Compiling with CUDA&OpenCL: 'Lepton' not declared in this scope

[jaimergp]

I am trying to provide the conda recipes as suggested in #6. This is happening in this PR: https://github.com/conda-forge/staged-recipes/pull/13486

After the release of Plumed 2.7, the Reference platform can build correctly (see temporary CPU builds in the CI logs for Linux and MacOS).

However, OpenCL and CUDA are not as lucky. There's an issue with lepton and its scope in the OpenMM headers:

Scanning dependencies of target OpenMMPlumedCUDA
[ 73%] Building CXX object platforms/cuda/CMakeFiles/OpenMMPlumedCUDA.dir/src/CudaPlumedKernelFactory.cpp.o
In file included from ${PREFIX}/include/openmm/cuda/CudaExpressionUtilities.h:30:0,
                 from ${PREFIX}/include/openmm/cuda/CudaContext.h:45,
                 from ${WORK}/platforms/cuda/src/CudaPlumedKernels.h:37,
                 from ${WORK}/platforms/cuda/src/CudaPlumedKernelFactory.cpp:35:
${PREFIX}/include/openmm/common/ExpressionUtilities.h:61:63: error: 'Lepton' was not declared in this scope
     std::string createExpressions(const std::map<std::string, Lepton::ParsedExpression>& expressions, const std::map<std::string, std::string>& variables,

Any ideas? This is the build script.

[peastman]

Nothing obvious comes to mind. That file includes several lepton headers, so it ought to be defined:

https://github.com/openmm/openmm/blob/e23dc9272ab4eafd72f91aace1320a7f1177ba7b/platforms/common/include/openmm/common/ExpressionUtilities.h#L32-L34

[jaimergp]

I have done some more digging. Turns out plumed's conda package also distributes lepton headers, which causes some kind of path confusion... Renaming that directory fixes it, I think.

[jaimergp]

Yep, that's it. Closing!