This section is out of date!! A further review will be completed in the near future. Sorry for the inconvenience.

More on Testing

MPAS-JEDI comes with an extensive suite of ctests. These tests are designed to ensure that as much as possible of the source code is regularly exercised, and that generic applications used in larger scale experiments are tested. Testing is an integral aspect of the development and experiment process. It is recommended that the tests are run every time the code is built. Developers working on a new feature in the MPAS-JEDI repository should ensure that all existing tests pass before submitting a pull request to github. We also request that some attempt is made to ensure that new code is exercised by an existing test or a new test. There are exceptions for code that is exercised extensively and continually in cycling experiments with verification, and for diagnostic tools that are not automatically tested yet.

Continuous integration

MPAS-JEDI is instrumented with a continuous integration (CI) suite running on Amazon Web Services. Each time a pull request is issued against the develop branch in the JCSDA-internal repository, the MPAS-JEDI part of the MPAS-BUNDLE package is built with Intel, GNU, and Clang compilers. Then all the MPAS-JEDI ctests are executed. A failure from any of the ctests blocks the pull request from being merged. At this time, MPAS-JEDI is not instrumented with a code coverage report.

Adding a test to MPAS-JEDI

When new codes that prove concepts or implement new features are added to MPAS-JEDI and MPAS-Model repositories used in the MPAS-BUNDLE, corresponding ctests should be added into the standard ctest set. This ensures that future modifications to either of those two repositories, or to the repositories on which MPAS-JEDI depends (i.e., OOPS, SABER, IODA, UFO, CRTM), do not break existing functionalities that are critical to users’ scientific experiments.

All the ctesting in MPAS-JEDI is controlled through mpas-jedi/test/CMakeLists.txt. A ctest may be either a unit test, which exercises an individual method in a given class, or an application test that executes a generic application. Benchmark results are provided accompanying the ctests. A unit ctest contains results in a reference log file based on analytical solutions or accurate numerical studies. Each application ctest has an associated reference based on a previous execution of the same test. To determine the pass or failure for a ctest, the actual output is compared against the reference log file within a prescribed small tolerance.

To simplify adding tests to MPAS-JEDI, two macro functions are provided — add_mpasjedi_unit_test adds a new unit test and add_mpasjedi_application_test adds a new application test. The reader is referred to mpas-jedi/test/CMakeLists.txt, where numerous examples exist for both. Note that the name of the yaml and reference files must match the name of the ctest, e.g., test_mpasjedi_forecast uses the configuration stored in mpas-jedi/test/testinput/forecast.yaml and is compared to the reference stored in mpas-jedi/test/testoutput/forecast.ref.

If a PR made to one of the repositories used by MPAS-BUNDLE causes the reference values of many tests to change, it is useful to use the RECALIBRATE_CTEST_REFS option in mpas-jedi/test/CMakeLists.txt.

Additional automated testing on Derecho

There are two additional testing mechanisms in place on Derecho that provide automated test coverage.

(1) A daily cron job builds MPAS-JEDI from JCSDA-internal/mpas-bundle::develop branch, then runs the standard ctest suite. To keep MPAS-JEDI up to date with the latest development from JEDI infrastructure, the develop branch from JEDI-core repos, including OOPS, SABER, IODA, UFO and CRTM, are checked out to build the code. This allows us to promptly identify any changes in upstream repositories that break MPAS-BUNDLE.

(2) A weekly cron job builds the JCSDA-internal/mpas-bundle::develop branch, then runs a 6-day 120-km resolution cycling DA experiment. The experiment uses 3DEnVar to assimilate conventional observations (sondes, aircraft, gnssro refractivity, satwind, surface pressure) and AMSUA clear-sky radiances (aqua, noaa-15, noaa-18, noaa-19, and metop-a). The results are automatically analyzed for statistical comparison to GFS analyses. This test ensures that the MPAS-BUNDLE performance does not diverge far from a benchmark.