- Spack's core package interface was previously overly stateful, in that
calling methods like `do_stage()` could change your working directory.
- This removes Stage's `chdir` and `chdir_to_source` methods and replaces
their usages with `with working_dir(stage.path)` and `with
working_dir(stage.source_path)`, respectively. These ensure the
original working directory is preserved.
- This not only makes the API more usable, it makes the tests more
deterministic, as previously a test could leave the current working
directory in a bad state and cause subsequent tests to fail
mysteriously.
- Assertion would search for root through all possible paths.
- It's also really slow.
- This isn't needed anymore. We're pretty good at ensuring single-rooted
DAGs, and this assertion has never been thrown.
- This shaves another 6 seconds off r-rminer concretization
- This reduces concretization time for r-rminer from over 1 minute to
only 16 seconds.
- OrderedDict ends up taking a *lot* of time to compare larger specs.
- An OrderedDict isn't actually needed here. It's actually not possible
to find duplicates, and we end up sorting the contents of the
OrderedDict anyway.
- This is an optimization to the way traverse_edges iterates over
successors.
- Previous version called dependencies_dict(), which involved a lot of
redundant work (creating dicts and calling caonical_deptype)
- Spack ends up constructing compilers frequently from YAML data.
- This caches the result of parsing the compiler config
- The logic in compilers/__init__.py could use a bigger cleanup, but this
makes concretization much faster for now.
- on macOS, this also ensures that xcrun is called only twice, as opposed
to every time a new compiler object is constructed.
This adds a workflow section on how to use spack on Docker.
It provides an example on the best-practices I collected over the
last months and circumvents the common pitfalls I tapped in.
Works with MPI, CUDA, Modules, execution as root, etc.
Background: Developed initially for PIConGPU.
* Make --trusted default when running spack gpg list
Currently running `spack gpg list` with no arguments returns nothing. You must supply either the `--trusted` or the `--signing` options. The idea here is to return some initial data to the user when the command is run. The alternative is to return an error, telling the user to select one of the two options.
* Add an extra test case for the empty list command
Fixes the issue with code coverage
This isn't a rework of the concretizer but it speeds things up a LOT.
The main culprits were:
1. Variant code, `provider_index`, and `concretize.py` were calling
`spec.package` when they could use `spec.package_class`
- `spec.package` looks up a package instance by `Spec`, which requires a
(fast-ish but not that fast) DAG compare.
- `spec.package_class` just looks up the package's class by name, and you
should use this when all you need is metadata (most of the time).
- not really clear that the current way packages are looked up is
necessary -- we can consider refactoring that in the future.
2. `Repository.repo_for_pkg` parses a `str` argument into a `Spec` when
called with one, via `@_autospec`, but this is not needed.
- Add some faster code to handle strings directly and avoid parsing
This speeds up concretization 3-9x in my limited tests. Still not super
fast but much more bearable:
Before:
- `spack spec xsdk` took 33.6s
- `spack spec dealii` took 1m39s
After:
- `spack spec xsdk` takes 6.8s
- `spack spec dealii` takes 10.8s
* Do not call setup_package for fake installs
- setup package could fail if ``setup_dependent_environment`` or other
routines expected to use executables from dependencies
- xpetsc and boost try to get python config variables in
`setup_dependent_package`; this would cause them not to be
fake-installable
* Remove vestigial deptype_query argument to Spec.traverse()
- The `deptype_query` argument isn't used anymore -- it's only passed
around and causes confusion when calling traverse.
- Get rid of it and just keep the `deptypes` argument
* Don't print redundant messages when installing dependencies
- `do_install()` was originally depth-first recursive, and printed "<pkg>
already installed in ..." multiple times for packages as recursive
calls encountered them.
- For much cleaner output, use spec.traverse(order='post') to install
dependencies instead
* Add package for aspell and ass't dictionaries
Add a package definition for aspell.
Add a handful of dictionaries to convince myself that the support for
a bunch of dictionaries works.
* Flake8 cleanup
* Use six's version of urlparse
`urlparse` is not python3 friendly. This works around it (stolen from
`.../cmd/md5.py`).
* Fix incorrect trimming regexp
* Clean up dictionary build
- more parsimonious use of `which` (`make()` has already been made)
- use `sh` instead of `bash`
* Use a helper method to generate info for variants
I figured out my issues with static methods. I *think* that it this
is pythonic.
* Convert aspell to an extendable package
Convert aspell to be extendable and rework the dictionaries to be
extensions.
As it stands, there's a great deal of cut and paste in the
dictionaries, I'll abstract that out next.
The {de,}activate methods copy a great deal of code out of
package.py. Perhaps there's a better way....
* Create AspellDictPackage and use it for the dictionaries
Reduce the repeated code, pull it into a base class.
I'm confused about why 'from spack import *' wasn't more useful in the
base class.
* Oops, -de & -es should be AspellDictPackages too
* Typo: pakcage -> package
* Address some commentary
* Update copyright dates, 2016->2017
* spec and spec.package.spec can refer to different objects in the
database. When these two instances of spec differ in terms of
the value of the 'concrete' property, Spec._mark_concrete can
fail when checking Spec.package.installed (which requires
package.spec to be concrete). This skips the check for
spec.package.installed when _mark_concrete is called with
'True' (in other words, when the database is marking all specs
as being concrete).
* add test to confirm this fixes#5293
* edits to address issues where spack concretization attempts to set properties on already-installed specs
* most added checks only need to check if the spec is concrete; they dont also need to check if the package is installed
* add test to ensure that patches are not applied to an installed spec
* add test to ensure that an error is detected when a dependent requests a dependency constraint which conflicts with a requested installed dependency
Fixes#5455
All methods within setup_package use an EnvironmentModifications object
to control the environment. Those modifications are applied at the end
of setup_package. Module loads for the build environment need to be
done after the rest of the environment modifications are applied, as
otherwise Spack may unset variables set by those modules (for example
LD_LIBRARY_PATH).
closes#2884closes#4684
In #1848 we decided to use `Spec.format` to expand certain tokens in
the module file naming scheme or in the environment variable name.
Not all the tokens that are allowed in `Spec.format` make sense in
module file generation. This PR restricts the set of tokens that can
be used, and adds tests to check that the intended behavior is respected.
Additionally, the names of environment variables set/modified by module
files were, up to now, always uppercase. There are packages though that
require case sensitive variable names to honor certain behaviors (e.g.
OpenMPI). This PR restricts the uppercase transformation in variable
names to `Spec.format` tokens.
fixes#5587
In trying to preserve patch ordering, #5476 made equality inconsistent
for the added 'patches' variant. This commit maintains the original
weak ordering of patch applications while preserving consistency of
comparisons. The ordering DOES NOT enter the hashing mechanism. It's
supposed to be a hotfix, while we think of a cleaner and more-permanent
solution.
- This steals the magic regular expressions that CTest uses to parse log
files and addds them to Spack. See here:
https://github.com/Kitware/CMake/blob/master/Source/CTest/cmCTestBuildHandler.cxx
These are BSD licensed, so the port is in `externa/ctest_log_parser.py`
- We currently use these to do better filtering of errors from build
output. Plan is to use them to generate good CDash output.
`spack blame` prints out the contributors to a package.
By modification time:
```
$ spack blame --time llvm
LAST_COMMIT LINES % AUTHOR EMAIL
3 days ago 2 0.6 Andrey Prokopenko <andrey.prok@gmail.com>
3 weeks ago 125 34.7 Massimiliano Culpo <massimiliano.culpo@epfl.ch>
3 weeks ago 3 0.8 Peter Scheibel <scheibel1@llnl.gov>
2 months ago 21 5.8 Adam J. Stewart <ajstewart426@gmail.com>
2 months ago 1 0.3 Gregory Becker <becker33@llnl.gov>
3 months ago 116 32.2 Todd Gamblin <tgamblin@llnl.gov>
5 months ago 2 0.6 Jimmy Tang <jcftang@gmail.com>
5 months ago 6 1.7 Jean-Paul Pelteret <jppelteret@gmail.com>
7 months ago 65 18.1 Tom Scogland <tscogland@llnl.gov>
11 months ago 13 3.6 Kelly (KT) Thompson <kgt@lanl.gov>
a year ago 1 0.3 Scott Pakin <pakin@lanl.gov>
a year ago 3 0.8 Erik Schnetter <schnetter@gmail.com>
3 years ago 2 0.6 David Beckingsale <davidbeckingsale@gmail.com>
3 days ago 360 100.0
```
Or by percent contribution:
```
$ spack blame --percent llvm
LAST_COMMIT LINES % AUTHOR EMAIL
3 weeks ago 125 34.7 Massimiliano Culpo <massimiliano.culpo@epfl.ch>
3 months ago 116 32.2 Todd Gamblin <tgamblin@llnl.gov>
7 months ago 65 18.1 Tom Scogland <tscogland@llnl.gov>
2 months ago 21 5.8 Adam J. Stewart <ajstewart426@gmail.com>
11 months ago 13 3.6 Kelly (KT) Thompson <kgt@lanl.gov>
5 months ago 6 1.7 Jean-Paul Pelteret <jppelteret@gmail.com>
3 weeks ago 3 0.8 Peter Scheibel <scheibel1@llnl.gov>
a year ago 3 0.8 Erik Schnetter <schnetter@gmail.com>
3 years ago 2 0.6 David Beckingsale <davidbeckingsale@gmail.com>
3 days ago 2 0.6 Andrey Prokopenko <andrey.prok@gmail.com>
5 months ago 2 0.6 Jimmy Tang <jcftang@gmail.com>
2 months ago 1 0.3 Gregory Becker <becker33@llnl.gov>
a year ago 1 0.3 Scott Pakin <pakin@lanl.gov>
3 days ago 360 100.0
```
- A package can depend on a special patched version of its dependencies.
- The `Spec` YAML (and therefore the hash) now includes the sha256 of
the patch in the `Spec` YAML, which changes its hash.
- The special patched version will be built separately from a "vanilla"
version of the same package.
- This allows packages to maintain patches on their dependencies
without affecting either the dependency package or its dependents.
This could previously be accomplished with special variants, but
having to add variants means the hash of the dependency changes
frequently when it really doesn't need to. This commit allows the
hash to change *just* for dependencies that need patches.
- Patching dependencies shouldn't be the common case, but some packages
(qmcpack, hpctoolkit, openspeedshop) do this kind of thing and it
makes the code structure mirror maintenance responsibilities.
- Note that this commit means that adding or changing a patch on a
package will change its hash. This is probably what *should* happen,
but we haven't done it so far.
- Only applies to `patch()` directives; `package.py` files (and their
`patch()` functions) are not hashed, but we'd like to do that in the
future.
- The interface looks like this: `depends_on()` can optionally take a
patch directive or a list of them:
depends_on(<spec>,
patches=patch(..., when=<cond>),
when=<cond>)
# or
depends_on(<spec>,
patches=[patch(..., when=<cond>),
patch(..., when=<cond>)],
when=<cond>)
- Previously, the `patch()` directive only took an `md5` parameter. Now
it only takes a `sha256` parameter. We restrict this because we want
to be consistent about which hash is used in the `Spec`.
- A side effect of hashing patches is that *compressed* patches fetched
from URLs now need *two* checksums: one for the downloaded archive and
one for the content of the patch itself. Patches fetched uncompressed
only need a checksum for the patch. Rationale:
- we include the content of the *patch* in the spec hash, as that is
the checksum we can do consistently for patches included in Spack's
source and patches fetched remotely, both compressed and
uncompressed.
- we *still* need the patch of the downloaded archive, because we want
to verify the download *before* handing it off to tar, unzip, or
another decompressor. Not doing so is a security risk and leaves
users exposed to any arbitrary code execution vulnerabilities in
compression tools.
- Functions returned by directives were all called `_execute`, which made
reading stack traces hard because you couldn't tell what directive a
frame came from.
- renamed them all to `_execute_<directive>`
- Exceptions in directives were only really used in one or two places --
get rid of the boilerplate init functions and let the callsite specify
the message.
- move `spack.cmd.checksum.get_checksums` to `spack.util.web.spider_checksums`
- move `spack.error.NoNetworkError` to `spack.util.web.NoNetworkError` since
it is only used there.
- Previously, dependencies and dependency_types were stored as separate
dicts on Package.
- This means a package can only depend on another in one specific way,
which is usually but not always true.
- Prior code unioned dependency types statically across dependencies on
the same package.
- New code stores dependency relationships as their own object, with a
spec constraint and a set of dependency types per relationship.
- Dependency types are now more precise
- There is now room to add more information to dependency relationships.
- New Dependency class lives in dependency.py, along with deptype
definitions that used to live in spack.spec.
Move deptype definitions to spack.dependency
* Add '--test=all' and '--test=root' options to test either the root or the root and all dependencies.
* add a test dependency type that is only used when --test is enabled.
* test dependencies are not added to the spec, but they are provided in the test environment.
closes#5473
Prior to this PR we were not exiting early for external packages, which
caused the `configure_options` property of the contexts to fail with
e.g. a key error because the DAG gets truncated for them. More
importantly Spack configure options don't make any sense for externals.
Now we exit early, and leave a message in the module file clarifying
that this package has been installed outside of Spack.
closes#5201
Currently, if a user sets an external package to have a prefix that is
one of the system paths (like '/usr') the module files that are
generated will prepend '/usr/bin' to 'PATH', etc. This is particularly
nasty at the time when a module file is unloaded, and e.g. paths like
'/usr/bin' will be discarded from PATH.
This PR solves the issue skipping system paths when a prefix inspection
is made to generate module files.
* Module files now are generated using a template engine refers #2902#3173
jinja2 has been hooked into Spack.
The python module `modules.py` has been splitted into several modules
under the python package `spack/modules`. Unit tests stressing module
file generation have been refactored accordingly.
The module file generator for Lmod has been extended to multi-providers
and deeper hierarchies.
* Improved the support for templates in module files.
Added an entry in `config.yaml` (`template_dirs`) to list all the
directories where Spack could find templates for `jinja2`.
Module file generators have a simple override mechanism to override
template selection ('modules.yaml' beats 'package.py' beats 'default').
* Added jinja2 and MarkupSafe to vendored packages.
* Spec.concretize() sets mutual spec-package references
The correct place to set the mutual references between spec and package
objects at the end of concretization. After a call to concretize we
should now be ensured that spec is the same object as spec.package.spec.
Code in `build_environment.py` that was performing the same operation
has been turned into an assertion to be defensive on the new behavior.
* Improved code and data layout for modules and related tests.
Common fixtures related to module file generation have been extracted
in `conftest.py`. All the mock configurations for module files have been
extracted from python code and have been put into their own yaml file.
Added a `context_property` decorator for the template engine, to make
it easy to define dictionaries out of properties.
The default for `verbose` in `modules.yaml` is now False instead of True.
* Extendable module file contexts + short description from docstring
The contexts that are used in conjunction with `jinja2` templates to
generate module files can now be extended from package.py and
modules.yaml.
Module files generators now infer the short description from package.py
docstring (and as you may expect it's the first paragraph)
* 'module refresh' regenerates all modules by default
`module refresh` without `--module-type` specified tries to
regenerate all known module types. The same holds true for `module rm`
Configure options used at build time are extracted and written into the
module files where possible.
* Fixed python3 compatibility, tests for Lmod and Tcl.
Added test for exceptional paths of execution when generating Lmod
module files.
Fixed a few compatibility issues with python3.
Fixed a bug in Tcl with naming_scheme and autoload + unit tests
* Updated module file tutorial docs. Fixed a few typos in docstrings.
The reference section for module files has been reorganized. The idea is
to have only three topics at the highest level:
- shell support + spack load/unload use/unuse
- module file generation (a.k.a. APIs + modules.yaml)
- module file maintenance (spack module refresh/rm)
Module file generation will cover the entries in modules.yaml
Also:
- Licenses have been updated to include NOTICE and extended to 2017
- docstrings have been reformatted according to Google style
* Removed redundant arguments to RPackage and WafPackage.
All the callbacks in `RPackage` and `WafPackage` that are not build
phases have been modified not to accept a `spec` and a `prefix`
argument. This permits to leverage the common `configure_args` signature
to insert by default the configuration arguments into the generated
module files. I think it's preferable to handling those packages
differently than `AutotoolsPackage`. Besides only one package seems
to override one of these methods.
* Fixed broken indentation + improved resiliency of refresh
Fixed broken indentation in `spack module refresh` (probably a rebase
gone silently wrong?). Filter the writers for blacklisted specs before
searching for name clashes. An error with a single writer will not
stop regeneration, but instead will print a warning and continue
the command.
- converted `log_path` and `env_path` to properties of PackageBase.
- InstallErrors in build_environment are now annotated with the package
that caused them, in the 'pkg' attribute.
- Add `--show-log-on-error` option to `spack install` that catches
InstallErrors and prints the log to stderr if it exists.
Note that adding a reference to the Pakcage allows a lot of stuff
currently handled by do_install() and build_environment to be handled
externally.
- '\b' in regular expression needs to be in a raw string (r'\b')
- Regression test that would've caught this was unintentionally disabled
- This fixes the string and the test
