* Make use of new QE multi-valued HDF5 variant interface
* Remove explicit CUDA dependency (which is handled by inheriting
from CudaPackage)
* Update cuda_arch unset variant check
libbeagle compiles against CUDA by default but no there is no mention
of it in the package recipe. This PR adds explicit cuda paths and
variants, and fixes the target architecture as well (for those who
don't have compute_13)
Serial HDF5 in serial QE only works in develop version. This adds
a conflict to the QE package to prevent Spack from trying to build
specs which don't satisfy this constraint.
This restores the use of Package.headers when computing -I options
for building a package that was added in #8136 and reverted in
#10604. #8136 used utility logic that located all header files in
an installation prefix, and calculated the -I options as the
immediate roots containing those header files.
In some cases, for a package containing a directory structure like
prefix/
include/
ex1.h
subdir/
ex2.h
dependents may expect to include ex2.h relative to 'include', and
adding 'prefix/include/subdir' as a -I was causing errors,
in particular if ex2.h has the same name as a system header.
This updates header utility logic to by default return the base
"include" directory when it exists, rather than subdirectories.
It also makes it possible for package implementers to override
Package.headers to return the subdirectory when it is required
(for example with libxml2).
Update older md5 checksums to sha256. Require CMake >= 3.6.0 (updated
from 3.5.0). Remove some third-party depends_on statements (which
were commented out before).
Extract implicit xenv dependency into a new package and add it as an
explicit dependency. Update Gaudi to explicitly depend on older
versions of xenv rather than the latest develop because that
conflicts with Spack's compiler wrappers.
Spack warns users when a dependency package updates CPATH. This
warning message is generating bug reports and alarm in cases where
there is no problem. For now this downgrades the warning message to
the debug level, so it only shows up if something goes wrong for the
user and they ask for more information from Spack.
The nvptx installation support phase (added in 040b658) was always
running for gcc regardless of whether +nvptx was enabled. This
removes the "@when" syntax (which apparently is not working) and
adds a check inside of the "nvptx_install" function to bail out if
+nvptx is not enabled.
This spack command adds a new schema for a file which describes the
builder containers available, along with the compilers availabe on
each builder. The release-jobs command then generates the .gitlab-ci.yml
file by first expanding the release spec set, concretizing each spec
(in an appropriate docker container if --this-machine-only argument is
not provided on command line), and then combining and staging all the
concrete specs as jobs to be run by gitlab.
The built images are set up with fairly recent versions of gcc and
clang:
- centos_7: [ gcc@5.5.0 (built from src), clang@6.0.0 (spack-built from src) ]
- ubuntu_18.04: [ gcc@5.5.0 (system), clang@6.0.0-1ubuntu2 (system) ]
Adds four new sub-commands to the buildcache command:
1. save-yaml: Takes a root spec and a list of dependent spec names,
along with a directory in which to save yaml files, and writes out
the full spec.yaml for each of the dependent specs. This only needs
to concretize the root spec once, then indexes it with the names of
the dependent specs.
2. check: Checks a spec (via either an abstract spec or via a full
spec.yaml) against remote mirror to see if it needs to be rebuilt.
Comparies full_hash stored on remote mirror with full_hash computed
locally to determine whether spec needs to be rebuilt. Can also
generate list of specs to check against remote mirror by expanding
the set of release specs expressed in etc/spack/defaults/release.yaml.
3. get-buildcache-name: Makes it possible to attempt to read directly
the spec.yaml file on a remote or local mirror by providing the path
where the file should live based on concretizing the spec.
4. download: Downloads all buildcache files associated with a spec
on a remote mirror, including any .spack, .spec, and .cdashid files
that might exist. Puts the files into the local path provided on
the command line, and organizes them in the same hierarchy found on
the remote mirror
This commit also refactors lib/spack/spack/util/web.py to expose
functionality allowing other modules to read data from a url.
- add CombinatorialSpecSet in spack.util.spec_set module.
- class is iterable and encaspulated YAML parsing and validation.
- Adjust YAML format to be more generic
- YAML spec-set format now has a `matrix` section, which can contain
multiple lists of specs, generated different ways. Including:
- specs: a raw list of specs.
- packages: a list of package names and versions
- compilers: a list of compiler names and versions
- All of the elements of `matrix` are dimensions for the build matrix;
we take the cartesian product of these lists of specs to generate a
build matrix. This means we can add things like [^mpich, ^openmpi]
to get builds with different MPI versions. It also means we can
multiply the build matrix out with lots of different parameters.
- Add a schema format for spec-sets
On macOS, the build of libxxhash is performed by a bare `make`
launched by `MakePackage`s default `build` phase. As `prefix`
is not set at this point, the library gets compiled with a hard-coded
install name of `/usr/local/libxxhash.dylib`. Downstream clients,
in the tested case Root, will then fail to build as the install name
is incorrect.
Move setting of `prefix` from `install` to `edit` so that it
propagates to subsequent `build` and `install` phases.
* QE hdf5 is now a multi-valued variant. Update conflicts accordingly.
* Update dependencies for hdf5 multi-valued variant. Bug fix: high-level hdf5 library needs to be a hard dependecy.
* Support for HDF5 serial interface. Note that this is not supported in QE autoconf, but will be fixed in the next official release.
