Obtain raw MPI_Comm from mpl::communicator - Version 2 about mpl HOT 4 CLOSED

from mpl.

You did, but my intention was to provide a solution that hopefully made no external mods to the wrapped comm_. Thereby satisfying what I believe to be the design constraints of MPL while also allowing users such as me the ability to operate with MPI and MPL within the same code base. It would be trivial to extend my proposal to say (untested and uncompiled):

    MPI_Comm& get_mpi_comm_nonconst()
    {
        return comm_;
    }

    /// \brief Allow raw MPI commands to be run 
    /// that need access to the MPI_Comm from this
    /// \param userCode a functor provided by the user taking a MPI_Comm&
    template<typename UserCode>
    void execute_raw_modifiable_mpi(const UserCode& userCode)
    {
        userCode(get_mpi_comm_nonconst());
    }

Here the purpose is to make it clear that what you get can change the wrapped comm_ with reference to 'execute_raw_mpi` which cannot I believe mod the communicator, only use it.

from mpl.

Although, I fully understand the desire to find a way to mix MPL with MPI code I will not incorporate this in any way into MPL. The more I think about it the more skeptical I get. It might be very useful in the short run, it will likely cause many issues in the long run.

The problem lies not so much in introducing an execute_raw_modifiable_mpi method or something similar. Problems will arise when removing it once MPL has become close to feature complete (It's long way ahead, I know.) This might be a huge breaking change for those who rely on this feature. I do not want to introduce features into MPL, where I know from day one on that I want to get rid of them at a later stage.

Furthermore, allowing a mixed programming style that uses MPI and MPL eliminates the need to develop MPL further.

from mpl.

Understood

from mpl.