error when using update_and_rebuild_libmesh.sh using msys2 on windows

[starkekr]

I am following the MOOSE installation instructions here. I have progressed through the installation instructions and am stuck on the libmesh installation. I am following the instructions for building libmesh as a submodule of MOOSE.

I have the following errors when I run ./scripts/update_and_rebuild_libmesh.sh at the ending step of the rebuilding MOOSE.

make[2]: Leaving directory '/home/stark/projects/moose/libmesh/build/contrib/eigen/eigen'
make[1]: *** [Makefile:1033: all-recursive] Error 1
make[1]: Leaving directory '/home/stark/projects/moose/libmesh/build/contrib'
make: *** [Makefile:33131: all-recursive] Error 1

It gets caught up making eigen. I would greatly appreciate any help.

[starkekr]

I have disabled eigen and netcdf using --disable-eigen --disable-netcdf but now I have a new problem.

../src/base/libmesh.C: In constructor 'libMesh::LibMeshInit::LibMeshInit(int, const char* const*, TIMPI::communicator, int)':
../src/base/libmesh.C:648:3: error: 'setenv' was not declared in this scope; did you mean 'getenv'?
648 | setenv("HDF5_USE_FILE_LOCKING", "FALSE", /overwrite=false/0);
| ^~~~~~
| getenv
make[1]: *** [Makefile:26063: src/base/libmesh_opt_la-libmesh.lo] Error 1

any suggestions?

[jwpeterson]

Looks like we picked up this msys2 regression in d457766. I guess a temporary fix could be to change the line:

#ifdef LIBMESH_HAVE_HDF5

to

#if defined(LIBMESH_HAVE_HDF5) && !defined(_MSC_VER)

That should simply skip the setenv command when you are compiling on windows. Not sure what the correct equivalent to a setenv call on Windows would actually be.

[jwpeterson]

I created PR #3473 to fix the pid_t typedef issue.

Regarding the Eigen error, it sounds like Eigen is not optional for MOOSE currently, so you won't be able to build libmesh without Eigen if you want to use MOOSE. I'm not sure what is actually causing the error you are seeing above, since there is nothing to compile for Eigen (it is a header-only library).

I'm wondering if it is some simple filesystem issue with symbolic links in the git submodule that we use for Eigen... currently we have contrib/eigen/eigen -> contrib/eigen/gitshim and contrib/eigen/gitshim/Eigen -> contrib/eigen/git/Eigen so if symlinks aren't working for whatever reason, I could see this maybe giving the error you reported.

[jwpeterson]

I also opened #3474 to fix the setenv error that you encountered.

[starkekr]

I tried to verify the symlinks that you specified. I don't know if this is the correct way of doing this but I have attached a picture of my terminal window

I also tried passing the eigen path to the configure using
../configure --with-methods='opt' --prefix=$HOME/projects/libmesh/installed --with-static=yes --with-shared=no --enable-static --disable-shared --enable-unique-id --disable-maintainer-mode --enable-petsc-hypre-required --enable-metaphysicl-required --disable-netcdf --disable-netcdf-4 --with-eigen-include=/home/stark/projects/libmesh/contrib/eigen/3.3.9/Eigen/src && make && make install

but it didn't work. I may be using the wrong directory though.

[starkekr]

Ok, so i got past the eigen problem by not relying on the symlinks and just copying the eigen files to /build/contrib/eigen/eigen. Its messy but it works. Now when I went to build MOOSE I get the following error

C:/msys64/home/stark/projects/moose/framework/src/outputs/formatters/ExodusFormatter.C:18:10: fatal error: libmesh/exodusII.h: No such file or directory
18 | #include "libmesh/exodusII.h"

So I think I need to build libmesh with netcdf enabled or maybe there is some way to disable netcdf on the MOOSE side? The ideal scenario would be that I compile libmesh with netcdf enabled so I started trying to do that. But I ran into another problem, I run the configure script with the following additional arguments:

--enable-netcdf-required NETCDF_DIR=C:/msys64/mingw64

and the the configure runs fine. Once I run make I get the following error:

make[4]: Entering directory '/home/stark/projects/libmesh/build/contrib/netcdf/v4/libsrc'
CC libnetcdf3_la-memio.lo
../../../../../contrib/netcdf/v4/libsrc/memio.c: In function 'memio_new':
../../../../../contrib/netcdf/v4/libsrc/memio.c:148:7: error: unknown type name 'SYSTEM_INFO'
148 | SYSTEM_INFO info;
| ^~~~~~~~~~~
../../../../../contrib/netcdf/v4/libsrc/memio.c:149:7: warning: implicit declaration of function 'GetSystemInfo' [-Wimplicit-function-declaration]
149 | GetSystemInfo (&info);
| ^~~~~~~~~~~~~
../../../../../contrib/netcdf/v4/libsrc/memio.c:150:22: error: request for member 'dwPageSize' in something not a structure or union
150 | pagesize = info.dwPageSize;

Any thoughts?

[jwpeterson]

We don't currently support building with system NetCDF (or any NetCDF other than the one distributed with libmesh). So I don't think this approach of using NETCDF_DIR=C:/msys64/mingw64 is likely to work unfortunately. Can you remind me what the reason for not using the NetCDF that comes with libmesh is? Is it also a symlink issue?

[mkhon]

@jwpeterson my changes were to compile libmesh using Visual Studio. When msys2 is used many posix headers are available, so pid_t definition should be additionally ifdef'ed out for msys2