Neko  0.8.0-rc2
A portable framework for high-order spectral element flow simulations
Testing

pFUnit

Neko uses the software pFUnit for unit testing. To install the software you can use the following commands, in which you should just set the desired installation path

git clone https://github.com/Goddard-Fortran-Ecosystem/pFUnit.git -b v4.4.2
cd pFUnit && mkdir b && cd b && cmake -DCMAKE_INSTALL_PREFIX=/pfunit_install_path .. && make -j$(nproc) && make install

You will now have a dirctory called PFUNIT-4.4 in your intall path.

Configuring Neko

To use pFUnit with Neko, we need to specify its location during the configure phase of the build. The following configuration is given here as an example, where again you need to provide the path

./configure FC=${FC} FCFLAGS="-O2 -pedantic -std=f2008" --with-pfunit=/pfunit_install_path/PFUNIT-4.4

Make sure that in the output you see it says pFUnit ... yes at some point.

Running the tests

To run the tests you should execute

make check

This will first compile the software and the tests, and then execute them. In the end you should get a nice list of tests that were run and a summary, which hopefully says that all tests have passed!

Adding a new test

The first step is of course to read the documentation for pFUnit to understand how it works. It is up to you to learn how to write .pf files. Here, we only cover how to incorporate the test into the Neko build system.

The tests are located in the tests folder. Unlike some other software, the structure of the tests folder does not follow that of src. Instead each folder loosely corresponds to a single class or sometimes a single abstract class and its children (see the stack folder, for example).

We now go through the steps of adding a test. The instructions will differ somewhat depending on whether you test uses MPI or not.

  1. In Add a new folder in tests, corresponding to the classes that will be tested.
  2. Put the .pf files you've written inside that folder.
  3. Copy over a Makefile.in file from some over test folder. If your test does not use MPI, you can copy from e.g. hex, otherwise take it from field. As usual, the Makefile.in will turn into a Makefile during the configure phase of the build.
  4. The Makefile.in looks very much like the Makefile examples in the documentation of pFUnit.
    1. The check target should be changed to the name of your test. If yor test does not use MPI, name it the same as the name of the folder + _test. If you need MPI, then don't use test, but rather something else, like _suite.
    2. The lines defining _TESTS, _OTHER_LIBRARIES, and the other pFUnit stuff should be prefixed with the name selected above. Similarly with the eval statement.
    3. In the clean target, also put the same name as in check.
  5. If you use MPI, you now need to create a runner script. Copy field/field_test tp start with, which looks like this
    #!/bin/sh
    if which mpirun >/dev/null; then
    mpirun -np 1 ./field/field_suite
    else
    mpiexec -np 1 ./field/field_suite
    fi
    You need to change ./field/field_suite to the name of the directory for your test and the name of the check target in the Makefile.in.
  6. Now you have to edit tests/Makefile.am. This file contains three lists.
    1. The first one is SUBDIRS, to which you should add the directory with your tests.
    2. The second one is TESTS, here you should add the file, which ends with _test. If you followed the conventions above it will be the output from pFUnit for a simple test, or the runner you created in the case of an MPI test.
    3. The last list is EXTRA_DIST. Here, you should add all your .pf files. That is it for a simple test, but if you use MPI you should also copy the entry you made in TESTS to this list. Again, you can use field_test as an example.
  7. Finally, open the configure.ac file in the root folder of neko. Find # Config tests line and the AC_CONFIG_FILES list below it. Add the path to a Makefile (note, no .in!!) in your test folder.
  8. Please also add the compilation products of your test to the tests/.gitignore file. This helps keeps the version controlled file list unpolluted.