Spack currently tries to unify everything in the DAG, but this is too strict for build dependencies, where it is fine to build a dependency with a tool that conflicts with a version fo that tool for a dependent's build.
To enable a workaround for conflicts among build dependencies, so that users can install in multiple steps to avoid these conflicts, make the following changes:
* Dont apply package dependency constraints for build deps of installed packages
* Avoid applying constraints for installed packages vs. concrete packages
* Mark all dependencies of installed packages as visited in normalization method
* don't remove dependency links for concrete specs in flat_dependencies
Also add tests:
* Update test to ensure that link dependencies of installed packages have constraints applied
* Add test to check for proper handling of transitive dependencies (which is currently not the case)
ADIOS2 is the latest implementation of the ADaptable Input Output System,
ADIOS. This brand new architecture was designed to continue supporting the
performance legacy of ADIOS, and extend its current capabilities to address
current and future input/output (IO) challenges in the scientific data
lifecycle through effective research and development (R&D) activities.
ADIOS2 was recently added as a Third Party Library of Trilinos when it
was integrated to IOSS, a library inside the SEACAS package.
- spack.compilers.find_compilers now uses a multiprocess.pool.ThreadPool to execute
system commands for the detection of compiler versions.
- A few memoized functions have been introduced to avoid poking the filesystem multiple
times for the same results.
- Performance is much improved, and Spack no longer fork-bombs the system when doing a `compiler find`
- We use `spack list --foramt=html` now, as it is much faster and doesn't
make the docs build take forever.
- Remove `spack list --format=rst` as it is no longer used.
- `stage.source_path` was previously overloaded; it returned `None` if it
didn't exist and this was used by client code
- we want to be able to know the `source_path` before it's created
- make stage.source_path available before it exists.
- use a well-known stage source path name, `$stage_path/src` that is
available when `Stage` is instantiated but does not exist until it's
"expanded"
- client code can now use the variable before the stage is created.
- client code can test whether the tarball is expanded by using the new
`stage.expanded` property instead of testing whether `source_path` is
`None`
- add tests for the new source_path semantics
- make tty.msg, tty.info, etc. print the exception type and stringified
message if the message argument is an exception.
- simplify parts of the code that call tty.debug(str(e))
- add extra tty.debug statements in places where exceptions were
previously ignored
- `spack graph --static` (and `spack.graph.dot_graph`) now do the "right
thing" and print the possible dependency graph of provided packages.
- `spack graph --static` no longer concretizes specs, as it only relies
on class level metadata
- Previously the behavior was not consistent -- `spack graph --static`
would graph possible dependencies of concrete specs, but would only
include some of them. The new code properly pursues all possible
dependencies, and allows traversing by different dependency types.
- `spack dependencies` can now take a --deptype argument to only traverse
particular deptypes
- add a new "common" argument for deptype in spack.cmd.common.arguments
- Database.installed_relatives() can now also take a deptype argument
- this is used by `spack dependencies --installed`
- `PackageBase.possible_dependencies` now:
- accepts a deptype param that controls dependency types traversed
- returns a dict mapping possible depnames to their immediate possible
dependencies (this lets you build a graph easily)
- Add tests for PackageBaes
- The 'name' attribute for packages was being set in DirectiveMeta, which
wasn't consistent with other class properties (like fullname, etc.)
- Move it to be a class property of `PackageMeta`, and add the
corresponding property method wrapper on `PackageBase`
* add c99_flag, c11_flag to compiler class
* implement c99_flag, c11_flag for gcc
* implement c99_flag, c11_flag for arm
* implement c99_flag for cce
* implement c99_flag, c11_flag for clang
* implement c99_flag, c11_flag for intel
* implement c99_flag, c11_flag for xl
Previously, module files were not set with the same permissions as the package installation. For world-readable packages, this would not cause a problem. For group readable packages, it does:
```
packages:
mypackage:
permissions:
group: mygroup
read: group
write: group
```
In this case, the modulefile is unreadable by members of the group other than the one who installed it. Add logic to the modulefile writers to set the permissions based on the configuration in `packages.yaml`
Due to the large number of issues this migration has caused, this rolls back
to the older autotools build. However, it still keeps the updates wrt
variants and virtual packages.
* Apply patch to git to make it search for config files from its current location instead of the locations hard coded at install.
* Add provenance info for patch
* Pass needed flags to make
* Environment variables need because of relocation
* Build cache: relocate path to spack/bin/sbang in text files.
* Found in testing.
* update packaging test
* Make sbang replacement including #!/bin/bash. Add an additional spack prefix replacement to fix stage directory references.
* flake8
* Use buildinfo.get() so old buildcaches without buildinfo['spackprefix'] can be read.
* Updating the TAU package with 2.28 and public repo
* Cleaning up TAU package and fixing binutils to install extras
The extras variant will install the extra ELF headers needed
by the TAU package. The TAU package has been cleaned up and
extended.
* Adding some comments to the changes in binutils
* Adding correct python support, finding include and lib paths.
* Added PDT 3.25.1.
* Added TAU v2.28.1 with libdwarf and libelf.
* When +libdwarf is activated, +libelf is not automatically activated. It checks separately for libelf.
* Xrootd: set PYTHON_EXECUTABLE with +python variant
When building with python3 PYTHON_EXECUTABLE needs to be set for deprecated FindPythonInterp to work as expected.
* Add comment
* Flake8
* More flake8
* Update package.py
* typo
* Also add versions 0.9.9.104, 0.9.9.52, 0.9.9.0, and 'develop'
* Replace version 0.9.7 with 0.9.7.0
* Update Rose to be an Autotools package
* Update boost dependency to 1.60.0 or later (previously 1.47.0)
* Add EDG binary identifiers for ROSE versions installing from
archives (rather than git)
* Add variants for mvapich2 backend, c++11, debug symbols,
optimization, and ROSE-based projects (codethorn, autopar, and
polyopt)
* Remove patch add_spack_compiler_recognition
* Improve C++ standard computation in ACTS package
* Use a more clever strategy for aligning ACTS cxxstd with ROOT cxxstd
* Add new ACTS releases
* ACTS is not compatible with the new Boost CMake build system yet
* Explain why Boost 1.70 does not work currently
* Add LSF package, which cannot be installed by Spack and must be
system-installed. The package install will fail if no external
LSF is registered in packages.yaml (LSF may not be installed in a
well-known location and the external entry helps Spack locate it
for dependents).
* Add LSF dependency to OpenMPI when schedulers=lsf is chosen
