I think I hit the wrong key and my email disappeared, apologies if you get this twice
> The package config file should have only declarative effects and not actually modify the loading project's build, so that advice was correct. < Understood. This seems sensible. > The _INCLUDE_DIRS, _LIBRARIES, etc. variables that hdf5-config.cmake provides should already be populated with the transitive dependencies. It should be able to do this without find_package(MPI) too. < OK, that's great. I've been doing this in my builds but thought it was somehow 'wrong' in the public release. Since hdf5 already knows which mpi libs/includes are being pulled in, all is ok. One thing : Suppose hdf5 pulls in the mpi lib and includes, and the project using it requests mpi separately. Is there a correct - or standard - way of making sure the user does not choose a different mpi lib - If we know that the project using hdf5 will call find_package(MPI) then we could too in the config, and that'd actually be a good way of making sure the same one is used. If the calling project does need or use mpi itself, it can/would just confuse the user... Is there a correct way of handling this? (I'm guessing the best we can do is export a variable HDF5_MPI_INCLUDE etc etc which the user can test if concerned that the user mpi doesn't match the hdf5 mpi) Thanks again JB -- Powered by www.kitware.com Visit other Kitware open-source projects at http://www.kitware.com/opensource/opensource.html Please keep messages on-topic and check the CMake FAQ at: http://www.cmake.org/Wiki/CMake_FAQ Follow this link to subscribe/unsubscribe: http://public.kitware.com/cgi-bin/mailman/listinfo/cmake-developers
