Skip to content

Latest commit

 

History

History
216 lines (136 loc) · 6.42 KB

installation.rst

File metadata and controls

216 lines (136 loc) · 6.42 KB
Raw

Installation

sisl is easy to install using any of your preferred methods.

Required dependencies

Running sisl requires these versions:

Optional dependencies:

sisl implements certain methods in Cython which speeds up the execution. Cython is required if one wishes to re-generate the C-sources with a different Cython version. Note that this is not a necessary step and should typically only be considered by developers of Cython modules.

Installation of the stable sisl releases can be done by following the common conventions using pip or conda methods:

pip

pip will install from pre-build wheels when they are found, if not it will try and install from a source-distribution.

python3 -m pip install sisl
# for better analysis
python3 -m pip install sisl[analysis]
# for advanced plotting functionality
python3 -m pip install sisl[viz]

conda

Conda enviroments are clever, but fragile. It is recommended to contain the sisl installation in a separate environment to decouple it from other fragile components. Their inter-dependencies may result in problematic installations.

conda create -n sisl
conda activate sisl
conda config --add channels conda-forge
conda install -c conda-forge python sisl

dev|pip

This is equivalent to a development installation which requires a C and fortran compiler as well as some other packages:

python3 -m pip install setuptools_scm "scikit-build-core[pyproject]" Cython
python3 -m pip install git+https://github.com/zerothi/sisl.git --prefix <prefix>

The remaining dependencies should automatically be installed.

dev|conda

Using conda as development environment can be done, but may be a bit more cumbersome to work with. To install sisl from sources one needs a conda environment with the following content:

conda create -n sisl
conda activate sisl
conda config --add channels conda-forge
conda install -c conda-forge fortran-compiler c-compiler python scikit-build-core pyproject-metadata
conda install -c conda-forge cython scipy netcdf4 cftime plotly matplotlib

subsequent installations of sisl should follow dev|pip tab

editable|pip

Editable installs are currently not fully supported by scikit-build-core and is considered experimental. One may get it to work by doing:

git clone git+https://github.com/zerothi/sisl.git
cd sisl
python3 -m pip install -e .

Passing options to the build-system through pip should de done with the following convention

pip>=22.1

python3 -m pip install --config-settings=cmake.define.CMAKE_BUILD_PARALLEL_LEVEL=5 ...

pip<22.1

python3 -m pip install --global-option=cmake.define.CMAKE_BUILD_PARALLEL_LEVEL=5 ...

In the above case the compilation of the C/Fortran sources are compiled in parallel using 5 cores. This may greatly reduce compilation times.

There exists a set of compile time definitions that may be handy for developers. These are all CMake definitions and can be added like this:

pip>=22.1

python3 -m pip install --config-settings=cmake.define.WITH_FORTRAN=YES .

pip<22.1

python3 -m pip install --global-option=cmake.define.WITH_FORTRAN=YES .

The options are:

  • WITH_FORTRAN default to ON If OFF, no fortran sources will be compiled, this may be useful in debug situations, but are required for full support with externally created fortran files, such as output files from DFT codes.
  • WITH_F2PY_REPORT_EXIT default to OFF If ON, the compile definition -DF2PY_REPORT_ATEXIT will be set.
  • WITH_F2PY_REPORT_COPY default to OFF If ON, error messages will be printed while running when the array size has some certain size (see F2PY_REPORT_ON_ARRAY_COPY)
  • F2PY_REPORT_ON_ARRAY_COPY default 10 Minimum (total) number of array elements an array should have before an error is created when reporting a copy, WITH_F2PY_REPORT_COPY must also be ON for this to take effect.
  • WITH_LINE_DIRECTIVES default to OFF Add line-directives when cythonizing sources
  • WITH_GDB default to OFF Add information for the GDB debugger
  • WITH_ANNOTATE default to OFF create annotation output (html format) that can be viewed

Warning

Only developers should play with these flags at install time.

And in particular using WITH_FORTRAN=OFF will reduce the functionality of sisl (no fortran binary file support).

Windows

To install sisl on Windows one will require a specification of the compilers used. Typically one may do

python3 -m pip install . --prefix=<prefix>

but sometimes setuptools does not intercept the flags in the build process. Since 3.12 distutils has been deprecated and one needs to pass explicit linker flags to the CMake environment. If problems arise, please help out the community by figuring out how sisl can be compiled on Windows.

Testing your installation

After installation (by either of the above listed methods) you are encouraged to perform the shipped tests to ensure everything got installed correctly.

Note that pytest needs to be installed to run the tests. Testing the installation may be done by:

pytest --pyargs sisl

The above will run the default test-suite which covers most of the sisl tool-box. Additional tests may be runned by cloning the sisl-files and setting the environment variable SISL_FILES_TESTS as the tests path to the repository.

A basic procedure would be:

git clone https://github.com/zerothi/sisl-files.git
SISL_FILES_TESTS=$(pwd)/sisl-files/tests pytest --pyargs sisl