NEST module generated from NESTML fails to build on macOS when NEST is installed from conda-forge

[pnbabu]

Describe the bug
After installing NEST through conda-forge and installing NESTML via pip in the same conda environment, the generated NEST module from NESTML fails to compile on macOS with the following error:

clang: error: unsupported argument 'libomp' to option 'fopenmp='
make[2]: *** [CMakeFiles/nestmlmodule_lib.dir/nestmlmodule.o] Error 1
make[1]: *** [CMakeFiles/nestmlmodule_lib.dir/all] Error 2
make: *** [all] Error 2

To Reproduce
Steps to reproduce the behavior: https://nestml.readthedocs.io/en/latest/installation.html#anaconda-installation

conda create --name nestml_conda
conda activate nestml_conda
conda install -c conda-forge nest-simulator ipython cxx-compiler pyqt wxpython
pip install nestml --upgrade

The C++ path and the $PYTHONPATH are updated as mentioned in the above link. Then running the below command gives the compilation error.

nestml --input_path models/neurons/iaf_psc_exp.nestml --target_path target_iaf --logging_level INFO

Desktop/Environment (please complete the following information):

• OS: macOS Monterey 12.6
• Python-Version: Python 3.10
• NEST-Version: 3.3
• Installation: NEST installed through conda-forge

Additional context
I suspect the error is because $HOME/miniconda3/envs/nestml_conda/lib/libomp.dylib is not linked during the compilation. The library is also not listed under $HOME/miniconda3/envs/nestml_conda/bin/nest-config --libs. Although, manually adding it to the nest-config file doesn't resolve the error.

[heplesser]

The problem is that nest-config --libs from the conda-forge build returns ... -fopenmp=libomp ..., while it should contain ... -Xclang -fopenmp .... If NEST 3.3 is built locally with clang, this works out. There are some other problems with the libs provide, which have been fixed by #2468 (not available via conda yet).

@steffengraber Is it possible to get the output from the cmake run for the conda builds? Maybe that can help to find out why we get the wrong flags.

[steffengraber]

Here are the complete build logs for osx_64 and osx_arm64:

osx_64 python3.10
https://dev.azure.com/conda-forge/84710dde-1620-425b-80d0-4cf5baca359d/_apis/build/builds/566950/logs/50

osx_arm64 python3.10
https://dev.azure.com/conda-forge/84710dde-1620-425b-80d0-4cf5baca359d/_apis/build/builds/566950/logs/49

[heplesser]

Should disappear with wheel-based installation.

[github-actions]

Issue automatically marked stale!

[steffengraber]

I hope the problem will be solved with the conda package for NEST 3.4.

[terhorstd]

Did the 3.4_rc1 solve this?

[steffengraber]

It is hard for me to test this in macOS. With the new version, the compiler is added to the run environment and the path inside nest-config is overwritten. This should fix the problem, also in macOS.

[nicolossus]

I did a conda installation of 3.4_rc1 on macOS, and the NESTML model still fails to compile with the same error:

clang: error: unsupported argument 'libomp' to option 'fopenmp='
make[2]: *** [CMakeFiles/nestmlmodule_module.dir/build.make:76: CMakeFiles/nestmlmodule_module.dir/nestmlmodule.o] Error 1
make[1]: *** [CMakeFiles/Makefile2:112: CMakeFiles/nestmlmodule_module.dir/all] Error 2
make: *** [Makefile:156: all] Error 2

However, if I change -fopenmp=libomp to -Xclang -fopenmp under both --cflags and --libs in the nest-config file, the NESTML model compiles successfully.

The NESTML documentation on Anaconda installation (https://nestml.readthedocs.io/en/latest/installation.html#anaconda-installation) already says one has to:

Edit nest-config and correct the entry under --compiler with the output returned by which c++

Perhaps a temporary workaround could be for the documentation to instruct macOS users to also replace -fopenmp=libomp with -Xclang -fopenmp in the nest-config.

Steps to reproduce:

Download 3.4_rc1 feedstock build conda_artifacts_20230213.5.1_osx_64_python3.10.____cpython and unpack the file and the archive directory within (I also renamed the archive directory of this particular file to build_artifacts).

Create conda environment and install:

conda create -n nestml_conda python=3.10
conda install -c conda-forge c-compiler compilers cxx-compiler make cmake zlib boost cython gsl libtool matplotlib-base readline ipython scipy statsmodels openssh pyqt wxpython
conda install --use-local build_artifacts/osx-64/nest-simulator-3.4_rc1-py310h69b982f_0.conda
pip install nestml --upgrade

Then replace --compiler entry in ~/opt/anaconda3/envs/nestml_conda/bin/nest-config with output from which c++ and -fopenmp=libomp to -Xclang -fopenmp under --cflags and --libs.

Finally, compile the NESTML model from the NESTML source directory:

$ nestml --input_path models/neurons/iaf_psc_exp.nestml --target_path target_iaf --logging_level INFO

[github-actions]

Issue automatically marked stale!

[steffengraber]

@pnbabu Can the issue be closed?

[pnbabu]

@steffengraber Yes, it can be closed.