On 2021-01-11-M, at 13:05, Ruben Di Battista <[email protected]> wrote:
> I personally would prefer to have a direct dependency on the Macports > libevent, if there are not compatibility issues. That's what we also do with > VTK, where all the compatible dependencies that can be externalized are > externalized. That helps avoiding rebuilding libraries → less disk space in > case two ports share the same dep. > > On Mon, Jan 11, 2021 at 6:51 PM Christopher Nielsen > <[email protected]> wrote: >> >> There’s an open ticket for OpenMPI, related to opportunistic use of libevent: >> >> https://trac.macports.org/ticket/59626 >> >> It turns out that, as of OpenMPI v4.0, the cmake scripts now prefer the use >> of an external libevent, when available. Which means we end up with >> inconsistent behavior, depending on whether the user already has libevent >> installed. >> >> This obviously needs to be fixed. The question is, do we want OpenMPI to >> build and use its own private libevent, or add a formal dependency on ours? >> >> The developer docs don’t provide a strong recommendation either way: >> >> https://www.open-mpi.org/faq/?category=building#libevent-or-hwloc-errors-when-linking-fortran >> >> Given that the internal version was always built and used prior to v4.0, >> using that one might be a good choice. But I don’t have an opinion. >> >> So I was hoping that you folks can help me make this call. Thoughts? Sounds good to me. Anyone else?
