For those wishing to build and run the UFO bundle on a standalone Linux PC outside of a container
- Download ecbuild from: https://github.com/ecmwf/ecbuild
- Install it in a place mentioned in $PATH
- Clone the ufo bundle with your github username: % git clone https://<username>@github.com/JCSDA/ufo-bundle.git
- As of this writing (7/17/2018), an example build environment for Linux PC exists only in a feature branch, though hopefully it'll be merged to the development branch soon. For now, switch branches like this after cd-ing into the ufo bundle: % git checkout feature/buildtools
- cd into the tools/ directory and edit the file module_setup_linuxpc_gcc_openmpi.sh The naming of this file indicates that it is for a Linux PC using the GCC compiler suite and openmpi MPI library. You'll need to ensure that gcc, gcc-fortran, and openmpi are installed before proceeding. If you wish to use a different MPI stack that shouldn't be too hard, but may require some changes to, and renaming of, the module_setup* file.
- The first 6 "export" entries in module_setup_linuxpc_gcc_openmpi.sh indicate installation locations for packages required by the UFO bundle. The following steps outline what I did for a successful build on my Ubuntu/LinuxMint PC:
- For eigen: % sudo apt-get install libeigen3-dev
- For netcdf: % sudo apt-get install libnetcdf-c++4-dev; sudo apt-get install libnetcdff-dev
- For CXX_COMPILER and C_COMPILER and Fortran_COMPILER: %sudo apt-get install gcc-5-base; sudo apt-get install libgfortran5
- For boost: I was not able to get a package install approach to work, so instead built it from scratch. Rev. 1.67.* or greater should work. I won't kid you: getting this monster to build and install correctly is not simple. Unfortunately I didn't take notes on what I did to get the install to work, or I'd have included them here.
- Still in the tools/ directory, run the script build.sh This script was originally designed to simplify the build process on a few known machines, and has been augmented here for a generic Linux PC. The script takes 4 mandatory arguments and one optional. The first is hostname (linuxpc), the second is compiler suite (gcc), the third is MPI distribution (for me this was openmpi), the fourth can be either debug or release. The fifth optional argument is the number of threads to allow "make" to use. This can speed up the build process substantially. I generally set it to the number of processors available on the machine on which the build is being done. So, the following works for me:
% build.sh linuxpc gcc openmpi release 4
- After a successful build, you can run the full test suite by doing the following:
- % source module_setup_linuxpc_gcc_openmpi.sh (For csh users: % source module_setup_linuxpc_gcc_openmpi.csh)
- Run the test suite: % cd ..; ctest