The idea is that `spack -e env add ./concrete-spec.json` would list the
full hash in the specs, so that (a) it's not ambiguous and (b) it could
in principle results in constant time lookup instead of linear time
substring match in large build caches.
* Add a descriptor to have a class level constant
This descriptor helps intercept places where we set a value on instances.
It does not really behave like "const" in C-like languages, but is the
simplest implementation that might still be useful.
* Add a descriptor to deprecate properties/attributes of an object
This descriptor is used as a base class. Derived classes may implement a
factory to return an adaptor to the attribute being deprecated. The
descriptor can either warn, or raise an error, when usage of the deprecated
attribute is intercepted.
---------
Co-authored-by: Harmen Stoppels <me@harmenstoppels.nl>
Currently, `spack solve` has different spec selection semantics than `spack spec`.
`spack solve` currently does not allow specifying a single spec when an environment is active.
This PR modifies `spack solve` to inherit the interface from `spack spec`, and to use
the same spec selection logic. This will allow for better use of `spack solve --show opt`
for debugging.
---------
Co-authored-by: Todd Gamblin <tgamblin@llnl.gov>
Originally, concretization failed if the splice config points to an invalid replacement.
This PR defers the check until we know the splice is needed, so that irrelevant splices
with bad config cannot stop concretization.
While I was at it, I improved an error message from an assert to a ValueError.
* Normalize Spack Win entrypoints
Currently Spack has multiple entrypoints on Windows that in addition to
differing from *nix implementations, differ from shell to shell on
Windows. This is a bit confusing for new users and in general
unnecessary.
This PR adds a normal setup script for the batch shell while preserving
the previous "click from file explorer for spack shell" behavior.
Additionally adds a shell title to both powershell and cmd letting users
know this is a Spack shell
* remove doskeys
#44588 we added logic to suppress deprecation warnings for the
Intel classic compilers. This depended on matching against
* The compiler names (looking for icc, icpc, ifort)
* The compiler version
When using an Intel compiler with fortran wrappers, the first check
always fails. To support using the fortran wrappers (in combination
with the classic Intel compilers), we remove the first check and
suppress if just the version matches. This works because:
* The newer compilers like icx can handle (ignore) the flags that
suppress deprecation warnings
* The Cray wrappers pass the underlying compiler version (e.g. they
report what icc would report)
Connection objects are Python version, platform and multiprocessing
start method independent, so better to use those than a mix of plain
file descriptors and inadequate guesses in the child process whether it
was forked or not.
This also allows us to delete the now redundant MultiProcessFd class,
hopefully making things a bit easier to follow.
This allows the following
```python
cache.init_entry("my/cache")
with cache.read_transaction("my/cache") as f:
data = f.read() if f is not None else None
```
mirroring `write_transaction`, which returns a tuple `(old, new)` where
`old` is `None` if the cache file did not exist yet.
The alternative that requires less defensive programming on the call
site would be to create the "old" file upon first read, but I did not
want to think about how to safely atomically create the file, and it's
not unthinkable that an empty file is an invalid format (for instance
the call site may expect a json file, which requires at least {} bytes).
This PR is in response to a question in the `environments` slack channel (https://spackpm.slack.com/archives/CMHK7MF51/p1729200068557219) about inadequate CLI help/documentation for one specific subcommand.
This PR uses the approach I took for the descriptions and help for `spack test` subcommands. Namely, I use the first line of the relevant docstring as the description, which is shown per subcommand in `spack env -h`, and the entire docstring as the help. I then added, where it seemed appropriate, help. I also tweaked argument docstrings to tighten them up, make consistent with similar arguments elsewhere in the command, and elaborate when it seemed important. (The only subcommand I didn't touch is `loads`.)
For example, before:
```
$ spack env update -h
usage: spack env update [-hy] env
positional arguments:
env name or directory of the environment to activate
optional arguments:
-h, --help show this help message and exit
-y, --yes-to-all assume "yes" is the answer to every confirmation request
```
After the changes in this PR:
```
$ spack env update -h
usage: spack env update [-hy] env
update the environment manifest to the latest schema format
update the environment to the latest schema format, which may not be
readable by older versions of spack
a backup copy of the manifest is retained in case there is a need to revert
this operation
positional arguments:
env name or directory of the environment
optional arguments:
-h, --help show this help message and exit
-y, --yes-to-all assume "yes" is the answer to every confirmation request
```
---------
Co-authored-by: Todd Gamblin <tgamblin@llnl.gov>
Commit aa0825d642 accidentally added a semicolon
to the ANSI escape sequence even if the color code was `None` or unknown, breaking the
bold, uncolored font-face. This PR restores the old behavior.
---------
Co-authored-by: Todd Gamblin <tgamblin@llnl.gov>
* Add type hints to all query* methods
* Inline docstrings
* Change defaults from `any` to `None` so they can be type hinted in old Python
* Pre-filter on given hashes instead of iterating over all db specs
* Fix a bug where the `--origin` option of uninstall had no effect
* Fix a bug where query args were not applied when searching by concrete spec
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Fixes a change in behavior/bug in
70412612c7, where partial environment
installs would mark the selected spec as explicitly installed, even if
it was not a root of the environment.
The desired behavior is that roots by definition are the to be
explicitly installed specs. The specs on the `spack -e ... install x`
command line are just filters for partial installs, so leave them
implicitly installed if they aren't roots.
ci: Remove deprecated logic from the ci module
Remove the following from the ci module, schema, and tests:
- deprecated ci stack and handling of old ci config
- deprecated mirror handling logic
- support for artifacts buildcache
- support for temporary storage url
Both `multiprocessing.connection.Connection.__del__` and `io.IOBase.__del__` called `os.close` on the same file descriptor. As of Python 3.13, this is an explicit warning. Ensure we close once by usef `os.fdopen(..., closefd=False)`
* stacks: add a stack for devtools on darwin
After getting this whole mess building on darwin, let's keep it that
way, and maybe make it so we have some non-ML darwin binaries in spack
as well.
* reuse: false for devtools
* dtc: fix darwin dylib name and id
On mac the convention is `lib<name>.<num>.dylib`, while the makefile
creates a num suffixed one by default. The id in the file is also a
local name rather than rewritten to the full path, this fixes both
problems.
* node-js: make whereis more deterministic
* relocation(darwin): catch Mach-O load failure
The MachO library can throw an exception rather than return no headers,
this happened in an elf file in the test data of go-bootstrap. Trying
catching the exception and moving on for now. May also need to look
into why we're trying to rewrite an elf file.
* qemu: add darwin flags to clear out warnings
There's a build failure for qemu in CI, but it's invisible because of
the immense mass of warning output. Explicitly specify the target macos
version and remove the extraneous unknown-warning-option flag.
* dtc: libyaml is also a link dependency
libyaml is required at runtime to run the dtc binary, lack of it caused
the ci for qemu to fail when the library wasn't found.
fixes#47101
The bug was introduced in #33495, where `spack find was not updated,
and wasn't caught by unit tests.
Now a Database can accept a custom predicate to select the installation
records. A unit test is added to prevent regressions. The weird convention
of having `any` as a default value has been replaced by the more commonly
used `None`.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
#44327 made sure to always run `set_package_py_globals` on all
packages before running `setup_dependent_package` for any package,
so that packages implementing the latter could depend on variables
like `spack_cc` being defined.
This ran into an undocumented dependency: `std_cmake_args` is set in
`set_package_py_globals` and makes use of `cmake_prefix_paths` (if it
is defined in the package); `py-torch`es implementation of
`cmake_prefix_paths` depends on a variable set by
`setup_dependent_package` (`python_platlib`).
This generally restores #44327, and corrects the resulting issue by
moving assignment of `std_cmake_args` to after both actions have been
run.
Remove the constraint for concrete specs and simply take the
max(version) if a version is not given. This should default to the
highest infinity version which is also the logical best guess for
doing development.
* Remove concrete verision constriant for develop, set docs
* Add unit-test
* Update lib/spack/docs/environments.rst
Co-authored-by: kwryankrattiger <80296582+kwryankrattiger@users.noreply.github.com>
* Update lib/spack/spack/cmd/develop.py
Co-authored-by: Greg Becker <becker33@llnl.gov>
* Consolidate env collection in cmd
* Style
---------
Co-authored-by: kwryankrattiger <80296582+kwryankrattiger@users.noreply.github.com>
Co-authored-by: Greg Becker <becker33@llnl.gov>
The CMake builder in Spack actually adds incorrect rpaths. They are
unfiltered and incorrectly ordered compared to what the compiler wrapper
adds.
There is no need to specify paths to dependencies in `CMAKE_INSTALL_RPATH`
because of two reasons:
1. CMake preserves "toolchain" rpaths, which includes the rpaths injected
by our compiler wrapper.
2. We use `CMAKE_INSTALL_RPATH_USE_LINK_PATH=ON`, so libraries we link
to are rpath'ed automatically.
However, CMake does not create install rpaths to directories in the package's
own install prefix, so we set `CMAKE_INSTALL_RPATH` to the educated guess
`<prefix>/{lib,lib64}`, but omit dependencies.
Some assertion are not testing DAG invariants, and they are passing only
because of the simple structure of the builtin.mock repository on develop.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
This PR allows users to configure explicit splicing replacement of an abstract spec in the concretizer.
concretizer:
splice:
explicit:
- target: mpi
replacement: mpich/abcdef
transitive: true
This config block would mean "for any spec that concretizes to use mpi, splice in mpich/abcdef in place of the mpi it would naturally concretize to use. See #20262, #26873, #27919, and #46382 for PRs enabling splicing in the Spec object. This PR will be the first place the splice method is used in a user-facing manner. See https://spack.readthedocs.io/en/latest/spack.html#spack.spec.Spec.splice for more information on splicing.
This will allow users to reuse generic public binaries while splicing in the performant local mpi implementation on their system.
In the config file, the target may be any abstract spec. The replacement must be a spec that includes an abstract hash `/abcdef`. The transitive key is optional, defaulting to true if left out.
Two important items to note:
1. When writing explicit splice config, the user is in charge of ensuring that the replacement specs they use are binary compatible with whatever targets they replace. In practice, this will likely require either specific knowledge of what packages will be installed by the user's workflow, or somewhat more specific abstract "target" specs for splicing, to ensure binary compatibility.
2. Explicit splices can cause the output of the concretizer not to satisfy the input. For example, using the config above and consider a package in a binary cache `hdf5/xyzabc` that depends on mvapich2. Then the command `spack install hdf5/xyzabc` will instead install the result of splicing `mpich/abcdef` into `hdf5/xyzabc` in place of whatever mvapich2 spec it previously depended on. When this occurs, a warning message is printed `Warning: explicit splice configuration has caused the the concretized spec {concrete_spec} not to satisfy the input spec {input_spec}".
Highlighted technical details of implementation:
1. This PR required modifying the installer to have two separate types of Tasks, `RewireTask` and `BuildTask`. Spliced specs are queued as `RewireTask` and standard specs are queued as `BuildTask`. Each spliced spec retains a pointer to its build_spec for provenance. If a RewireTask is dequeued and the associated `build_spec` is neither available in the install_tree nor from a binary cache, the RewireTask is requeued with a new dependency on a BuildTask for the build_spec, and BuildTasks are queued for the build spec and its dependencies.
2. Relocation is modified so that a spack binary can be simultaneously installed and rewired. This ensures that installing the build_spec is not necessary when splicing from a binary cache.
3. The splicing model is modified to more accurately represent build dependencies -- that is, spliced specs do not have build dependencies, as spliced specs are never built. Their build_specs retain the build dependencies, as they may be built as part of installing the spliced spec.
4. There were vestiges of the compiler bootstrapping logic that were not removed in #46237 because I asked alalazo to leave them in to avoid making the rebase for this PR harder than it needed to be. Those last remains are removed in this PR.
Co-authored-by: Nathan Hanford <hanford1@llnl.gov>
Co-authored-by: Gregory Becker <becker33@llnl.gov>
Co-authored-by: Tamara Dahlgren <dahlgren1@llnl.gov>
* cuda: Add 12.6.2
* Update cuda build system
- Remove gcc@6 conflict that was only a deprecation (probably has be added again with cuda@13)
- Update cuda_arch support by CUDA version
- Kepler support has ended with cuda@12
- Recently added 90a Hopper "experimental" features architecture was
missing the dependency on cuda@12:
* CMake: Improve incremental build speed.
CMake automatically embeds an updated configure step into make/ninja that will be called during the build phase. By default if a `CMakeCache.txt` file exists in the build directory CMake will use it and this + `spec.is_develop` is sufficient evidence of an incremental build.
This PR removes duplicate work/expense from CMake packages when using `spack develop`.
* Update cmake.py
* [@spackbot] updating style on behalf of psakievich
* Update cmake.py
meant self not spec...
---------
Co-authored-by: psakievich <psakievich@users.noreply.github.com>
* Environment.clear: ensure clearing is passed through to manifest
* test/cmd/env: make test_remove_commmand round-trip to disk
* cleanup vestigial variables
Some Windows Python installations may store the Python exe in Scripts/
rather than the base directory. Update `.command` to search in both
locations on Windows. On all systems, the search is now done
recursively from the search root: on Windows, that is the base install
directory, and on other systems it is bin/.
* Remove "modify_object_macholib"
According to documentation, this function is used when installing
Mach-O binaries on linux. The implementation seems questionable at
least, and the code seems to be never hit (Spack currently doesn't
support installing Mach-O binaries on linux).
* Fix relocation on macOS, when store projection changes
---------
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Enable tests for symlink-based views (this works with almost no
modifications to the view logic). View logic is not yet robust
for hardlink/junction-based views, so those are disabled for now
(both in the tests and as subcommands to `spack view`).
Also adds support for Paraview and CMake to build with Qt support on
Windows.
The remaining edits are to enable building of Qt itself on Windows:
* Several packages needed to update `.libs` to properly locate
libraries on Windows
* Qt needed a patch to allow it to build using a Python with a space
in the path
* Some Qt dependencies had not been ported to Windows yet
(e.g. `harfbuzz` and `lcms`)
This PR does not provide a sufficient GL for Qt to use Qt Quick2, as
such Qt Quick2 is disabled on the Windows platform by this PR.
---------
Co-authored-by: Dan Lipsa <dan.lipsa@kitware.com>
* Bugfix/Installer: properly track task queueing
* Move ordinal() to llnl.string; change time to attempt
* Convert BuildTask to use kwargs (after pkg); convert STATUS_ to BuildStats enum
* BuildTask: instantiate with keyword only args after the request
* Installer: build request is required for initializing task
* Installer: only the initial BuildTask cannnot have status REMOVED
* Change queueing check
* ordinal(): simplify suffix determination [tgamblin]
* BuildStatus: ADDED -> QUEUED [becker33]
* BuildTask: clarify TypeError for 'installed' argument
`spack gc` has so far been a global or environment-specific thing.
This adds the ability to restrict garbage collection to specific specs,
e.g. if you *just* want to get rid of all your unused python installations,
you could write:
```console
spack gc python
```
- [x] add `constraint` arg to `spack gc`
- [x] add a simple test
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
There was a bit of mystery surrounding the arguments for `_setup_pkg_and_run`. It passes
two file descriptors for handling the `gmake`'s job server in child processes, but they are
unsed in the method.
It turns out that there are good reasons to do this -- depending on the multiprocessing
backend, these file descriptors may be closed in the child if they're not passed
directly to it.
- [x] Document all args to `_setup_pkg_and_run`.
- [x] Document all arguments to `_setup_pkg_and_run`.
- [x] Add type hints for `_setup_pkg_and_run`.
- [x] Refactor exception handling in `_setup_pkg_and_run` so it's easier to add type
hints. `exc_info()` was problematic because it *can* return `None` (just not
in the context where it's used). `mypy` was too dumb to notice this.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
* Revert "`cc`: ensure that RPATHs passed to linker are unique"
This reverts commit 2613a14c43.
* Revert "`cc`: simplify ordered list handling"
This reverts commit a76a48c42e.
Updated the terminology for the two types of environments to be
consistent with that used in the tutorial for the last three years.
Additionally:
* changed 'anonymous' to 'independent in environment command+test for consistency.
#45205 already removed previous use of single letter packages
from unit tests, in view of reserving `c` as a language (see #45191).
Some use of them has been re-introduced accidentally in #46382, and
is making unit-tests fail in the feature branch #45189 since there
`c` is a virtual package.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
On sysroot systems like gentoo prefix, as well as nix/guix, our "is
system path" logic is broken cause it's static.
Talking about "the system paths" is not helpful, we have to talk
about default search paths of the dynamic linker instead.
If glibc is recent enough, we can query the dynamic loader's default
search paths, which is a much more robust way to avoid registering
rpaths to system dirs, which can shadow Spack dirs.
This PR adds an **additional** filter on rpaths the compiler wrapper
adds, dropping rpaths that are default search paths. The PR **does
not** remove any of the original `is_system_path` code yet.
This fixes issues where build systems run just-built executables
linked against their *not-yet-installed libraries*, typically:
```
LD_LIBRARY_PATH=. ./exe
```
which happens in `perl`, `python`, and other non-cmake packages.
If a default path is rpath'ed, it takes precedence over
`LD_LIBRARY_PATH`, and a system library gets loaded instead
of the just-built library in the stage dir, breaking the build. If
default paths are not rpath'ed, then LD_LIBRARY_PATH takes
precedence, as is desired.
This PR additionally fixes an inconsistency in rpaths between
cmake and non-cmake packages. The cmake build system
computed rpaths by itself, but used a different order than
computed for the compiler wrapper. In fact it's not necessary
to compute rpaths at all, since we let cmake do that thanks to
`CMAKE_INSTALL_RPATH_USE_LINK_PATH`. This covers rpaths
for all dependencies. The only install rpaths we need to set are
`<install prefix>/{lib,lib64}`, which cmake unfortunately omits,
although it could also know these. Also, cmake does *not*
delete rpaths added by the toolchain (i.e. Spack's compiler
wrapper), so I don't think it should be controversial to simplify
things.
The current `Spec.splice` model is very limited by the inability to splice specs that
contain multiple nodes with the same name. This is an artifact of the original
algorithm design predating the separate concretization of build dependencies,
which was the first feature to allow multiple specs in a DAG to share a name.
This PR provides a complete reimplementation of `Spec.splice` to avoid that
limitation. At the same time, the new algorithm ensures that build dependencies
for spliced specs are not changed, since the splice by definition cannot change
the build-time information of the spec. This is handled by splitting the dependency
edges and link/run edges into separate dependencies as needed.
Signed-off-by: Gregory Becker <becker33@llnl.gov>
* CI: Add documentation for adding new stacks and runners
* More docs for runner registration
---------
Co-authored-by: Zack Galbreath <zack.galbreath@kitware.com>
Co-authored-by: Bernhard Kaindl <contact@bernhard.kaindl.dev>
This PR shorten the string representation for concrete specs,
in order to make it more legible.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
macOS Sequoia's linker will complain if RPATHs on the CLI are specified more than once.
To avoid errors due to this, make `cc` only append unique RPATHs to the final args list.
This required a few improvements to the logic in `cc`:
1. List functions in `cc` didn't have any way to append unique elements to a list. Add a
`contains()` shell function that works like our other list functions. Use it to implement
an optional `"unique"` argument to `append()` and an `extend_unique()`. Use that to add
RPATHs to the `args_list`.
2. In the pure `ld` case, we weren't actually parsing `RPATH` arguments separately as we
do for `ccld`. Fix this by adding *another* nested case statement for raw `RPATH`
parsing. There are now 3 places where we deal with `-rpath` and friends, but I don't
see a great way to unify them, as `-Wl,`, `-Xlinker`, and raw `-rpath` arguments are
all ever so slightly different.
3. Fix ordering of assertions to make `pytest` diffs more intelligible. The meaning of
`+` and `-` in diffs changed in `pytest` 6.0 and the "preferred" order for assertions
became `assert actual == expected` instead of the other way around.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
`cc` divides most paths up into system paths, spack managed paths, and other paths.
This gets really repetitive and makes the code hard to read. Simplify the script
by adding some functions to do most of the redundant work for us.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
SimpleFilesystemView was producing an error due to looking for a
<prefix>/lib/.spack folder. Also, view_destination had no effect and
wasn't called. Changed this by instead patching in the correct
installation prefix for dictionaries.
Since aspell is using the resolved path of the executable prefix, the
runtime environment variable ASPELL_CONF is set to correct the prefix
when in a view. With this change aspell can now find installed
dictionaries. Verified with:
aspell dump config
aspell dump dicts
If `add_padding()` is allowed to return a path with a trailing path
separator, it will get collapsed elsewhere in Spack. This can lead to
buildcache entries that have RPATHS that are too short to be replaced by
other users whose install root happens to be padded to the correct
length. Detect this and replace the trailing path separator with a
concrete path character.
Signed-off-by: Samuel E. Browne <sebrown@sandia.gov>
We've seen `getfqdn()` cause slowdowns on macOS in CI when added elsewhere. It's also
called by database.py every time we write the DB file.
- [x] replace the call with a memoized version so that it is only called once per process.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
This PR introduces a new heuristic for the solver, which behaves better when
compilers are treated as nodes. Apparently, it performs better also on `develop`,
where compilers are still node attributes.
The new heuristic:
- Sets an initial priority for guessing a few attributes. The order is "nodes" (300),
"dependencies" (150), "virtual dependencies" (60), "version" and "variants" (30), and
"targets" and "compilers" (1). This initial priority decays over time during the solve, and
falls back to the defaults.
- By default, it considers most guessed facts as "false". For instance, by default a node
doesn't exist in the optimal answer set, or a version is not picked as a node version etc.
- There are certain conditions that override the default heuristic using the _priority_ of
a rule, which previously we didn't use. For instance, by default we guess that a
`attr("variant", Node, Variant, Value)` is false, but if we know that the node is already
in the answer set, and the value is the default one, then we guess it is true.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
The `spack.target.Target` class is a weird entity, that is just needed to:
1. Sort microarchitectures in lists deterministically
2. Being able to use microarchitectures in hashed containers
This PR removes it, and uses `archspec.cpu.Microarchitecture` directly. To sort lists, we use a proper `key=` when needed. Being able to use `Microarchitecture` objects in sets is achieved by updating the external `archspec`.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
While the existing getting started guide does in fact reference the
powershell support, it's a footnote and easily missed. This PR adds
explicit, upfront mentions of the powershell support. Additionally
this PR adds notes about some of the issues with certain components
of the spec syntax when using CMD.
Removes `spack.package_base.PackageBase.do_{install,deprecate}` in favor of
`spack.installer.PackageInstaller.install` and `spack.installer.deprecate` resp.
That drops a dependency of `spack.package_base` on `spack.installer`, which is
necessary to get rid of circular dependencies in Spack.
Also change the signature of `PackageInstaller.__init__` from taking a dict as
positional argument, to an explicit list of keyword arguments.
Continuing the work started in #40326, his changes the structure
of Variant metadata on Packages from a single variant definition
per name with a list of `when` specs:
```
name: (Variant, [when_spec, ...])
```
to a Variant definition per `when_spec` per name:
```
when_spec: { name: Variant }
```
With this change, everything on a package *except* versions is
keyed by `when` spec. This:
1. makes things consistent, in that conditional things are (nearly)
all modeled in the same way; and
2. fixes an issue where we would lose information about multiple
variant definitions in a package (see #38302). We can now have,
e.g., different defaults for the same variant in different
versions of a package.
Some notes:
1. This required some pretty deep changes to the solver. Previously,
the solver's job was to select value(s) for a single variant definition
per name per package. Now, the solver needs to:
a. Determine which variant definition should be used for a given node,
which can depend on the node's version, compiler, target, other variants, etc.
b. Select valid value(s) for variants for each node based on the selected
variant definition.
When multiple variant definitions are enabled via their `when=` clause, we will
always prefer the *last* matching definition, by declaration order in packages. This
is implemented by adding a `precedence` to each variant at definition time, and we
ensure they are added to the solver in order of precedence.
This has the effect that variant definitions from derived classes are preferred over
definitions from superclasses, and the last definition within the same class sticks.
This matches python semantics. Some examples:
```python
class ROCmPackage(PackageBase):
variant("amdgpu_target", ..., when="+rocm")
class Hipblas(ROCmPackage):
variant("amdgpu_target", ...)
```
The global variant in `hipblas` will always supersede the `when="+rocm"` variant in
`ROCmPackage`. If `hipblas`'s variant was also conditional on `+rocm` (as it probably
should be), we would again filter out the definition from `ROCmPackage` because it
could never be activated. If you instead have:
```python
class ROCmPackage(PackageBase):
variant("amdgpu_target", ..., when="+rocm")
class Hipblas(ROCmPackage):
variant("amdgpu_target", ..., when="+rocm+foo")
```
The variant on `hipblas` will win for `+rocm+foo` but the one on `ROCmPackage` will
win with `rocm~foo`.
So, *if* we can statically determine if a variant is overridden, we filter it out.
This isn't strictly necessary, as the solver can handle many definitions fine, but
this reduces the complexity of the problem instance presented to `clingo`, and
simplifies output in `spack info` for derived packages. e.g., `spack info hipblas`
now shows only one definition of `amdgpu_target` where before it showed two, one of
which would never be used.
2. Nearly all access to the `variants` dictionary on packages has been refactored to
use the following class methods on `PackageBase`:
* `variant_names(cls) -> List[str]`: get all variant names for a package
* `has_variant(cls, name) -> bool`: whether a package has a variant with a given name
* `variant_definitions(cls, name: str) -> List[Tuple[Spec, Variant]]`: all definitions
of variant `name` that are possible, along with their `when` specs.
* `variant_items() -> `: iterate over `pkg.variants.items()`, with impossible variants
filtered out.
Consolidating to these methods seems to simplify the code a lot.
3. The solver does a lot more validation on variant values at setup time now. In
particular, it checks whether a variant value on a spec is valid given the other
constraints on that spec. This allowed us to remove the crufty logic in
`update_variant_validate`, which was needed because we previously didn't *know* after
a solve which variant definition had been used. Now, variant values from solves are
constructed strictly based on which variant definition was selected -- no more
heuristics.
4. The same prevalidation can now be done in package audits, and you can run:
```
spack audit packages --strict-variants
```
This turns up around 18 different places where a variant specification isn't valid
given the conditions on variant definitions in packages. I haven't fixed those here
but will open a separate PR to iterate on them. I plan to make strict checking the
defaults once all existing package issues are resolved. It's not clear to me that
strict checking should be the default for the prevalidation done at solve time.
There are a few other changes here that might be of interest:
1. The `generator` variant in `CMakePackage` is now only defined when `build_system=cmake`.
2. `spack info` has been updated to support the new metadata layout.
3. split out variant propagation into its own `.lp` file in the `solver` code.
4. Add better typing and clean up code for variant types in `variant.py`.
5. Add tests for new variant behavior.
Openmpi provider statements were changed in #46102. The package change
was fine in and of itself, but apparently one of our tests depends on
the precise constraints used in those statements. I updated the test
to remove the checks for constraints that were removed.
* CUDA: support Grace Hopper 9.0a compute capability
* Fix other packages
* Add type annotations
* Support ancient Python versions
* isort
* spec -> self.spec
Co-authored-by: Andrew W Elble <aweits@rit.edu>
* [@spackbot] updating style on behalf of adamjstewart
---------
Co-authored-by: Andrew W Elble <aweits@rit.edu>
Co-authored-by: adamjstewart <adamjstewart@users.noreply.github.com>
The option config:install_missing_compilers is currently buggy,
and has been for a while. Remove it, since it won't be needed
when compilers are treated as dependencies.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
`spack reindex` relies on projections from configuration to locate
installed specs and prefixes. This is problematic because config can
change over time, and we have reasons to do so when turning compilers
into depedencies (removing `{compiler.name}-{compiler.version}` from
projections)
This commit makes reindex recursively search for .spack/ metadirs.
When a package is running `setup_dependent_package` on a parent, ensure
that module variables like `spack_cc` are available. This was often
true prior to this commit, but externals were an exception.
---------
Co-authored-by: John Parent <john.parent@kitware.com>
Allow flags from different sources (compilers, `require:`, command-line
specs, and `depends_on`) to be merged together, and enforce a consistent
order among them.
The order is based on the sources, e.g. flags on specs from the command
line always come last. Some flag order consistency issues are fixed:
1. Flags from `compilers.yaml` and the command line were always intra- and
inter-source order consistent.
2. Flags from dependents and packages.yaml (introduced via `require:`)
were not: for `-a -b` from one source and `-c` from another, the final
result might rearrange `-a -b`, and would also be inconsistent in terms
of whether `-c` came before or after.
(1) is/was handled by going back to the original source, i.e., flags are
retrieved directly from the command line spec rather than the solver.
(2) is addressed by:
* Keeping track of grouped flags in the solver
* Keeping track of flag sources in the solver on a per-flag basis
The latter info is used in this PR to enforce DAG ordering on flags
applied from multiple dependents to the same package, e.g., for this
graph:
```
a
/|\
b | c
\|/
d
```
If `a`, `b`, and `c` impose flags on `d`, the combined flags on `d` will
contain the flags of `a`, `b`, and `c` -- in that order.
Conflicting flags are allowed (e.g. -O2 and -O3). `Spec.satisifes()` has
been updated such that X satisfies Y as long as X has *at least* all of
the flags that Y has. This is also true in the solver constraints.
`.satisfies` does not account for how order can change behavior (so
`-O2 -O3` can satisfy `-O3 -O2`); it is expected that this can be
addressed later (e.g. by prohibiting flag conflicts).
`Spec.constrain` and `.intersects` have been updated to be consistent
with this new definition of `.satisfies`.
Spack can now bootstrap two new dependencies on Windows: GnuPG, and file.
These dependencies are modeled as a separate package, and they install a cross-compiled binary.
Details on how they binaries are built are in https://github.com/spack/windows-bootstrap-resources
* Allow deprecating more than one property in config
This internal change allows the customization of errors
and warnings to be printed when deprecating a property.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
* fix
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
* Use a list comprehension for "issues"
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
---------
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
fixes#40791
Currently stacks behave differently if used in unify:false
environments, which leads to inconsistencies during concretization.
For instance, we might have two abstract user specs that do not
intersect with each other map to the same concrete spec in the
environment. This is clearly wrong.
This PR removes the best effort expansion, so that user specs
are always applied strictly.
