Skip to content

Latest commit

 

History

History
347 lines (205 loc) · 8.96 KB

INSTALL_LINUX.rst

File metadata and controls

347 lines (205 loc) · 8.96 KB
Raw

Installation on Linux

Pre-build binary install with conda

.. only:: html

  .. image:: https://anaconda.org/gimli/pygimli/badges/installer/conda.svg
      :target: https://conda.anaconda.org/gimli

  .. image:: https://anaconda.org/gimli/pygimli/badges/downloads.svg
      :target: https://anaconda.org/gimli/pygimli


On Linux platforms, the most comfortable way to install pygimli is via the conda package manager contained in the Anaconda distribution. Anaconda is scientific Python distribution with more than 100 Python packages included (~400 Mb). You can also use the lightweight alternative Miniconda (~35 Mb) and only install the packages you like to use.

Install Miniconda (only once):

wget http://repo.continuum.io/miniconda/Miniconda3-latest-Linux-x86_64.sh -O miniconda.sh
chmod +x miniconda.sh
./miniconda.sh -b -p $HOME/miniconda
export PATH=$HOME/miniconda/bin:$PATH # Note: Add this to your .bashrc for permanent use

Add the gimli and conda-forge channel (only once):

conda config --add channels gimli --add channels conda-forge

Install pygimli (and its dependencies) and test if everything works correctly via:

conda install -f pygimli
python -c "import pygimli; pygimli.test()"

Update your pygimli installation if want to have the newest functionality:

conda update -f pygimli

The only drawback using conda is that you are bound to the rhythm we update the binary packages. Conda also can be seen as a sandbox Linux inside your system and it might be difficult to combine system python packages and conda GIMLi. If you like to keep your GIMLi version more recent (including all possible drawbacks of versions that are actively developed) you should compile GIMli using your systems toolchain.

Compile your own with the curl installer

If you are not using Anaconda, you can build pyGIMLi from source in the current directory via:

curl -Ls install.pygimli.org | bash

This script accepts a few more options. For help see:

curl -Ls install.pygimli.org | bash -s help

If something goes wrong please take a look at the error message. In most cases there are missing or outdated packages. Please have a look at the prerequisites tab.

If the installation fails you can try the following instructions for manual installation.

Detailed Installation on Debian Stretch

In order to build pygimli (and gimli) for Python 2.7 and Python 3.5, install the required packages:

sudo apt-get install wget subversion git cmake mercurial \
    libboost-all-dev libblas-dev liblapack-dev \
    python python-setuptools \
    python-numpy python-matplotlib \
    libedit-dev clang llvm-dev python3-dev \
    python3  python3-numpy python3-matplotlib \
    python3-setuptools

Create a directory for your installation, e.g., $HOME/src

mkdir -p ~/src
cd src
mkdir -p gimli
cd gimli

Checkout the current sources for libgimli:

git clone https://github.com/gimli-org/gimli.git

We use cmake for compilation. We recommend using a build directory parallel to the gimli (trunk) path:

mkdir -p build

The main directory structure should looks like this:

gimli/gimli
gimli/build

Change to the build path

cd build

and configure the build for Python 2.7 with:

cmake ../gimli

If you want to compile for Python 3.5, alternatively use:

cmake ../gimli -DPYVERSION=3.5

If the output complains about missing dependencies, install these and repeat the the last step. To build the library just run make.

make

To speed up the build process using more CPUs, use the -j flag, e.g.:

make -j 8

The libraries will be installed in build/lib and some test applications are installed in build/bin. If you want to build the Python bindings, call:

make pygimli

You might add J=8 (make pygimli J=8) for using 8 jobs in parallel to speed up the build (adapt this to the number of real cores of the computer). The library _pygimli_.so library will be copied into the source path ../gimli/python/pygimli in the subdirectory core.

To use the gimli installation you need to set some environment variables (this example assumes that the src directory resides in your home directory):

export PYTHONPATH=$PYTHONPATH:$HOME/src/gimli/gimli/python
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:$HOME/src/gimli/build/lib
export PATH=$PATH:$HOME/src/gimli/build/bin

If you want to use the C++ command line applications, call

make apps

Compiled binaries will be written to build/bin.

You can do a quick test of the pygimli build and installation with:

python -c 'import pygimli as pg; print(pg.__version__)'

You can test your gimli build with:

make check

Note that the test will be very silent if you don't have cppunit installed.

If you install pytest with

sudo apt-get install python-pytest python3-pytest

then you can run the internal test suite with

python -c "import pygimli; pygimli.test()"

Using Docker to build in Debian stretch (for advanced users only!)

If you want to use a Docker container to build (and possibly use) pyGIMLi, you can use the Dockerfile found in the scripts/ subdirectory named Dockerfile_DebianStretch. Please refer to the file for further instructions.

Example Installation on Ubuntu

sudo apt-get install libc-dev subversion git cmake mercurial
sudo apt-get install libboost-all-dev libblas-dev liblapack-dev libedit-dev
sudo apt-get install python3-dev python3-matplotlib python3-numpy

mkdir -p ~/src/gimli
cd ~/src/gimli
git clone https://github.com/gimli-org/gimli.git

mkdir -p build
cd build
cmake ../gimli
make -j 4 gimli
make pygimli J=4

Troubleshooting

If you experience runtime problems on starting pygimli like:

ImportError: /usr/lib/libboost_python.so: undefined symbol: PyClass_Type

It may happen that CMake estimates the wrong libboost_python version by choosing py2 version instead of py3. You can force cmake to select the correct version with:

cmake ../gimli -DBoost_PYTHON_LIBRARY=/usr/lib64/libboost_python3.so

If the build misses libedit:

/usr/bin/ld: cannot find -ledit

Install libedit, e.g. 'apt-get install libedit' on Debian/Ubuntu.

castXML

castXML (https://github.com/CastXML/CastXML/) is needed to generate the code for the python bindings. Some systems provide castxml binary so the build system should detect it if installed. As fallback solution the build system tries to install castxml binaries or try to compile there own if the binaries don't work. You can enforce the local binary installation with:

cmake ../../src/castXML/ -DCASTXML_LOCAL=1
make

or the local binary compilation with:

cmake ../../src/castXML/ -DCASTXML_LOCALSRC=1
make

If castXML build complains about missing clang or llvm command, please go into $(GIMLISRC)/../thirdParty/build-XXX-XXX/castXML and try configure and build cmake manually

CC=clang-3.6 CXX=clang++-3.6 cmake ../../src/castXML/
make

If you build castXML manually you can provide this binary to cmake via

cmake ../gimli -DCASTER_EXECUTABLE=$(PATH_TO_CASTXML)

Useful cmake settings

You can rebuild and update all local generated third party software by setting the CLEAN environment variable:

CLEAN=1 cmake ../gimli

Use alternative c++ compiler.

CC=clang CXX=clang++ cmake ../gimli

Define alternative python version. On default the version of your active python version will be chosen. You will need numpy and boost-python builds with your desired python version.

cmake ../gimli -DPYVERSION=3.3

Build the library with debug and profiling flags

cmake ../gimli -DCMAKE_BUILD_TYPE=Debug

Build the library with gcc build.in sanity check

cmake ../gimli -DCMAKE_BUILD_TYPE=Debug -DASAN=1

Useful make commands

More verbose build output to view the complete command line:

make VERBOSE=1