Updates to environments were not multi-process safe, which prevented them from taking advantage of parallel builds as implemented in #13100. This is a minimal set of changes to enable `spack install` in an environment to be parallelized:
- [x] add an internal lock, stored in the `.spack-env` directory,
to synchronize updates to `spack.yaml` and `spack.lock`
- [x] add `Environment.write_transaction` interface for this lock
- [x] makes use of `Environment.write_transaction` in `install`,
`add`, and `remove` commands
- `uninstall` is not synchronized yet; that is left for a future PR.
Spack commands referring to upstream-installed specs by hash have
been broken since 6b619da (merged September 2019), which added a new
Database function specifically for parsing hashes from command-line
specs; this function was inappropriately attempting to acquire locks
on upstream databases.
This PR updates the offending function to avoid locking upstream
databases and also updates associated tests to catch regression
errors: the upstream database created for these tests was not
explicitly set as an upstream (i.e. initialized with upstream=True)
so it was not guarding against inappropriate accesses.
* Unified environment modifications in config files
fixes#13357
This commit factors all the code that is involved in
the validation (schema) and parsing of environment modifications
from configuration files in a single place. The factored out
code is then used for module files and compiler configuration.
Attributes were separated by dashes in `compilers.yaml` files and
by underscores in `modules.yaml` files. This PR unifies the syntax
on attributes separated by underscores.
Unit testing of environment modifications in compilers
has been refactored and simplified.
Using `sys.executable` to run Python in a sub-shell doesn't always work in a virtual environment as the `sys.executable` Python is not necessarily compatible with any loaded spack/other virtual environment.
- revert use of sys.executable to print out subshell environment (#14496)
- try instead to use an available python, then if there *is not* one, use `sys.executable`
- this addresses RHEL8 (where there is no `python` and `PYTHONHOME` issue in a simpler way
Openblas target is now determined automatically upon inspection of
`TargetList.txt`. If the spack target is a generic architecture family
(like x86_64 or aarch64) the DYNAMIC_ARCH setting is used
instead of targeting a specific microarchitecture.
Instead of another script, this adds a simple argument to `spack
commands` that updates the completion script. Developers can now just
run:
spack commands --update-completion
This should make it simpler for developers to remember to run this
*before* the tests fail. Also, this version tab-completes.
Previously the `spack load` command was a wrapper around `module load`. This required some bootstrapping of modules to make `spack load` work properly.
With this PR, the `spack` shell function handles the environment modifications necessary to add packages to your user environment. This removes the dependence on environment modules or lmod and removes the requirement to bootstrap spack (beyond using the setup-env scripts).
Included in this PR is support for MacOS when using Apple's System Integrity Protection (SIP), which is enabled by default in modern MacOS versions. SIP clears the `LD_LIBRARY_PATH` and `DYLD_LIBRARY_PATH` variables on process startup for executables that live in `/usr` (but not '/usr/local', `/System`, `/bin`, and `/sbin` among other system locations. Spack cannot know the `LD_LIBRARY_PATH` of the calling process when executed using `/bin/sh` and `/usr/bin/python`. The `spack` shell function now manually forwards these two variables, if they are present, as `SPACK_<VAR>` and recovers those values on startup.
- [x] spack load/unload no longer delegate to modules
- [x] refactor user_environment modification calculations
- [x] update documentation for spack load/unload
Co-authored-by: Todd Gamblin <tgamblin@llnl.gov>
This PR adds a `--format=bash` option to `spack commands` to
auto-generate the Bash programmable tab completion script. It can be
extended to work for other shells.
Progress:
- [x] Fix bug in superclass initialization in `ArgparseWriter`
- [x] Refactor `ArgparseWriter` (see below)
- [x] Ensure that output of old `--format` options remains the same
- [x] Add `ArgparseCompletionWriter` and `BashCompletionWriter`
- [x] Add `--aliases` option to add command aliases
- [x] Standardize positional argument names
- [x] Tests for `spack commands --format=bash` coverage
- [x] Tests to make sure `spack-completion.bash` stays up-to-date
- [x] Tests for `spack-completion.bash` coverage
- [x] Speed up `spack-completion.bash` by caching subroutine calls
This PR also necessitates a significant refactoring of
`ArgparseWriter`. Previously, `ArgparseWriter` was mostly a single
`_write` method which handled everything from extracting the information
we care about from the parser to formatting the output. Now, `_write`
only handles recursion, while the information extraction is split into a
separate `parse` method, and the formatting is handled by `format`. This
allows subclasses to completely redefine how the format will appear
without overriding all of `_write`.
Co-Authored-by: Todd Gamblin <tgamblin@llnl.gov>
The gpg2 command isn't always around; it's sometimes called gpg. This is
the case with the brew-installed version, and it's breaking our tests.
- [x] Look for both 'gpg2' and 'gpg' when finding the command
- [x] If we find 'gpg', ensure the version is 2 or higher
- [x] Add tests for version detection.
- [x] Factored to a common place the fixture `testing_gpg_directory`, renamed it as
`mock_gnupghome`
- [x] Removed altogether the function `has_gnupg2`
For `has_gnupg2`, since we were not trying to parse the version from the output of:
```console
$ gpg2 --version
```
this is effectively equivalent to check if `spack.util.gpg.GPG.gpg()` was found. If we need to ensure version is `^2.X` it's probably better to do it in `spack.util.gpg.GPG.gpg()` than in a separate function.
Despite trying very hard to keep dicts out of our hash algorithm, we seem
to still accidentally add them in ways that the tests can't catch. This
can cause errors when hashes are not computed deterministically.
This fixes an error we saw with Python 3.5, where dictionary iteration
order is random. In this instance, we saw a bug when reading Spack
environment lockfiles -- The load would fail like this:
```
...
File "/sw/spack/lib/spack/spack/environment.py", line 1249, in concretized_specs
yield (s, self.specs_by_hash[h])
KeyError: 'qcttqplkwgxzjlycbs4rfxxladnt423p'
```
This was because the hashes differed depending on whether we wrote `path`
or `module` first when recomputing the build hash as part of reading a
Spack lockfile. We can fix it by ensuring a determistic iteration order.
- [x] Fix two places (one that caused an issue, and one that did
not... yet) where our to_node_dict-like methods were using regular python
dicts.
- [x] Also add a check that statically analyzes our to_node_dict
functions and flags any that use Python dicts.
The test found the two errors fixed here, specifically:
```
E AssertionError: assert [] == ['Use syaml_dict instead of ...pack/spack/spec.py:1495:28']
E Right contains more items, first extra item: 'Use syaml_dict instead of dict at /Users/gamblin2/src/spack/lib/spack/spack/spec.py:1495:28'
E Full diff:
E - []
E + ['Use syaml_dict instead of dict at '
E + '/Users/gamblin2/src/spack/lib/spack/spack/spec.py:1495:28']
```
and
```
E AssertionError: assert [] == ['Use syaml_dict instead of ...ack/architecture.py:359:15']
E Right contains more items, first extra item: 'Use syaml_dict instead of dict at /Users/gamblin2/src/spack/lib/spack/spack/architecture.py:359:15'
E Full diff:
E - []
E + ['Use syaml_dict instead of dict at '
E + '/Users/gamblin2/src/spack/lib/spack/spack/architecture.py:359:15']
```
Rework Spack's continuous integration workflow to be environment-based.
- Add the `spack ci` command, which replaces the many scripts in `bin/`
- `spack ci` decouples the CI workflow from the spack instance:
- CI is defined in a spack environment
- environment is in its own (single) git repository, separate from Spack
- spack instance used to run the pipeline is up to the user
- A new `gitlab-ci` section in environments allows users to configure how
specs in the environment should be mapped to runners
- Compilers can be bootstrapped in the new pipeline workflow
- Add extensive documentation on pipelines (see `pipelines.rst` for further details)
- Add extensive tests for pipeline code
* Reorder GNU mirrors (#14395)
As @adamjstewart commented in #14395, GNU suggests to use
their mirror. So reorder the mirror to the top.
GNU Doc: https://www.gnu.org/prep/ftp.en.html
* Use spack.util.url.join for URLs in GNU mirrors (#14395)
One should not use os.path.join for URLs. This does only
work on POSIX systems.
Instead use spack.util.url.join.
So every part in spack uses the same url joining method.
When removing packages from a view, extensions were being deactivated
in an arbitrary order. Extensions must be deactivated in preorder
traversal (dependents before dependencies), so when this order was
violated the view update would fail.
This commit ensures that views deactivate extensions based on a
preorder traversal and adds a test for it.
* Spack can uninstall unused specs
fixes#4382
Added an option to spack uninstall that removes all unused specs i.e.
build dependencies or transitive dependencies that are left
in the store after the specs that pulled them in have been removed.
* Moved the functionality to its own command
The command has been named 'spack autoremove' to follow the naming used
for the same functionality by other widely known package managers i.e.
yum and apt.
* Speed-up autoremoving specs by not locking and re-reading the scratch DB
* Make autoremove work directly on Spack's store
* Added unit tests for the new command
* Display a terser output to the user
* Renamed the "autoremove" command "gc"
Following discussion there's more consensus around
the latter name.
* Preserve root specs in env contexts
* Instead of preserving specs, restrict gc to the active environment
* Added docs
* Added a unit test for gc within an environment
* Updated copyright to 2020
* Updated documentation according to review
Rephrased a couple of sentences, added references to
`spack find` and dependency types.
* Updated function naming and docstrings
* Simplified computation of unused specs
Since the new approach uses private attributes of the DB
it has been coded as a method of that class rather than a
freestanding function.
The imports in `spec.py` are getting to be pretty unwieldy.
- [x] Remove all of the `import from` style imports and replace them with
`import` or `import as`
- [x] Remove a number names that were exported by `spack.spec` that
weren't even in `spack.spec`
Previously, `spack test` automatically passed all of its arguments to
`pytest -k` if no options were provided, and to `pytest` if they were.
`spack test -l` also provided a list of test filenames, but they didn't
really let you completely narrow down which tests you wanted to run.
Instead of trying to do our own weird thing, this passes `spack test`
args directly to `pytest`, and omits the implicit `-k`. This means we
can now run, e.g.:
```console
$ spack test spec_syntax.py::TestSpecSyntax::test_ambiguous
```
This wasn't possible before, because we'd pass the fully qualified name
to `pytest -k` and get an error.
Because `pytest` doesn't have the greatest ability to list tests, I've
tweaked the `-l`/`--list`, `-L`/`--list-long`, and `-N`/`--list-names`
options to `spack test` so that they help you understand the names
better. you can combine these options with `-k` or other arguments to do
pretty powerful searches.
This one makes it easy to get a list of names so you can run tests in
different orders (something I find useful for debugging `pytest` issues):
```console
$ spack test --list-names -k "spec and concretize"
cmd/env.py::test_concretize_user_specs_together
concretize.py::TestConcretize::test_conflicts_in_spec
concretize.py::TestConcretize::test_find_spec_children
concretize.py::TestConcretize::test_find_spec_none
concretize.py::TestConcretize::test_find_spec_parents
concretize.py::TestConcretize::test_find_spec_self
concretize.py::TestConcretize::test_find_spec_sibling
concretize.py::TestConcretize::test_no_matching_compiler_specs
concretize.py::TestConcretize::test_simultaneous_concretization_of_specs
spec_dag.py::TestSpecDag::test_concretize_deptypes
spec_dag.py::TestSpecDag::test_copy_concretized
```
You can combine any list option with keywords:
```console
$ spack test --list -k microarchitecture
llnl/util/cpu.py modules/lmod.py
```
```console
$ spack test --list-long -k microarchitecture
llnl/util/cpu.py::
test_generic_microarchitecture
modules/lmod.py::TestLmod::
test_only_generic_microarchitectures_in_root
```
Or just list specific files:
```console
$ spack test --list-long cmd/test.py
cmd/test.py::
test_list test_list_names_with_pytest_arg
test_list_long test_list_with_keywords
test_list_long_with_pytest_arg test_list_with_pytest_arg
test_list_names
```
Hopefully this stuff will help with debugging test issues.
- [x] make `spack test` send args directly to `pytest` instead of trying
to do fancy things.
- [x] rework `--list`, `--list-long`, and add `--list-names` to make
searching for tests easier.
- [x] make it possible to mix Spack's list args with `pytest` args
(they're just fancy parsing around `pytest --collect-only`)
- [x] add docs
- [x] add tests
- [x] update spack completion
Test configuration files (except modules.yaml) were in the root level of
test/data, but should really just be in their own directory. The absence
of modules.yaml was also breaking module tests if we got module
preferences after tests started, as the mock modules.yaml was not in the
test directory.
The module hook would previously fail if there were no enabled module types.
- Instead of looking for a `KeyError`, default to empty list when the
config variable is not present.
- Convert lambdas to real functions for clarity.
- Remove legacy yaml_version_check() hook
- Remove the pre_run hook from `hook/__init__.py` and `main.py`
We want to discourage the use of pre-run hooks because they have to run
at startup. To keep Spack fast, we should do things like this lazily
instead of in hooks that require spidering directories full of modules.
Continuing to shave small bits of time off startup --
`spack.cmd.common.arguments` constructs many `Args` objects at module
scope, which has to be done for all commands that import it. Instead of
doing this at load time, do it lazily.
- [x] construct Args objects lazily
- [x] remove the module-scoped argparse fixture
- [x] make the mock config scope set dirty to False by default (like the
regular scope)
This *seems* to reduce load time slightly
Previously, fixtures like `config`, `database`, and `store` were
module-scoped, but frequently used as test function arguments. These
fixtures swap out global on setup and restore them on teardown. As
function arguments, they would do the right set-up, but they'd leave the
global changes in place for the whole module the function lived in. This
meant that if you use `config` once, other functions in the same module
would inadvertently inherit the mock Spack configuration, as it would
only be torn down once all tests in the module were complete.
In general, we should module- or session-scope the *STATE* required for
these global objects (as it's expensive to create0, but we shouldn't
module-or session scope the activation/use of them, or things can get
really confusing.
- [x] Make generic context managers for global-modifying fixtures.
- [x] Make session- and module-scoped fixtures that ONLY build filesystem
state and create objects, but do not swap out any variables.
- [x] Make seeparate function-scoped fixtures that *use* the session
scoped fixtures and actually swap out (and back in) the global
variables like `config`, `database`, and `store`.
These changes make it so that global changes are *only* ever alive for a
singlee test function, and we don't get weird dependencies because a
global fixture hasn't been destroyed.
`PackagePrefs` has had a class-level cache of data from `packages.yaml` for
a long time, but it complicates testing and leads to subtle errors,
especially now that we frequently manipulate custom config scopes and
environments.
Moving the cache to instance-level doesn't slow down concretization or
the test suite, and it just caches for the life of a `PackagePrefs`
instance (i.e., for a single cocncretization) so we don't need to worry
about global state anymore.
- [x] Remove class-level caches from `PackagePrefs`
- [x] Add a cached _spec_order object on each `PackagePrefs` instance
- [x] Remove all calls to `PackagePrefs.clear_caches()`
Commands like `spack blame` were printig poorly when redirected to files,
as colify reverts to a single column when redirected. This works for
list data but not tables.
- [x] Force a table by always passing `tty=True` from `colify_table()`
In "spack info" the Variants header currently has two blank
lines under it. That's too much. It looks like the actual
content belongs to something else.
Instead underline the headers to make things more obvious.
This commit removes the `python_version.py` unit test module
and the vendored dependencies `pyqver2.py` and `pyqver3.py`.
It substitutes them with an equivalent check done using
`vermin` that is run as a separate workflow via Github Actions.
This allows us to delete 2 vendored dependencies that are unmaintained
and substitutes them with a maintained tool.
Also, updates the list of vendored dependencies.
`ViewDescriptor.regenerate()` calls `get_all_specs()`, which reads
`spec.yaml` files, which is slow. It's fine to do this once, but
`view.remove_specs()` *also* calls it immediately afterwards.
- [x] Pass the result of `get_all_specs()` as an optional parameter to
`view.remove_specs()` to avoid reading `spec.yaml` files twice.
`ViewDescriptor.regenerate()` was copying specs and stripping build
dependencies, which clears `_hash` and other cached fields on concrete
specs, which causes a bunch of YAML hashes to be recomputed.
- [x] Preserve the `_hash` and `_normal` fields on stripped specs, as
these will be unchanged.
`spack install` previously concretized, writes the entire environment
out, regenerated views, then wrote and regenerated views
again. Regenerating views is slow, so ensure that we only do that once.
- [x] add an option to env.write() to skip view regeneration
- [x] add a note on whether regenerate_views() shouldn't just be a
separate operation -- not clear if we want to keep it as part of write
to ensure consistency, or take it out to avoid performance issues.
Environments need to read the DB a lot when installing all specs.
- [x] Put a read transaction around `install_all()` and `install()`
to avoid repeated locking
Our `LockTransaction` class was reading overly aggressively. In cases
like this:
```
1 with spack.store.db.read_transaction():
2 with spack.store.db.write_transaction():
3 ...
```
The `ReadTransaction` on line 1 would read in the DB, but the
WriteTransaction on line 2 would read in the DB *again*, even though we
had a read lock the whole time. `WriteTransaction`s were only
considering nested writes to decide when to read, but they didn't know
when we already had a read lock.
- [x] `Lock.acquire_write()` return `False` in cases where we already had
a read lock.
If a write transaction was nested inside a read transaction, it would not
write properly on release, e.g., in a sequence like this, inside our
`LockTransaction` class:
```
1 with spack.store.db.read_transaction():
2 with spack.store.db.write_transaction():
3 ...
4 with spack.store.db.read_transaction():
...
```
The WriteTransaction on line 2 had no way of knowing that its
`__exit__()` call was the last *write* in the nesting, and it would skip
calling its write function.
The `__exit__()` call of the `ReadTransaction` on line 1 wouldn't know
how to write, and the file would never be written.
The DB would be correct in memory, but the `ReadTransaction` on line 4
would re-read the whole DB assuming that other processes may have
modified it. Since the DB was never written, we got stale data.
- [x] Make `Lock.release_write()` return `True` whenever we release the
*last write* in a nest.
Lock transactions were actually writing *after* the lock was
released. The code was looking at the result of `release_write()` before
writing, then writing based on whether the lock was released. This is
pretty obviously wrong.
- [x] Refactor `Lock` so that a release function can be passed to the
`Lock` and called *only* when a lock is really released.
- [x] Refactor `LockTransaction` classes to use the release function
instead of checking the return value of `release_read()` / `release_write()`
`ViewDescriptor.regenerate()` checks repeatedly whether packages are
installed and also does a lot of DB queries. Put a read transaction
around the whole thing to avoid repeatedly locking and unlocking the DB.
`Environment.added_specs()` has a loop around calls to
`Package.installed()`, which can result in repeated DB queries. Optimize
this with a read transaction in `Environment`.
Checks for deprecated specs were repeatedly taking out read locks on the
database, which can be very slow.
- [x] put a read transaction around the deprecation check
BundlePackages use a noop fetch strategy. The mirror logic was assuming
that the fetcher had a resource to cach after performing a fetch. This adds
a special check to skip caching if the stage is associated with a
BundleFetchStrategy. Note that this should allow caching resources
associated with BundlePackages.
When updating a mirror, Spack was re-retrieving all patches (since the
fetch logic for patches is separate). This updates the patch logic to
allow the mirror logic to avoid this.
Since cache_mirror does the fetch itself, it also needs to do the
checksum itself if it wants to verify that the source stored in the
mirror is valid. Note that this isn't strictly required because fetching
(including from mirrors) always separately verifies the checksum.
The targets for the cosmetic paths in mirrrors were being calculated
incorrectly as of fb3a3ba: the symlinks used relative paths as targets,
and the relative path was computed relative to the wrong directory.
When creating a cosmetic symlink for a resource in a mirror, remove
it if it already exists. The symlink is removed in case the logic to
create the symlink has changed.
* Some packages (e.g. mpfr at the time of this patch) can have patches
with the same name but different contents (which apply to different
versions of the package). This appends part of the patch hash to the
cache file name to avoid conflicts.
* Some exceptions which occur during fetching are not a subclass of
SpackError and therefore do not have a 'message' attribute. This
updates the logic for mirroring a single spec (add_single_spec)
to produce an appropriate error message in that case (where before
it failed with an AttributeError)
* In various circumstances, a mirror can contain the universal storage
path but not a cosmetic symlink; in this case it would not generate
a symlink. Now "spack mirror create" will create a symlink for any
package that doesn't have one.
`ViewDescriptor.regenerate()` calls `get_all_specs()`, which reads
`spec.yaml` files, which is slow. It's fine to do this once, but
`view.remove_specs()` *also* calls it immediately afterwards.
- [x] Pass the result of `get_all_specs()` as an optional parameter to
`view.remove_specs()` to avoid reading `spec.yaml` files twice.
`ViewDescriptor.regenerate()` was copying specs and stripping build
dependencies, which clears `_hash` and other cached fields on concrete
specs, which causes a bunch of YAML hashes to be recomputed.
- [x] Preserve the `_hash` and `_normal` fields on stripped specs, as
these will be unchanged.
`spack install` previously concretized, writes the entire environment
out, regenerated views, then wrote and regenerated views
again. Regenerating views is slow, so ensure that we only do that once.
- [x] add an option to env.write() to skip view regeneration
- [x] add a note on whether regenerate_views() shouldn't just be a
separate operation -- not clear if we want to keep it as part of write
to ensure consistency, or take it out to avoid performance issues.
Environments need to read the DB a lot when installing all specs.
- [x] Put a read transaction around `install_all()` and `install()`
to avoid repeated locking
Our `LockTransaction` class was reading overly aggressively. In cases
like this:
```
1 with spack.store.db.read_transaction():
2 with spack.store.db.write_transaction():
3 ...
```
The `ReadTransaction` on line 1 would read in the DB, but the
WriteTransaction on line 2 would read in the DB *again*, even though we
had a read lock the whole time. `WriteTransaction`s were only
considering nested writes to decide when to read, but they didn't know
when we already had a read lock.
- [x] `Lock.acquire_write()` return `False` in cases where we already had
a read lock.
If a write transaction was nested inside a read transaction, it would not
write properly on release, e.g., in a sequence like this, inside our
`LockTransaction` class:
```
1 with spack.store.db.read_transaction():
2 with spack.store.db.write_transaction():
3 ...
4 with spack.store.db.read_transaction():
...
```
The WriteTransaction on line 2 had no way of knowing that its
`__exit__()` call was the last *write* in the nesting, and it would skip
calling its write function.
The `__exit__()` call of the `ReadTransaction` on line 1 wouldn't know
how to write, and the file would never be written.
The DB would be correct in memory, but the `ReadTransaction` on line 4
would re-read the whole DB assuming that other processes may have
modified it. Since the DB was never written, we got stale data.
- [x] Make `Lock.release_write()` return `True` whenever we release the
*last write* in a nest.
Lock transactions were actually writing *after* the lock was
released. The code was looking at the result of `release_write()` before
writing, then writing based on whether the lock was released. This is
pretty obviously wrong.
- [x] Refactor `Lock` so that a release function can be passed to the
`Lock` and called *only* when a lock is really released.
- [x] Refactor `LockTransaction` classes to use the release function
instead of checking the return value of `release_read()` / `release_write()`
`ViewDescriptor.regenerate()` checks repeatedly whether packages are
installed and also does a lot of DB queries. Put a read transaction
around the whole thing to avoid repeatedly locking and unlocking the DB.
Users can now list mirrors of the main url in packages.
- [x] Instead of just a single `url` attribute, users can provide a list (`urls`) in the package, and these will be tried by in order by the fetch strategy.
- [x] To handle one of the most common mirror cases, define a `GNUMirrorPackage` mixin to handle all the standard GNU mirrors. GNU packages can set `gnu_mirror_path` to define the path within a mirror, and the mixin handles setting up all the requisite GNU mirror URLs.
- [x] update all GNU packages in `builtin` to use the `GNUMirrorPackage` mixin.
- Add an optional argument so that `possible_dependencies()` will report
missing dependencies.
- Add a test to ensure it works.
- Ignore missing dependencies in `possible_dependencies()` by default.
- this version allows getting possible dependencies of multiple packages
or specs at once.
- New method handles calling `PackageBase.possible_dependencies` multiple
times and passing `visited` dict around.
`Environment.added_specs()` has a loop around calls to
`Package.installed()`, which can result in repeated DB queries. Optimize
this with a read transaction in `Environment`.
Checks for deprecated specs were repeatedly taking out read locks on the
database, which can be very slow.
- [x] put a read transaction around the deprecation check
