Currently, to force Spack to use an external MPI, you have to specify `buildable: False`
for every MPI provider in Spack in your packages.yaml file. This is both tedious and
fragile, as new MPI providers can be added and break your workflow when you do a
git pull.
This PR allows you to specify an entire virtual dependency as non-buildable, and
specify particular implementations to be built:
```
packages:
all:
providers:
mpi: [mpich]
mpi:
buildable: false
paths:
mpich@3.2 %gcc@7.3.0: /usr/packages/mpich-3.2-gcc-7.3.0
```
will force all Spack builds to use the specified `mpich` install.
* geant4: new version 10.6 plus simplifications
Add new 10.6.0 release, migrating download of source to use Geant4's
public release repo on CERN GitLab. Change versioning scheme to use
clearer and standard semantic scheme.
Update geant4-data and g4XXX data packages with new versions. Migrate
geant4-data as a BundlePackage of the g4XXX packages, installing links
to each under a single directory under share for geant4-data. Ensure
each g4XXX package exports the environment variable pointing to its
location expected by Geant4.
Remove "data" variant from Geant4 package and always use geant4-data.
Simplify cxxstd variant transport to dependencies.
* g4<DATA>: Use self to resolve correct prefix
* geant4, data: Fix flake8 errors
* g4photonevaporation: flake8 fix
* geant4: vecgeom version depends_on
Geant4 major.minor versions have specific dependencies on vecgeom
versions. Add missing vecgeom version for geant4 10.5, and match
version requirements for vecgeom in geant4 depends_on.
* geant4: c++17 patch specific for 10.4.3
* geant4: simplify geant4-data setup
* geant4: Use new define_from_variant function
* geant4: fix flake8 errors
* helics: add new package
* Remove FIXME boilerplate
* Use open @master: verison range for git dependency and remove mpi fix branch version
* Add blank line after spack import
* py-onnx: depends on cmake >= 3.1
* Update var/spack/repos/builtin/packages/py-onnx/package.py
Co-Authored-By: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
* Aded Option to Disable Shared Lua library
Added option to disable generation of shared object library for lua to
avoid build issues on static only platforms
* Fixed Flake8 Issue with Lua Spackage
Fixed indentation issue with lua spackage
The current implementation of `spack-python` will leave an extra shell
around while it runs. That shell should really replace itself with
spack.
- [x] add exec to spack-python script
* Add initial attempt at intel-mpi-benchmarks package
* Add more checksummed versions
* Changes to how makefile is handled
* First working install version. Needs tuning to support building specific benchmarks
* Add variant for building specific benchmarks rather than all of them
* Minor syntax change
* New package: gdrcopy
provides the userspace libraries for gdrcopy.
* Update var/spack/repos/builtin/packages/gdrcopy/package.py
Co-Authored-By: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
* XIOS: add new versions
Patch has been removed because it was not applied to any previously
existing versions and it actually breaks the new versions added by this
PR.
* Sort versions from newest to oldest
This allows the llvm build to support:
* clang cuda
* libomptarget for:
* current host
* cuda
* bitcode compilation of libomptarget device runtime for inlining by
bootstrapping libomptarget
* split dwarf information support as an option for debug builds, if you need a
debug build, for the love of all that's good in the universe use this flag
* adds necessary dependencies for shared library builds and libomp and
libomp target to build correctly
* new version of z3 to make it sufficient to build recent llvm
The actual change is much smaller than the diff, this is because it's been formatted with black. I realize this kinda sucks right now, but I'm hoping it will make future updates here less painful.
the gcc package.py includes patches for a sanitizer related bug that look
like they've been fixed in gcc 8.4.0, which caused `spack install` to fail.
This PR excludes patching gcc >= 8.4.0 and < 9.0.0.
* py-statsmodels: update to 0.102 and fix dependencies
* Update var/spack/repos/builtin/packages/py-statsmodels/package.py
Co-Authored-By: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
* Update URL for new py2neo versions
* Use pypi for py-py2neo
* Add version 4.3.0
* Update py2neo dependencies
* Apply suggestions from code review
Co-Authored-By: Adam J. Stewart <ajstewart426@gmail.com>
Co-authored-by: Adam J. Stewart <ajstewart426@gmail.com>
* NETCDF: Remove maxdims maxvars variant
I'm not sure of the correct protocol to do this, so decided to make a stab and hopefully it works or I'm told the correct way...
The `maxdims` and `maxvars` variants for the NetCDF package were, to the best of my knowledge, only ever used for the Exodus library in the SEACAS package. In versions of NetCDF prior to 4.4.0, Exodus required that the `NC_MAX_DIMS` and `NC_MAX_VARS` be increased over the default values. This requirement was removed in 4.4.0 and later.
I do not know of any way to make a variant depend on the version and since the `maxdims` and `maxvars` variants are integer values and not boolean, then every build of NetCDF will have these variants. Typically `maxdims=1024 maxvars=8192` and the build will patch the `netcdf.h` include file for every build even though it is (almost) never needed.
The SEACAS package has a NetCDF version requirement of >4.6.2, so it no longer specifies the `maxdims` or `maxvars` variant and I could find no other package in spack that uses this variant either, so removal should not break anything *in* spack. However, there is no guarantee that some other external package doesn't use the variant, so I'm not sure of the correct way to remove the variant.
For this PR, I simply removed the variants. If there is a way to specify use of the variant tied to a specific version, I couldn't find it anywhere...
* Address review comment
Removed `is_integral` and `import numbers` since `is_integral` was only place it was used.
* Add blank line for flake8
* magma now extends CudaPackage class, taking care of the gcc conflicts
* enforce +cuda; thus cuda is dependency via CudaPackage class
* add conflict
* use cuda_arch to set GPU_TARGET build option
* get rid of unnecessary constraint
* flake8
* impose cuda version dependency found empirically
* add variant description
* add conflict
Co-authored-by: Sinan81 <Sinan81@github>
Co-authored-by: Sinan81 <sbulut@3vgeomatics.com>
