This PR improves compatibility with specs installed before #45189, and with externals specifying a compiler, by using the annotated compiler to "satisfy" a spec query.
On top of that, the PR adds a new flag for:
```console
$ spack find --specfile-format -I %gcc
-- linux-ubuntu20.04-icelake / gcc@10.5.0 -----------------------
[+] [v4] ca-certificates-mozilla@2023-05-30 [e] [v4] cmake@3.31.6 [+] [v4] gcc-runtime@10.5.0 [e] [v4] glibc@2.31 [+] [v4] gmake@4.4.1 [+] [v4] hdf5@1.14.5 [+] [v4] pkgconf@2.2.0 [+] [v4] zlib-ng@2.2.1
==> 8 installed packages
```
which shows the specfile format of the specs being retrieved.
Fixes a logic bug where a -> b was assumed to imply not a -> not b
in conditional requirements.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
When trying to use Spack to build Intel TBB on the EOS network file
system we use at CERN, I am facing the following error:
```
Launching the installer...
Installation directory is not empty.
The product cannot be installed into nonempty directory
'/eos/home-s/sswatman/spack/opt/spack/linux-x86_64_v2/...
intel-oneapi-tbb-2021.12.0-3jlx6hlr3z6di42f3qy35eizcse7u2tk'.
```
This error appears to happen because Spack tries to set `$HOME` to the
prefix of the package, into which the Intel installer is also trying to
install it's software. I've found that an easy fix is to set `$HOME` to
a directory in the build stage instead, as this can be guaranteed to
remain empty.
* `x=*` constrained by `+x` now produces a boolean valued variant instead of a multi-valued variant.
* Values are now always stored as a tuple internally, whether bool, single or multi-valued.
* Value assignment has a stricter api to prevent ambiguity / type issues related to
`variant.value = "x"` / `variant.value = ["x"]` / `variant.value = ("x",)`. It's now `variant.set("x", ...)` for
single and multi-valued variants.
* The `_original_value` prop is dropped, since it was unused.
* The wildcard `*` is no longer a possible variant value in any type of variant, since the *parser*
deals with it and creates a variant with no values.
The second change technically affects non-Windows, but the
behavior should be exactly the same:
* Packages no longer have access to `.msbuild` and `.nmake`
automatically; they now get them via a dependency on `msvc`.
* Update two CMake-based packages that call `make test` to
instead call `ctest` (`netcdf-cxx4` and `pegtl`).
CMake-based packages should do this because on Windows
`make test` will not generally work, but `ctest` does.
* Fix `openssl` "make test" on Windows (WRT prior point: not
a CMake-based package).
Show progress meter for fetches when `stdout` is a `tty`.
* fetch_strategy.py: show progress
* "Fetched: x MB at y MB/s"
* add tests, show % if content-length
This PR fixes the issues with `%` and reused specs, due to https://github.com/spack/spack/issues/49847#issuecomment-2774640234
It does so by adding another layer of indirection, so that whenever a spec
`foo %bar` is encountered, the `%bar` part is encoded as an
`attr("build_requirement", ...)`.
Then:
1. If `foo` is a node to be built, then the build requirement implies a dependency
2. Otherwise it implies looking e.g. reused specs metadata, and ensure it matches
---------
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
When the solver produces specs that do not satisfy the input
constraints, dump both input and output specs as json in an temporary
dir and ask the user to upload these files in a bug report.
Currently we inject runtimes when a package has a direct build dep on a
compiler, but what matters is whether the package depends on a language.
That way we can avoid recursion of injecting runtimes to runtimes
without a rule in the solver: runtimes don't depend on languages, they
just have a build dep on the same compiler.
In this way, to run them, we just need to run:
spack unit-test lib/spack/spack/test/concretization
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
* TestSuite: add type hints
* spack test run: add a --timeout argument
* pipelines: allow 2 minutes to run tests
* Fix docstrings, increase maximum pipelines time for tests to 5 mins.
* Use SIGTERM first, SIGKILL shortly after
* Add unit-tests for "start_build_process"
---------
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Multiple build systems have been part of Spack for a long
time now, and they are rarely the cause of a missing attribute.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Similar to the range-or-specific-version ambiguity of `@1.2` in the past,
which was solved with `@1.2` vs `@=1.2` we still have the ambiguity of
`name=a,b,c` in multi-valued variants. Do they mean "at least a,b,c" or
"exactly a,b,c"?
This issue comes up in for example `gcc languages=c,cxx`; there's no
way to exclude `fortran`.
The ambiguity is resolved with syntax `:=` to distinguish concrete from
abstract.
The following strings parse as **concrete** variants:
* `name:=a,b,c` => values exactly {a, b, c}
* `name:=a` => values exactly {a}
* `+name` => values exactly {True}
* `~name` => values exactly {False}
The following strings parse as **abstract** variants:
* `name=a,b,c` values at least {a, b, c}
* `name=*` special case for testing existence of a variant; values are at
least the empty set {}
As a reminder
* `satisfies(lhs, rhs)` means `concretizations(lhs)` ⊆ `concretizations(rhs)`
* `intersects(lhs, rhs)` means `concretizations(lhs)` ∩ `concretizations(rhs)` ≠ ∅
where `concretizations(...)` is the set of sets of variant values in this case.
The satisfies semantics are:
* rhs abstract: rhs values is a subset of lhs values (whether lhs is abstract or concrete)
* lhs concrete, rhs concrete: set equality
* lhs abstract, rhs concrete: false
and intersects should mean
* lhs and rhs abstract: true (the union is a valid concretization under both)
* lhs or rhs abstract: true iff the abstract variant's values are a subset of the concrete one
* lhs concrete, rhs concrete: set equality
Concrete specs with single-valued variants are printed `+foo`, `~foo` and `foo=bar`;
only multi-valued variants are printed with `foo:=bar,baz` to reduce the visual noise.
Return a single scope from environment.env_config_scope
Return a single scope from environment_path_scope
---------
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
This commit reorders ASP setup, so that rules from
possible compilers are collected first.
This allows us to know the dependencies that may be
injected before counting the possible dependencies,
so we can account for them too.
Proceeding this way makes it easier to inject
complex runtimes, like hip.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Deal with the "issue" that passing a str instance does not cause a
type check failure, because str is a subset of Sequence[str] and
Iterable[str]. Instead fix it by special casing the str instance.
* Check for LSF, FLux, and Slurm when determing MPI exec
* Make scheduler/MPI exec helper functions methods of CachedCMakeBuilder
* Remove axom workaround for running mpi on machines with flux
* Clearly split old and new hip settings requirements
* Apply generic rocm handling to every project
* make default logic for hip support more robust
* GPU_TARGET is only necessary under certain project specific conditions, it should not be necessary in general
* Update logic to find amdclang++
fixes#49717
If no compiler is listed in the 'packages' section of
the configuration, Spack will currently try to:
1. Look for a legacy compilers.yaml to convert
2. Look for compilers in PATH
in that order. If an entry in compilers.yaml is
corrupted, that should not result in an obscure
error.
Instead, it should just be skipped.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
This module references `spack.error.Something` in the same file, which happens to
work but is incorrect. Use `spack.error` to prevent that in the future.
I noticed that `abseil-cpp` was showing in `spack find` with "no compiler", and the only
difference between it and other nodes was that it *only* depends on `cxx` -- others
depend on `c` as well.
It turns out that the `select()` method on `EdgeMap` only takes `Sequence[str]` and doesn't
check whether they're actually just one `str`. So asking for, e.g., `cxx` is like asking for
`c` or `x` or `x`, as the `str` is treated like a sequence. This causes Spack to miss `cxx`
and `fortran` language virtuals in `DeprecatedCompilerSpec`.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Currently, externals show up in `spack find` and `spack spec` install status as a green
`[e]`, which is hard to distinguish from the green [+] used for installed packages.
- [x] Make externals magenta instead, so they stand out.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Concretization setup was checking whether any input spec has a dependency
that's *not* in the set of possible dependencies for all roots in the solve.
There are two reasons to check this:
1. The user could be asking for a dependency that none of the roots has, or
2. The user could be asking for a dependency that doesn't exist.
For abstract roots, (2) implies (1), and the check makes sense. For concrete
roots, we don't care, because the spec has already been built. If a `package.py`
no longer depends on something it did before, it doesn't matter -- it's already
built. If the dependency no longer exists, we also do not care -- we already
built it and there's an installation for it somewhere.
When you concretize an environment with a lockfile, *many* of the input specs
are concrete, and we don't need to build them. If a package changes its
dependencies, or if a `package.py` is removed for a concrete input spec, that
shouldn't cause an already-built environment to fail concretization.
A user reported that this was happening with an error like:
```console
spack concretize
==> Error: Package chapel does not depend on py-protobuf@5.28.2/a4rf4glr2tntfwsz6myzwmlk5iu25t74
```
Or, with traceback:
```console
File "/apps/other/spack-devel/lib/spack/spack/solver/asp.py", line 3014, in setup
raise spack.spec.InvalidDependencyError(spec.name, missing_deps)
spack.spec.InvalidDependencyError: Package chapel does not depend on py-protobuf@5.28.2/a4rf4glr2tntfwsz6myzwmlk5iu25t74
```
Fix this by skipping the check for concrete input specs. We already ignore conflicts,
etc. for concrete/external specs, and we do not need metadata in the solve for
concrete dependencies b/c they're imposed by hash constraints.
- [x] Ignore the package existence check for concrete input specs.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
* Skip packages removed for automatic checksum verification
* Unify finding modified or added packages with spack.repo logic
* Remove unused imports
* Fix unit-tests using shared modified function
* Update last remaining unit test to new format
## Summary
Compilers stop being a *node attribute*, and become a *build-only* dependency.
Packages may declare a dependency on the `c`, `cxx`, or `fortran` languages, which
are now treated as virtuals, and compilers would be *providers* for one or more of
those languages. Compilers can also inject runtime dependency, on the node being
compiled. An example graph for something as simple as `zlib-ng` is the following:
<p align="center">
<img src="https://github.com/user-attachments/assets/ee6471cb-09fd-4127-9f16-b9fe6d1338ac" alt="zlib-ng DAG" width="80%" height="auto">
</p>
Here `gcc` is used for both the `c`, and `cxx` languages. Edges are annotated with
the virtuals they satisfy (`c`, `cxx`, `libc`). `gcc` injects `gcc-runtime` on the nodes
being compiled. `glibc` is also injected for packages that require `c`. The
`compiler-wrapper` is explicitly represented as a node in the DAG, and is included in
the hash.
This change in the model has implications on the semantics of the `%` sigil, as
discussed in #44379, and requires a version bump for our `Specfile`, `Database`,
and `Lockfile` formats.
## Breaking changes
Breaking changes below may impact users of this branch.
### 1. Custom, non-numeric version of compilers are not supported
Currently, users can assign to compilers any custom version they want, and Spack
will try to recover the "real version" whenever the custom version fails some operation.
To deduce the "real version" Spack must run the compiler, which can add needless
overhead to common operations.
Since any information that a version like `gcc@foo` might give to the user, can also
be suffixed while retaining the correct numeric version, e.g. `gcc@10.5.0-foo`, Spack
will **not try** anymore to deduce real versions for compilers.
Said otherwise, users should have no expectation that `gcc@foo` behaves as
`gcc@X.Y.Z` internally.
### 2. The `%` sigil in the spec syntax means "direct build dependency"
The `%` sigil in the spec syntax means *"direct build dependency"*, and is not a node
attribute anymore. This means that:
```python
node.satisfies("%gcc")
```
is true only if `gcc` is a direct build dependency of the node. *Nodes without a compiler
dependency are allowed.*
### `parent["child"]`, and `node in spec`, will now only inspect the link/run sub-DAG
and direct build dependencies
The subscript notation for `Spec`:
```python
parent["child"]
```
will look for a `child` node only in the link/run transitive graph of `parent`, and in its
direct build dependencies. This means that to reach a transitive build dependency,
we must first pass through the node it is associated with.
Assuming `parent` does not depend on `cmake`, but depends on a `CMakePackage`,
e.g. `hdf5`, then we have the following situation:
```python
# This one raises an Exception, since "parent" does not depend on cmake
parent["cmake"]
# This one is ok
cmake = parent["hdf5"]["cmake"]
```
### 3. Externals differing by just the compiler attribute
Externals are nodes where dependencies are trimmed, and that _is not planned to
change_ in this branch. Currently, on `develop` it is ok to write:
```yaml
packages:
hdf5:
externals:
- spec: hdf5@1.12 %gcc
prefix: /prefix/gcc
- spec: hdf5@1.12 %clang
prefix: /prefix/clang
```
and Spack will account for the compiler node attribute when computing the optimal
spec. In this branch, using externals with a compiler specified is allowed only if any
compiler in the dag matches the constraints specified on the external. _The external
will be still represented as a single node without dependencies_.
### 4. Spec matrices enforcing a compiler
Currently we can have matrices of the form:
```yaml
matrix:
- [x, y, z]
- [%gcc, %clang]
```
to get the cross-product of specs and compilers. We can disregard the nature of the
packages in the first row, since the compiler is a node attribute required on each node.
In this branch, instead, we require a spec to depend on `c`, `cxx`, or `fortran` for the
`%` to have any meaning. If any of the specs in the first row doesn't depend on these
languages, there will be a concretization error.
## Deprecations
* The entire `compilers` section in the configuration (i.e., `compilers.yaml`) has been
deprecated, and current entries will be removed in v1.2.0. For the time being, if Spack
finds any `compilers` configuration, it will try to convert it automatically to a set of
external packages.
* The `packages:compiler` soft-preference has been deprecated. It will be removed
in v1.1.0.
## Other notable changes
* The tokens `{compiler}`, `{compiler.version}`, and `{compiler.name}` in `Spec.format`
expand to `"none"` if a Spec does not depend on C, C++, or Fortran.
* The default install tree layout is now
`"{architecture.platform}-{architecture.target}/{name}-{version}-{hash}"`
## Known limitations
The major known limitations of this branch that we intend to fix before v1.0 is that compilers
cannot be bootstrapped directly.
In this branch we can build a new compiler using an existing external compiler, for instance:
```
$ spack install gcc@14 %gcc@10.5.0
```
where `gcc@10.5.0` is external, and `gcc@14` is to be built.
What we can't do at the moment is use a yet to be built compiler, and expect it will be
bootstrapped, e.g. :
```
spack install hdf5 %gcc@14
```
We plan to tackle this issue in a following PR.
---------
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Signed-off-by: Harmen Stoppels <me@harmenstoppels.nl>
Co-authored-by: Harmen Stoppels <me@harmenstoppels.nl>
Co-authored-by: Todd Gamblin <tgamblin@llnl.gov>
Since we moved from creating clingo symbols directly to constructing a pure string
representation of the program, we don't need to make `AspFunctions` into symbols before
turning them into strings. We can just write strings like clingo would.
This cuts about 25% off the setup time by avoiding an unnecessary round trip.
- [x] create strings directly from `AspFunctions`
- [x] remove unused `symbol()` method on `AspFunction`
- [x] setup no longer tries to call `symbol()`
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Signed-off-by: Greg Becker <becker33@llnl.gov>
---------
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Co-authored-by: Greg Becker <becker33@llnl.gov>
* Add recursive argument to spack develop
This effort allows for a recursive develop call
which will traverse from the develop spec given back to the root(s)
and mark all packages along the path as develop.
If people are doing development across the graph then paying
fetch and full rebuild costs every time spack develop is called
is unnecessary and expensive.
Also 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.
Add a CI check to automatically verify the checksums of newly added
package versions:
- [x] a new command, `spack ci verify-versions`
- [x] a GitHub actions check to run the command
- [x] tests for the new command
This also eliminates the suggestion for maintainers to manually verify added
checksums in the case of accidental version <--> checksum mismatches.
----
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
The package was added in 2017, and never updated
substantially. It requires users to login into
a platform to download code.
Thus, instead of updating to new versions, and add
support for OneAPI, remove the package.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Fixes#49403.
When one scope included another, we were appending to a list stored on the scope to
track what was included, and we would clear the list when the scope was removed.
This assumes that the scopes are always strictly pushed then popped, but the order can
be violated when serializing config scopes across processes (and then activating
environments in subprocesses), or if, e.g., instead of removing each scope we simply
cleared the list of config scopes. Removal can be skipped, which can cause the list of
includes on a cached scope (like the one we use for environments) to grow every time it
is pushed, and this triggers an assertion error.
There isn't actually a need to construct and destroy the include list. We can just
compute it once and cache it -- it's the same every time.
- [x] Cache included scope list on scope objects
- [x] Do not dynamically append/clear the included scope list
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Right now the Spack %msvc compiler is inherently a hybrid compiler
that uses Intel's oneAPI fortran compiler.
This was addressed in Spacks MSVC compiler class, but detection has
since stopped using the compiler class, so this PR moves the logic
into the `msvc` compiler package (does not delete the original code
because that is handled in #45189).
This includes a change to the general detection logic to deprioritize
paths that include a symlink anywhere in the path, in order to prefer
"2025.0/bin" over "latest/bin" for the oneAPI compiler.
* style.py: add spack style --spec-strings for compat with v1.0
* add --fix also, and avoid infinite recursion and too large files
* tests: check identify and check edit files
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.
