The current use of git ref's as a version requires a search algorithm to pick the right matching version based on the tags in the git history of the package.
This is less than ideal for the use case where users already know the specific version they want the git ref to be associated with. This PR makes a new version syntax [package]@[ref]=[version] to allow the users to specify the exact hash they wish to use.
On some systems the shell in login mode wipes important parts of the
environment, such as PATH. This causes the build to fail since it can't
find `spack`.
For better robustness, don't use a login shell.
In a full CI job the final spack install is run in an environment formed by scripts running in this order:
export AWS_SECRET=... # 1. Load environment from GitLab project variables
source spack/share/spack/setup-env.sh # 2. Load Spack into the environment (PATH)
spack env activate -V concrete_env # 3. Activate the concrete environment
source /etc/profile # 4. Bash login shell (from -l)
spack install ...
Whereas when a user launches their own container with (docker|podman) run -it, they end up running spack install in an environment formed in this order:
source /etc/bash.bashrc # (not 4). Bash interactive shell (default with TTY)
export AWS_SECRET=... #~1. Manually load environment from GitLab project variables
source spack/share/spack/setup-env.sh # 2. Load Spack into the environment (PATH)
spack env activate -V concrete_env # 3. Activate the concrete environment
spack install ...
The big problem being that (4) has a completely different position and content (on Leap 15 and possibly other containers).
So in context, this PR removes (4) from the CI job case, leaving us with the simpler:
export AWS_SECRET=... # 1. Load environment from GitLab project variables
source spack/share/spack/setup-env.sh # 2. Load Spack into the environment (PATH)
spack env activate -V concrete_env # 3. Activate the concrete environment
spack install ...
* database: don't sort on return from query_local
* ASP-based solver: don't build the hash-lookup dictionary twice
Building this dictionary twice and traversing all the specs
might be time-consuming for large buildcaches.
This test relied on an old version of the `flake8_package` fixture that modified
the spack repository, but it doesn't do that anymore. There are other tests for
`changed_files()` that do a better job of mocking up a git repository with
changes, so we can just delete this one.
`spack style` tests were annoyingly brittle because we could not easily be
specific about which tools to run (we had to use `--no-black`, `--no-isort`,
`--no-flake8`, and `--no-mypy`). We should be able to specify what to run OR
what to skip.
Now you can run, e.g.:
spack style --tool black,flake8
or:
spack style --skip black,isort
- [x] Remove `--no-black`, `--no-isort`, `--no-flake8`, and `--no-mypy` args.
- [x] Add `--tool TOOL` argument.
- [x] Add `--skip TOOL` argument.
- [x] Allow either `--tool black --tool flake8` or `--tool black,flake8` syntax.
- [x] remove alignment spaces from tempaltes
- [x] replace single with double quotes
- [x] Makefile template now generates parsable code
(function body is `pass` instead of just a comment)
- [x] template checks now run black to check output
Previously we'd accept any version for bootstrapping black, but we need <= 21.
- [x] modify bootstrapping code to check black version before accepting an
executable from `PATH`.
- [x] add `.git-blame-ignore-revs` to ignore black reformatting
- [x] make `spack blame` respect `.git-blame-ignore-revs`
(even if the user hasn't configured git to do so)
Some of our tests rely on single vs. double quotes, and others rely on specific
line numbers in the source. These needed fixing after the switch to Black.
Black will automatically fix a lot of the exceptions we previously allowed for
directives, so we don't need them in our custom `flake8_formatter` anymore.
- [x] remove `E501` (long line) exceptions for directives from `flake8_formatter`,
as they won't help us now.
- [x] Refine exceptions for long URLs in the `flake8_formatter`.
- [x] Adjust the mock `flake8-package` to exhibit the exceptions we still allow.
- [x] Update style tests for new `flake8-package`.
- [x] Blacken style test.
Many noqa's in the code are no longer necessary now that the column limit is 99
characters. Others can easily be eliminated, and still more can just be made more
specific if they do not have to do with line length.
The only E501's still in the code are in the tests for `spack.util.path` and the tests
for `spack style`.
This adds necessary configuration for flake8 and black to work together.
This also sets the line length to 99, per the data here:
* https://github.com/spack/spack/pull/24718#issuecomment-876933636
Given the data and the spirit of black's 88-character limit, we set the limit to 99
characters for all of Spack, because:
* 99 is one less than 100, a nice round number, and all lines will fit in a
100-character wide terminal (even when the text editor puts a \ at EOL).
* 99 is just past the knee the file size curve for packages, and it means that packages
remain readable and not significantly longer than they are now.
* It doesn't seem to hurt core -- files in core might change length by a few percent but
seem like they'll be mostly the same as before -- just a bit more roomy.
- [x] set line length to 99
- [x] remove most exceptions from `.flake8` and add the ones black cares about
- [x] add `[tool.black]` to `pyproject.toml`
- [x] make `black` run if available in `spack style --fix`
Co-Authored-By: Tom Scogland <tscogland@llnl.gov>
In #31618 the idea was to determine the file extension heuristically by dropping query params etc from a url and then consider it as a file path. That broke for URLs that only have query params like http://example.com/?patch=x as it would result in empty string as basename. This PR reverts to the old behavior of saving files as ?patch=x in that case.
Co-authored-by: Stephen Sachs <stesachs@amazon.com>
* llvm: Use variant when clauses for many of the expressed conflicts
* llvm: Remove the shared variant as it wasn't really used
* llvm: Remove unnecessary deps and make explicit the ones that are
* llvm: Cleanup patch conditions
* pocl: Update for llvm cleanup
* unit-test: update unparse package hash with the updated llvm package
* llvm: Fix ppc long double patching and add clarifying comments
`self.archive_file` is (among others) a symlink to a tarball. `extension()` on a
symlink will result in no extension. This patch fixes the behavior introduced in
https://github.com/spack/spack/pull/31618.
Co-authored-by: Stephen Sachs <stesachs@amazon.com>
When
1. Spack installs libtool,
2. system libtool is installed too, and
3. system automake is used
Spack passes system automake's `-I <prefix>` flag to itself, even though
it's a default search path. This takes precedence over spack's libtool
prefix dir. This causes the wrong `libtool.m4` file to be used (since
system libtool is in the same prefix as system automake).
And that leads to error messages about incompatible libtool, something
something LT_INIT.
fixes#31627
spack.mirror.get_all_versions now uses the package class
instead of the package object in its implementation.
Ensure spec is concrete before staging for mirrors
Newer versions of botocore (>=1.23.47) support the full IOBase
interface, so the hacks added to supplement the missing attributes are
no longer needed. Conditionally disable the hacks if they appear to be
unnecessary based on the class hierarchy found at runtime.
* Add connection information to buildcache update command
Ensure that the s3 connection made when attempting to update the content
of a buildcache attempts to use the extra connection information
from the mirror creation.
* Add unique help for endpoint URL argument
Fix copy/paste error for endpoint URL help which was the same as
the access token
* Re-work URL checking for S3 mirrors
Due to the fact that nested bucket URLs would never match the string used
for checking that the mirror is the same, switch the check used.
Sort all mirror URLs by length to have the most specific cases first
and see if the desired URL "starts with" the mirror URL.
* Long line style fixes
Add execptions for long lines and fix other style errors
* Use format() function to rebuild URL
Use the format command to rebuild the url instead of crafing a
formatted string out of known values
* Add early exit for URL checking
When a valid mirror is found, break from the loop
For a long time the module configuration has had a few settings that use
`blacklist`/`whitelist` terminology. We've been asked by some of our users to replace
this with more inclusive language. In addition to being non-inclusive, `blacklist` and
`whitelist` are inconsistent with the rest of Spack, which uses `include` and `exclude`
for the same concepts.
- [x] Deprecate `blacklist`, `whitelist`, `blacklist_implicits` and `environment_blacklist`
in favor of `exclude`, `include`, `exclude_implicits` and `exclude_env_vars` in module
configuration, to be removed in Spack v0.20.
- [x] Print deprecation warnings if any of the deprecated names are in module config.
- [x] Update tests to test old and new names.
- [x] Update docs.
- [x] Update `spack config update` to fix this automatically, and include a note in the error
that you can use this command.
Python's built-in tarfile support doesn't address some general
cases of malformed tarfiles that are already handled by the system
'tar' utility; until these can be addressed, use that exclusively.
The goal of this PR is to make clearer where we need a package object in Spack as opposed to a package class.
We currently instantiate a lot of package objects when we could make do with a class. We should use the class
when we only need metadata, and we should only instantiate and us an instance of `PackageBase` at build time.
Modifications:
- [x] Remove the `spack.repo.get` convenience function (which was used in many places, and not really needed)
- [x] Use `spack.repo.path.get_pkg_class` wherever possible
- [x] Try to route most of the need for `spack.repo.path.get` through `Spec.package`
- [x] Introduce a non-data descriptor, that can be used as a decorator, to have "class level properties"
- [x] Refactor unit tests that had to be modified to reduce code duplication
- [x] `Spec.package` and `Repo.get` now require a concrete spec as input
- [x] Remove `RepoPath.all_packages` and `Repo.all_packages`
There's a race condition in `remove()` as the lockfile is removed after
releasing the lock, which is a problem when another process acquires a
write lock during deletion.
Also simplify life a bit in multiprocessing when a file is possibly
removed multiple times, which currently is an error on the second
deletion, so the proposed fix is to make remove(...) idempotent and not
error when deleting non-existing cache entries.
Don't tests for existence of lockfile, cause windows/linux behavior is different
Oversight in #31433, the non-phony `env` target was missing a file being
created for it, which can cause make to infinitely loop when including
multiple generated makefiles.
When no default editor is installed and no environment variable is set,
which_string would return None and this would be passed to os.execv
resulting in a TypeError. The message presented to the user would be:
Error: execv: path should be string, bytes or os.PathLike,
not NoneType
This change checks that which_string has returned successfully before
attempting to execute the result, resulting in a new error message:
Error: No text editor found! Please set the VISUAL and/or EDITOR
environment variable(s) to your preferred text editor.
It's not strictly necessary, but I've also changed try_exec to catch
all errors rather than just OSErrors. This would have provided slightly
more context for the original error message.
There were two choices: 1) remove '-p' from '-a' or 2) allow monkeypatching
the cleaning of the python cache since clean's test_function_calls isn't
supposed to be actually cleaning anything.
This commit supports the latter and adds a test case for `-p`.
Release branches and tags run protected pipelines, and we noticed
that those pipelines were generating all jobs in the stack, even
when the mirror contained all the built specs and an up to date
index. The problem was caused because the override mirror was
not present in spacks mirror configuration at the time when the
binary_distribution.update() method was called. This fixes that
by always adding the mirror override, if present, to the list of
configured mirrors.
* remove unhelpful comment
* Filter compiler duplicates while reading manifest
* more-specific version matching edited to use module-specific version (to avoid an issue where a user might add a compiler with the same version to the initial test configuration
* PythonPackage: add default libs/headers attributes
* Style fix
* libs and headers should be properties
* Check both platlib and include
* Fix variable reference
Building on #24639, this allows versions to be prefixed by `git.`. If a version begins `git.`, it is treated as a git ref, and handled as git commits are starting in the referenced PR.
An exception is made for versions that are `git.develop`, `git.main`, `git.master`, `git.head`, or `git.trunk`. Those are assumed to be greater than all other versions, as those prefixed strings are in other contexts.
This commit adds some changes which improve use of Spack-installed
oneAPI packages with Spack-generated modules, but not within Spack
(e.g. if you install some of these packages with Spack, then load
their associated modules to build other packages outside of Spack).
The majority of the PR diff is refactoring. The functional changes
are:
* intel-oneapi-mkl:
* setup_run_environment: update Intel compiler flags to RPATH the
mkl libs
* intel-oneapi-compilers: update the compiler configuration to RPATH
libraries installed by this package (note that Spack already handled
this when installing dependent packages with Spack, but this is
specifically to use the intel-oneapi-compilers package outside
of Spack). Specifically:
* inject_rpaths: this modifies the binaries installed as part of
the intel-oneapi-compilers package to RPATH libraries provided
by this package
* extend_config_flags: this instructs the compiler executables
provided by this package to RPATH those same libraries
Refactoring includes:
* intel-oneapi-compilers: in addition to the functional changes,
a large portion of this diff is reorganizing the creation of
resources/archives downloaded for the install
* The base oneAPI package renamed component_path, so several packages
changed as a result:
* intel-oneapi-dpl
* intel-oneapi-dnn
* extrae
* intel-oneapi-mpi:
* Perform file filtering in one pass rather than two
Allow `spack external find` (with no extra args) to proceed if the manifest file exists but
without sufficient permissions; in that case, print a warning. Also add a test for that behavior.
TODOs:
- [x] continue past any exception raised during manifest parsing as part of `spack external find`,
except for CTRL-C (and other errors that indicate immediate program termination)
- [x] Semi-unrelated but came up when discussing this with the user who reported this issue to
me: the manifest parser now accepts older schemas
See: https://github.com/spack/spack/issues/31191
fixes#30997
Instead of giving a penalty of 30 to all nodes when preferences
are not package specific, give a penalty of 100 to all targets
of a node where we have package specific preferences, if the target
is not explicitly preferred.
* Fixed a bug in the 'external find --all' command where the call failed
to find packages by both executable and library. The bug was that the
call `path.all_packages()` incorrectly turned the variable
`packages_to_check` into a generator rather than keeping it a list.
Thus the second call to `detection.by_library` had no work to do.
* Fixed the help message for the find external and compiler commands as
well as others that used the `scopes_metavar` field to define where
the results should be stored in configuration space. Specifically,
the fact that configuration could be added to the environment was not
mentioned in the help message.
* add test to verify fix works
* fix spec cflags/variants parsing test (breaking change)
* fix `spack spec` arg quoting issue
* add error report for deprecated cflags coalescing
* use .group(n) vs subscript regex group extraction for 3.5 compat
* add random test for untested functionality to pass codecov
* fix new test failure since rebase
Fix a bug introduced in #21720. `spack_json.dump()` calls `_strify()` on dictionaries to
convert `unicode` to `str`, but it constructs `dict` objects instead of
`collections.OrderedDict` objects, so in Python 2 (or earlier versions of 3) it can
scramble dictionary order.
This can cause hashes to differ between Python 2 and Python 3, or between Python 3.7
and earlier Python 3's.
- [x] use `OrderedDict` in `_strify`
- [x] add a regression test
The "submodules" argument of the "version" directive can now accept
a callable that returns a list of submodules, in addition to the usual
Boolean values
* bootstrap: account for disabled sources
Fix a bug introduced in #30192, which effectively skips
any prescription on disabled bootstrapping sources.
* Add unit test to avoid regression
Fixes#31021
With #25185, we no longer default to using tar when we can't
determine the extension type, opting to fail instead.
This broke fetching for the pcre package, where we couldn't determine
the extension. To determine the extension, we were attempting to
extract it from the destination filename; however, this file name
may omit details of the origin URL that are required to determine the
extension, so instead we examine the URL directly.
This also updates the decompressor_for method not to set ext=None
by default: it must now always be set by the caller.
