in Bash
profile="${HOME}/.bashrc"
echo 'export PATH="${PATH}:'"${PWD}/scripts\"" >> ${profile}
echo 'export PYTHONPATH="${PYTHONPATH}:'"${PWD}\"" >> ${profile}
source ${profile}

To see if the build succeeded, try

# kick the tires
python -m viscid --check
# run the full test suite
make check

To pull updates from github in the future, use

git pull
python setup.py build_ext -i

Verify Installation

You can always run the following to check for any installation warnings. It is most helpful when verifying whether or not the C / Fortran modules compiled successfully.

python -m viscid --check

Installing Mayavi (optional)

Warning

Do not install Mayavi using pip into an Anaconda environment. This will break your environment in a way that requires you to reinstall Anaconda. The issue is that pip happily clobbers some parts of pyqt that are hard linked to a cache. You have been warned.

Note

I do not recommend using the conda-forge channel for VTK or Mayavi. These binaries frequently have runtime problems caused by interactions with specific versions of vtk / pyqt.

Installing Mayavi can be a mine field of incompatible dependencies. Here is a table to help you choose your poison. If your environment is not in the table, then it is likely not supported by Mayavi / VTK.

.. cssclass:: align-left

.. cssclass:: table-striped

.. cssclass:: table-hover
OS Python Version OpenGL / MESA Installation Command
MacOS 3.5+ OpenGL 3.2+ conda install -c viscid-hub mayavi
MacOS 2.7 OpenGL 3.2+ conda install -c viscid-hub mayavi
MacOS 2.7 Any conda install mayavi vtk=6
Linux 3.5+ OpenGL 3.2+, MESA 11.2+ conda install -c viscid-hub mayavi
Linux 2.7 OpenGL 3.2+, MESA 11.2+ conda install -c viscid-hub mayavi
Linux 2.7 Any conda install mayavi vtk=6
Windows 3.5+ OpenGL 3.2+ conda install -c viscid-hub mayavi
Windows 2.7 Any conda install mayavi vtk=6

The Enthought Tool Suite requires some confusing environment variables. Effectively you can just cycle through these options until you find a combination that works.

export ETS_TOOLKIT='qt'  # or "qt4" or "wx" for older versions of pyface
export QT_API="pyqt5"  # or "pyqt" or "pyside"

Also, if you are using MESA to emulate the new OpenGL API for VTK 7+, you may need the following environment variables,

export MESA_GL_VERSION_OVERRIDE=3.2
export MESA_GLSL_VERSION_OVERRIDE=150

:ref:`Check here <functions-mayavi>` for a discussion of Viscid's wrapper functions and workarounds, :doc:`and here <tutorial/mayavi>` for an example of Mayavi in action.

Workarounds

Ubuntu

All Linux workarounds are currently incorporated in setup.py.

OS X

If you see a link error that says -lgcc_s.10.5 can't be found, try running:

sudo su root -c "mkdir -p /usr/local/lib && ln -s /usr/lib/libSystem.B.dylib /usr/local/lib/libgcc_s.10.5.dylib"