We use pFUnit for a software that simulates the airflow behind helicopter's rotor blades based upon a vortex method.
The software is parallelized with MPI, OpenMP and OpenACC (for GPUs).
As it is based on existing older (legacy) code, we can only add unit tests for newly developed stuff, unfortunately.
We have a nice build system based on CMake, and pFUnit is integrated smoothly using CMake's "ExternalProject" macro.
The tests are run in parallel using MPI (without setting the npes parameter),
but except for some parametrized tests we don't use many of the advanced features of pFUnit.
In my opinion, the implementation overhead for the parametrized tests (derived types for parameter and testcase) is still a minor drawback if you just want to 5 different integers.
We have a CI-server (Jenkins) that builds the software and runs the tests after every change/commit.
For the xml-output of pFUnit we added a minor patch to make it compatible with Jenkin's interpretation of JUnit xml.
This way, we obtain a nice overview of the test results in our CI web interface or as mail.
I could not unit-test my OpenACC parallelization because of bugs in the PGI compiler: I reported about 10+ bugs in various Fortran compilers over the last year(s) - most of them fixed by now and approximately half of them triggered by pFUnit.
We currently plan a new software where we also want to use pFUnit. Starting from scratch should allow us to directly maintain extensive unit tests for all the code and to apply a more test-driven development workflow.
Additional remark:
For C/C++ projects we use a patched version of googletest for parallel unit-tests with MPI, which provides a very powerfull environment (we easily end up with hundreds to thousends of tests based upon templated and parametrized test classes, such that in the end we have about 100.000 test results for a single CI-build - for this you really need a nice hierarchical overview of the results (like in Jenkins)).
So, I'm eager to see the development of the tests in our new Fortran+pFUnit-software.
Thanks for the nice work on pFUnit!
Melven
Simulation and Software Technology
German Aerospace Center (DLR)
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
We use pFUnit for a software that simulates the airflow behind helicopter's rotor blades based upon a vortex method.
The software is parallelized with MPI, OpenMP and OpenACC (for GPUs).
As it is based on existing older (legacy) code, we can only add unit tests for newly developed stuff, unfortunately.
We have a nice build system based on CMake, and pFUnit is integrated smoothly using CMake's "ExternalProject" macro.
The tests are run in parallel using MPI (without setting the npes parameter),
but except for some parametrized tests we don't use many of the advanced features of pFUnit.
In my opinion, the implementation overhead for the parametrized tests (derived types for parameter and testcase) is still a minor drawback if you just want to 5 different integers.
We have a CI-server (Jenkins) that builds the software and runs the tests after every change/commit.
For the xml-output of pFUnit we added a minor patch to make it compatible with Jenkin's interpretation of JUnit xml.
This way, we obtain a nice overview of the test results in our CI web interface or as mail.
I could not unit-test my OpenACC parallelization because of bugs in the PGI compiler: I reported about 10+ bugs in various Fortran compilers over the last year(s) - most of them fixed by now and approximately half of them triggered by pFUnit.
We currently plan a new software where we also want to use pFUnit. Starting from scratch should allow us to directly maintain extensive unit tests for all the code and to apply a more test-driven development workflow.
Additional remark:
For C/C++ projects we use a patched version of googletest for parallel unit-tests with MPI, which provides a very powerfull environment (we easily end up with hundreds to thousends of tests based upon templated and parametrized test classes, such that in the end we have about 100.000 test results for a single CI-build - for this you really need a nice hierarchical overview of the results (like in Jenkins)).
So, I'm eager to see the development of the tests in our new Fortran+pFUnit-software.
Thanks for the nice work on pFUnit!
Melven
Simulation and Software Technology
German Aerospace Center (DLR)