[CMake] FindMPI.cmake in 3.11.4

[Chuck Atkins]

Hi Burlen

Of course I read the documentation.
>

I certainly didn't mean for that to come off as an "rtfm", so my appologies
if it did, that wasn't my intent.



> why did they MPI_C_LIBRARIES and MPI_C_INCLUDE_DIRS/PATH get changed from
> cache variables to not chached variables? that seems to be the cause of the
> issue.
>

There was a completere-write of FindMPI from scratch to address a wide
range of issues with it.  In general many of the "old" behaviors were in
opposition to other established "best practices" so it was re-written with
many of these in mind.  As such, much of the old behavior w.r.t. variable
names may not have been preserved in an attemnt to bring the behavior more
inline with recommended current CMake practices.  So in some scenarios it
should be considdered a breaking change.


Alas after trying various settings described there in, nothing worked. my
> settings are ignored and erased.
>


>   Now how does one override the detection and tell it what to use in the
> new module?
>

There's several different scenarios to try to detect MPI and I believe they
happen in this order:

   1. Using the mpicc compiler wrapper as your "regular compiler"
      - libraries and includes should remain empty as nothing needs to be
      added for MPI to work
      - This is the case when using the CrayPE wrappers
   2. Locating mpicc in your path and interrogating it
      - Determined the set of includes and libraries to be added to the
      regular compiler to make MPI work
      - This is the typical behavior that get's used in *most* mpi
      environments.
   3. Falling back to manually searching for headers and libraries

Can you describe your environment and scenario more specifically so I can
try to reproduce it?

Thanks
- Chuck
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://cmake.org/pipermail/cmake/attachments/20180917/951ff0d2/attachment.html>