Tests for Vector<MultiPrecision> were removed by TO...

Fixed definition of static members in communicators when the header files are used in multiple translation units

Works fine with g++ 5.5 and nvcc 8.0 in Arch Linux. File a bug report on
Gitlab if you have problems.

This is getting really cumbersome and error-prone, somebody should
figure out how to do this more generally for all possible combinations
of CV-qualifications...

The link provided in the comment is not related - tnl-cuda-arch is not
built as a static binary.

Anyway, I don't see any problem with this. Either fix it properly or at
least file a full bug report on Gitlab. Upgrading cmake on your system
might also help...

It is not needed, since CUSPARSE is included in all versions of CUDA
since at least 7.0.

Also note that CUSPARSE is needed only for some tests and legacy code
(tnl-benchmark-spmv and wrappers in Matrices::CSR) and experimental code
(Preconditioners::ILU0). Everything is still guarded by the
HAVE_CUSPARSE macro, so the user can pass -DHAVE_CUSPARSE -lcusparse to
the compiler to enable the features.

This way they can use either yes/no, on/off, or true/false.