Issue With Cardinal Installation (Potentially Due to OpenMC Dependencies)

[jcstonehill]

Hello friends!

I'm running into an issue installing Cardinal on WSL (Ubuntu 24.04.1). It appears to be something related to the dependencies of OpenMC (specifically FMT), but I'm certainly no expert in compiling codes, etc. I have looked for similar threads but either there was no solution available or I tried the solution and it didn't work.

The error I'm getting:

/home/stonehill/cardinal/install/include/openmc/position.h:8:10: fatal error: fmt/format.h: No such file or directory 8 | #include "fmt/format.h" | ^~~~~~~~~~~~~~ compilation terminated. make: *** [/home/stonehill/cardinal/contrib/moose/framework/build.mk:150: /home/stonehill/cardinal/build/unity_src/postprocessors_Unity.x86_64-conda-linux-gnu.opt.lo] Error 1

My suspicion is something isn't going correctly in terms of moving FMT and other dependencies from the contrib/openmc/vendor/ folder (where the I can see the dependencies have been downloaded) to the cardinal/build/openmc/vendor folder (where there is only a Catch2 folder, no FMT folder).

Below are the series of commands I executed, and I have attached my console output.

conda create -n moose moose-dev
conda activate moose
git clone https://github.com/neams-th-coe/cardinal.git
cd cardinal
./scripts/get-dependencies.sh
export ENABLE_NEK=false
export HDF5_ROOT=$CONDA_PREFIX
export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
make -j2 >> out.txt 2>&1

out.txt

Thanks ahead of time for any help trying to debug this. Very excited to get started in Cardinal!

[jcstonehill]

Update! I completely re-installed Ubuntu and then follow the procedure I outlined above. Cardinal compiled on the first try! So, that's great. However, I have another machine where it would be much more difficult for me to re-install the OS. So, I'd still like to hear opinions on this problem if anyone has any?

My theory is that there must be something in the usr/bin that was affecting the install process? I had previously installed standalone OpenMC from Conda (openmc-dev) so I wonder if there were remnants of that? I'm not sure!

[jcstonehill]

I have identified the issue!

When Cardinal tried to build OpenMC, it uses CMake. Looking at the CMakeLists.txt file (cardinal/contrib/openmc/CMakeLists.txt), it was looking for libraries using the "find_package" function of CMake. It uses this methodology for libraries such as fmt, xtensor, xtl, and Catch2.

Here's a code snippet to illustrate the format in CMakeLists.txt:

#===============================================================================
# {fmt} library
#===============================================================================
find_package_write_status(fmt)
if (NOT fmt_FOUND)
     set(FMT_INSTALL ON CACHE BOOL "Generate the install target.")
     add_subdirectory(vendor/fmt)
endif()

The CMake find_package function found an fmt installation on my system, check this line from the out.txt file above:
-- Found fmt: /usr/local/lib/cmake/fmt (version 8.0.1)

Thus, it didn't look for the dependency in cardinal/contrib/openmc/vendor folder and instead used the installation on my system (a much older version).

I worked around this by editing the CMakeLists.txt file to not search for a package and instead pulled out all of the lines inside of the if statement. The code snippet above became this:

#===============================================================================
# {fmt} library
#===============================================================================
set(FMT_INSTALL ON CACHE BOOL "Generate the install target.")
add_subdirectory(vendor/fmt)

#find_package_write_status(fmt)
#if (NOT fmt_FOUND)
#  set(FMT_INSTALL ON CACHE BOOL "Generate the install target.")
#  add_subdirectory(vendor/fmt)
#endif()

I repeated this process for all of the relevant dependencies in the CMakeLists.txt file.

Obviously, this is not a very clean solution but with my limited knowledge of linux / compiled languages this is what I was able to get to work! If anyone has a cleaner way to force the compiler to only look in the cardinal folder for the OpenMC dependencies, please let me know so I can learn! Thanks!

[aprilnovak]

Hi @jcstonehill, I'm glad the issue was able to be resolved! Unfortunately this is a byproduct of the fact that MOOSE (and hence Cardinal) uses the Make build system, but OpenMC/NekRS/DAGMC use CMake. This deficiency is something on our to-do list to see if we can resolve (i.e. allow Make to inform CMake that fmt, eigen, etc. have been found on a user's system). For now, the fix you've identified is what I would have recommended!

[aprilnovak]

Hi @jcstonehill, @nglaser3 has added a fix into Cardinal which we think solves this issue. Would you be willing to help us test it on your machine to confirm if this solves this problem before we merge this in? (I'm not able to reproduce it on my system so we are trying to find volunteers who have run into this in the past).

To test, you'd basically just repeat the same steps to compile Cardinal that you did before, but use @nglaser3's branch. This could be done in a separate directory on your machine, for instance.

cd somewhere_else
git clone https://github.com/nglaser3/cardinal.git
cd cardinal
git switch build-update
./scripts/get-dependencies.sh

# the rest of the steps to build cardinal, but don't do anything custom in CMakeLists.txt

Thank you!

[jcstonehill]

I commented on the pull request. I followed your instructions above and was able to compile on the cluster where I was experiencing my issues previously without manipulating the CMakeLists.txt file. Looks like it solved all my problems!