Lisandro Dalcin

Here you have a reproducer ```bash tar -xf openmpi-5.0.2.tar.gz cd openmpi-5.0.2 export CC=cc export CXX=c++ PREFIX=$HOME/openmpi options=( CC=cc CXX=c++ --prefix="$PREFIX" --disable-dlopen --disable-oshmem --without-ofi --without-ucx --without-psm2 --without-cuda --without-rocm --with-pmix=internal --with-prrte=internal --with-libevent=internal...

I figured out the root of the issue. The failure occurs because of ```bash export MACOSX_DEPLOYMENT_TARGET="11.0" ``` before configure and make. Given that the last libevent release was in 2020-07-05,...

> What MacOS version are you running on? 14.3.1 > I gather that all is fine so long as you don't set that envar? Yes, all is fine without the...

`MACOSX_DEPLOYMENT_TARGET` is essential to binary distributors building a binaries on a NEWER macOS but wanting the generated binaries able to run on an OLDER macOS version. In my particular case,...

@jsquyres I have a hacky but legit solution. Untar the libevent tarball, run `autogen.sh` with a recent (latest) autotools, and create a patch file against the previous source tree. You...

> this is kinda the moral equivalent of shipping a patch. You should still consider the less intrusive alternative of actually shipping a patch alongside the tarball, applying the patch...

I believe I forgot to bump `OMPI_DATATYPE_MAX_PREDEFINED` by one.

@jsquyres Could you please reopen this issue? Thins still do not work quite as I would expect. Here you have another reproducer: ```c #include #include static int query_fn (void *ctx,...

> The grequest provided free function returned an error, why would you expect the request to be `MPI_REQUEST_NULL` ? For what we know the request was not freed. The free...

FWIW, the MPICH behavior is for `MPI_Wait` to set `request=MPI_REQUEST_NULL` and fail returning error. This is my preferred implementation from my previous comment. If you decide to do things differently,...