Quick Start for Developers

Unix Shell

Currently, compass only supports bash and related unix shells (such as ksh on the Mac). We do not support csh, tcsh or other variants of csh. An activation script for those shells will not be created.

If you normally use csh, tcsh or similar, you will need to temporarily switch to bash by calling /bin/bash each time you want to use compass.

Set up a compass repository: for beginners

To begin, obtain the master branch of the compass repository with:

git clone git@github.com:MPAS-Dev/compass.git
cd compass
git submodule update --init --recursive

The E3SM repository and a clone of E3SM for MALI development are submodules of the compass repository.

compass conda environment, compilers and system modules

As a developer, you will need your own environment with the latest dependencies for compass and a development installation of compass from the branch you’re working on.

The conda directory in the repository has a tool configure_compass_env.py that can get you started.

You will need to run ./conda/configure_compass_env.py each time you check out a new branch or create a new worktree with git. Typically, you will not need to run this command when you make changes to files within the compass python package. These will automatically be recognized because compass is installed into the conda environment in “editable” mode. You will need to run the command if you add new code files or data files to the package because these don’t get added automatically.

Whether you are on one of the Supported Machines or an “unknown” machine, you will need to specify a path where Miniconda3 either has already been installed or where the script can install it. You must have write permission in the base environment.

Note

It is very important that you not use a shared installation of Miniconda3 such as the base environment for E3SM-Unified for compass development. Most developers will not have write access to shared environments, meaning that you will get write-permission errors when you try to update the base environment or create the compass development environment.

For anyone who does have write permission to a shared environment, you would be creating your compass development environment in a shared space, which could cause confusion.

Please use your own personal installation of Miniconda3 for development, letting configure_compass_env.py download and install Miniconda3 for you if you don’t already have it installed.

Supported machines

If you are on one of the Supported Machines, run:

./conda/configure_compass_env.py --conda <base_path_to_install_or_update_conda> \
    -c <compiler>

The <base_path_to_install_or_update_conda> is typically ~/miniconda3. This is the location where you would like to install Miniconda3 or where it is already installed. If you have limited space in your home directory, you may want to give another path. If you already have it installed, that path will be used to add (or update) the compass test environment.

See the machine under Supported Machines for a list of available compilers to pass to -c. If you don’t supply a compiler, you will get the default one for that machine (usually Intel). Typically, you will want the default MPI flavor that compass has defined for each compiler, so you should not need to specify which MPI version to use but you may do so with -i if you need to.

If you are on a login node, the script should automatically recognize what machine you are on. You can supply the machine name with -m <machine> if you run into trouble with the automatic recognition (e.g. if you’re setting up the environment on a compute node, which is not recommended).

Unknown machines

If your are on an “unknown” machine, typically a Mac or Linux laptop or workstation, you will need to specify which flavor of MPI you want to use (mpich or openmpi):

./conda/configure_compass_env.py --conda <conda_path> --mpi <mpi>

Again, the <conda_path> is typically ~/miniconda3, and is the location where you would like to install Miniconda3 or where it is already installed. If you already have it installed, that path will be used to add (or update) the compass test environment.

We only support one set of compilers for Mac and Linux (gnu for Linux and clang with gfortran for Mac), so there is no need to specify them. See Other Machines for more details.

In addition, unknown machines require a config file to be specified when setting up the compass test environment. A config file can be specified using -f <filename>, where <filename> is an absolute or relative path to the file. More information, including example config files, can be found in Config Files.

What the script does

In addition to installing Miniconda and creating the conda environment for you, this script will also:

  • install the compass package from the local branch in “development” mode so changes you make to the repo are immediately reflected in the conda environment.

  • build the SCORPIO library if it hasn’t already been built. SCORPIO is needed building and running MPAS components.

  • build the ESMF library if it hasn’t already been built. ESMF with the system’s version of MPI is needed for making mapping files.

  • make an activation script called load_dev_compass_<version>_<machine>_<compiler>_<mpi>.sh, where <version> is the compass version, <machine> is the name of the machine (to prevent confusion when running from the same branch on multiple machines), <compiler> is the compiler name (e.g. intel or gnu), and mpi is the MPI flavor (e.g. impi, mvapich, openmpi).

  • optionally (with the --check flag), run some tests to make sure some of the expected packages are available.

Optional flags

--check

Check to make sure expected commands are present

--python

Select a particular python version (the default is currently 3.8)

--env-name

Set the name of the environment (and the prefix for the activation script) to something other than the default (dev_compass_<version> or dev_compass_<version>_<mpi>).

Activating the environment

Each time you want to work with compass, you will need to run:

source ./load_dev_compass_<version>_<machine>_<compiler>_<mpi>.sh

This will load the appropriate conda environment, load system modules for compilers, MPI and libraries needed to build and run MPAS components, and set environment variables needed for MPAS or compass. It will also set an environment variable LOAD_COMPASS_ENV that points to the activation script. compass uses this to make an symlink to the activation script called load_compass_env.sh in the work directory.

If you switch between different compass branches, it is safest to rerun ./conda/configure_compass_env.py with the same arguments as above to make sure dependencies are up to date and the compass package points to the current directory. If you are certain that no compass dependencies are different between branches, you can also simply source the activation script (load_dev_compass*.sh) in the branch.

Once you have sourced the activation script, you can run compass commands anywhere, and it always refers to that branch. To find out which branch you are actually running compass from, you should run:

echo $LOAD_COMPASS_ENV

This will give you the path to the load script, which will also tell you where the branch is. If you do not use the worktree approach, you will also need to check what branch you are currently on with git log, git branch or a similar command.

Note

If you switch branches and do not remember to recreate the conda environment (./conda/configure_compass_env.py) or at least source the activation script (load_dev_compass*.sh), you are likely to end up with an incorrect and possibly unusable compass package in your conda environment.

If you switch to another branch, you will need to rerun ./conda/configure_compass_env.py with the same arguments as above to make sure dependencies are up to date and the compass package points to the current directory.

Note

With the conda environment activated, you can switch branches and update just the compass package with:

python -m pip install -e .

The activation script will do this automatically when you source it in the root directory of your compass branch. This is substantially faster than rerunning ./conda/configure_compass_env.py ... but risks dependencies being out of date. Since dependencies change fairly rarely, this will usually be safe.

If you wish to work with another compiler, simply rerun the script with a new compiler name and an activation script will be produced. You can then source either activation script to get the same conda environment but with different compilers and related modules. Make sure you are careful to set up compass by pointing to a version of the MPAS model that was compiled with the correct compiler.

Troubleshooting

If you run into trouble with the environment or just want a clean start, you can run:

./conda/configure_compass_env.py --conda <conda_path> -c <compiler> --recreate

The --recreate flag will delete the conda environment and create it from scratch. This takes just a little extra time.

Creating/updating only the compass environment

For some workflows (e.g. for MALI development wih the Albany library), you may only want to create the conda environment and not build SCORPIO, ESMF or include any system modules or environment variables in your activation script. In such cases, run with the --env_only flag:

./conda/configure_compass_env.py --conda <conda_path> --env_only

Each time you want to work with compass, you will need to run:

source ./load_dev_compass_<version>.sh

This will load the appropriate conda environment for compass. It will also set an environment variable LOAD_COMPASS_ENV that points to the activation script. compass uses this to make a symlink to the activation script called load_compass_env.sh in the work directory.

If you switch to another branch, you will need to rerun:

./conda/configure_compass_env.py --conda <conda_path> --env_only

to make sure dependencies are up to date and the compass package points to the current directory.

Note

With the conda environment activated, you can switch branches and update just the compass package with:

python -m pip install -e .

This will be substantially faster than rerunning ./conda/configure_compass_env.py ... but at the risk that dependencies are not up-to-date. Since dependencies change fairly rarely, this will usually be safe.

Building MPAS components

The MPAS repository is a submodule of the compass repository. For example, to compile MPAS-Ocean:

source ./load_dev_compass_<version>_<machine>_<compiler>_<mpi>.sh
cd E3SM-Project/components/mpas-ocean/
make <mpas_make_target>

MALI can be compiled with or without the Albany library that contains the first-order velocity solver. The Albany first-order velocity solver is the only velocity option that is scientifically validated, but the Albany library is only installed on Badger, Grizzly, and Cori. Therefore, in some situations it is desirable to compile without Albany to run basic tests on platforms where Albany is not available. This basic mode of MALI can be compiled similarly to MPAS-Ocean, i.e.:

source ./load_dev_compass_<version>_<machine>_<compiler>_<mpi>.sh
cd MALI-Dev/components/mpas-albany-landice
make <mpas_make_target>

Compiling MALI with Albany has not yet been standardized. Some information is available at https://github.com/MALI-Dev/E3SM/wiki, and complete instructions will be added here in the future.

See the last column of the table in Supported Machines for the right <mpas_make_target> command for each machine and compiler.

Running compass from the repo

If you follow the procedure above, you can run compass with the compass command-line tool exactly like described in the User’s Guide Quick Start for Users and as detailed in Command-line interface.

To list test cases you need to run:

compass list

The results will be the same as described in Setting up test cases, but the test cases will come from the local compass directory.

To set up a test case, you will run something like:

compass setup -t ocean/global_ocean/QU240/mesh -m $MACHINE -w $WORKDIR -p $MPAS

where $MACHINE is an ES3M machine, $WORKDIR is the location where compass test cases will be set up and $MPAS is the directory where the MPAS model executable has been compiled. See compass setup for details.

To list available test suites, you would run:

compass list --suites

And you would set up a suite as follows:

compass suite -s -c ocean -t nightly -m $MACHINE -w $WORKDIR -p $MPAS

When you want to run the code, go to the work directory (for the suite or test case), log onto a compute node (if on an HPC machine) and run:

source load_compass_env.sh
compass run

The first command will source the same activation script (load_dev_compass_<version>_<machine>_<compiler>_<mpi>.sh) that you used to set up the suite or test case (load_compass_env.sh is just a symlink to that activation script you sourced before setting up the suite or test case).

Set up a compass repository with worktrees: for advanced users

This section uses git worktree, which provides more flexibility but is more complicated. See the beginner section above for the simpler version. In the worktree version, you will have many unix directories, and each corresponds to a git branch. It is easier to keep track of, and easier to work with many branches at once. Begin where you keep your repositories:

mkdir compass
cd compass
git clone git@github.com:MPAS-Dev/compass.git master
cd master

The MPAS-Dev/compass repo is now origin. You can add more remotes. For example:

git remote add mark-petersen git@github.com:mark-petersen/compass.git
git fetch mark-petersen

To view all your remotes:

git remote -v

To view all available branches, both local and remote:

git branch -a

We will use the git worktree command to create a new local branch in its own unix directory:

cd compass/master
git worktree add -b new_branch_name ../new_branch_name origin/master
cd ../new_branch_name

In this example, we branched off origin/master, but you could start from any branch, which is specified by the last git worktree argument.

There are two ways to build the MPAS executable:

  1. Compass submodule (easier): This guarantees that the MPAS commit matches compass. It is also the default location for finding the MPAS model so you don’t need to specify the -p flag at the command line or put the MPAS model path in your config file (if you even need a config file at all):

    git submodule update --init --recursive
    cd E3SM-Project/components/mpas-ocean/
    # load modules
    make gfortran
    

    For the “load modules” step, see Machines for specific instructions.

  2. Other E3SM directory (advanced): Create your own clone of the E3SM-Project/E3SM or MALI-Dev/E3SM repository elsewhere on disk. Either make an ocean.cfg or landice.cfg that specifies the absolute path to the path where the ocean_model or landice_model executable is found, or specify this path on the command line with -p. You are responsible for knowing if this particular version of MPAS component’s code is compatible with the version of compass that you are using. The simplest way to set up a new repo for MALI development in a new directory is:

    git clone git@github.com:MALI-Dev/E3SM.git your_new_branch
    cd your_new_branch
    git checkout -b your_new_branch origin/develop
    

    The equivalent for MPAS-Ocean development would be:

    git clone git@github.com:E3SM-Project/E3SM.git your_new_branch
    cd your_new_branch
    git checkout -b your_new_branch origin/master