Lisandro Dalcin

@bosilca Oh, man, sorry, I messed up indeed. Please `git fetch` and `git reset --hard` the mpi4py repo at branch `feature/MPI-5`. Now `src/lib-mpi/ulfm.h` has the right guard on the major...

> I found a typo in your commit that prevented things from working properly on my side Of course! Should have tried it on my side, I wouldn't have wasted...

Folks, please ping me once you submit a new PR to branch v.5.0.x.

Update: #395 improves the handling of `MSMPI_VER`. If `MSMPI_VER` is not found in the Windows registry, then as an environment variable (this is not standard practice for MSMPI), and if...

I'm wondering what's the output of `socket.gethostname()` in that system...

@hzhou Is there any upper bound for "port_name_size" ? I see 1024 being used in some defines, but I'm not sure whether that's the upper bound.

@hzhou Would it be possible to add a new `#define MPIX_MAX_PORT_NAME ` in `mpi.h`? If using ch4:ucx, then` ` should be 1024, but otherwise you set it just to `MPI_MAX_PORT_NAME`....

> How would MPIX_MAX_PORT_NAME help? It would help users in two ways: a) provide an actual safe upper bound than can be used to allocate the string buffer in the...

Maybe a `MPIX_Open_port(MPI_Info info, char *port_name, int len)` would also be useful?

Using an `MPI_Info` with key-value pairs seems like the ideal interface for MPI implementations to have absolute freedom to advertise GPU support in whatever format and detail they want. I'm...