The go team released 1.9.2 which includes fixes for some things
that 1.9.1 broke:
> ... include fixes to the compiler, linker, runtime, documentation, go command, and the crypto/x509, database/sql, log, and net/smtp packages. They include a fix to a bug introduced in Go 1.9.1 and Go 1.8.4 that broke "go get" of non-Git repositories under certain conditions.
* Exodus: skip the -G "Unix Makefiles" part
The problem is that spack passes -G "Unix Makefiles" into cmake, which normally
works. But in the Exodus package, it is being passed into a bash wrapper
script. In there, the $@ then loses the information about "Unix Makefiles"
being just one argument, and in effect passes -G Unix Makefiles into the cmake
(without quotes), and so cmake only sees -G Unix, and then fails. This is a
known problem with bash with no simple solutions. As a workaround, this patch
skips the first two arguments, i.e., -G and "Unix Makefiles". This makes it
work.
Fixes#5895.
* Port exodusii to cmake
The cmake options were taken from the cmake-exodus bash script and ported to
spack directly.
* Use variant forwarding to forward the 'mpi'
Now instead of
spack install exodusii~mpi^netcdf~mpi^hdf5~mpi
one can just use
spack install exodusii~mpi
* sw4lite: fix build errors and add variants
* sw4lite: change linking against blas and lapack
* change order of blas and lapack
* satisfy flake8 requirements
* Update package.py
* Add the custom paraview lib directory structure to the library paths in the paraview module file.
* Fixing flake8 issues.
* Checking if lib64 exists for paraview module file generation, else use lib.
* Fixing more flake8 problems I introduced.
Since LLVM 3.9 Clang can use the libc++ library by default using the
CLANG_DEFAULT_CXX_STDLIB cmake configuration variable, without having to
specify the -stdlib=libc++ option on the clang++ command line.
This commit makes clang++ use libc++ by default for LLVM 3.9 and later if the
libcxx variant is on.
Fixes#5942.
Chasing a performance regression has lead me to this change, going from default optimization gives a significant performance win. The sweet spot for zlib is apparently `-O2`, both `-Ofast` and `-O3` are slightly worse (regression is about 3% compared with `-O2) in my testing.
Happy to share my methodology with people so we can benchmark on a wider variety of systems.
