* Trilinos launch blocking + maintainers
Cuda launch blocking is not needed and slowing modern apps down.
More maintainers to spot issues like this.
---------
Co-authored-by: psakievich <psakievich@users.noreply.github.com>
Elpa's custom preprocessor createst temporary files for which it
assembles long filenames and then uses the last 250 characters. This
results in compilation errors when the first character happens to be a
dash.
* Slurm: extend spack external find support
On Debian srun/salloc --version returns 'slurm-wlm VERSION'. Check for both strings and return the first match.
* non-capturing group for slurm determine_version
Co-authored-by: Wouter Deconinck <wdconinc@gmail.com>
* slurm: add detection test
---------
Co-authored-by: Wouter Deconinck <wdconinc@gmail.com>
This commit removes the +fortran variant when building HDF5 for WRF.
This seems unnecessary, and prevents building WRF with some versions of
Intel MPI, as HDF5 doesn't appear to build with Fortran support and
Intel MPI.
* add new versions up to 1.5 and new variants
variant vtk: make vtk optional
variant shared: build shared libs
added patch to fix parmmg cmake so that it can be used by other software with find_package
* use +private for mmg@5.8: and parmmg@1.5:
* fix style and constraint mmg version
* add a condition on patch, use private_headers from mmg PR feelpp/spack#14
* add MET v12.0.0 and METplus v6.0.0
* Set correct dependencies for metplus@6 in var/spack/repos/builtin/packages/metplus/package.py
* Add missing dependency on proj for met@12
* Add met@12.0.1
* Change @6.0.0 to @6: for requirements in var/spack/repos/builtin/packages/metplus/package.py
* Address reviewer comments for met and metplus
---------
Co-authored-by: Rick Grubin <Richard.Grubin@noaa.gov>
This should help resolve the "No binary found when cache-only was specified"
errors we've recently seen in our GitLab CI pipelines.
example failing job here:
https://gitlab.spack.io/spack/spack/-/jobs/15570931#L370
This error is caused when a generate job finds a spec in the local root
binary mirror, and that spec does not yet exist in the stack-specific mirror.
The fix here is to instead locally cache the stack-specific mirrors and only
use the root-level mirror for public use.
Windows paths with drives were being interpreted as network protocols
in canonicalize_path (which was expanded to handle more general URLs
in #48784).
This fixes that and adds some tests for it.
In Spack v1.0 we plan to parse caret ^ and percent % the same. Their meaning is direct and transitive dependency respectively. It means that variants, versions, arch, platform, os, target and dag hash should go before the %, so that they apply to dependent not the %dependency.
