This updates the configuration loading/dumping logic (now called
load_config/dump_config) in spack_yaml to preserve comments (by using
ruamel.yaml's RoundTripLoader). This has two effects:
* environment spack.yaml files expect to retain comments, which
load_config now supports. By using load_config, users can now use the
':' override syntax that was previously unavailable for environment
configs (but was available for other config files).
* config files now retain user comments by default (although in cases
where Spack updates/overwrites config, the comments can still be
removed).
Details:
* Subclasses `RoundTripLoader`/`RoundTripDumper` to parse yaml into
ruamel's `CommentedMap` and analogous data structures
* Applies filename info directly to ruamel objects in cases where the
updated loader returns those
* Copies management of sections in `SingleFileScope` from #10651 to allow
overrides to occur
* Updates the loader/dumper to handle the processing of overrides by
specifically checking for the `:` character
* Possibly the most controversial aspect, but without that, the parsed
objects have to be reconstructed (i.e. as was done in
`mark_overrides`). It is possible that `mark_overrides` could remain
and a deep copy will not cause problems, but IMO that's generally
worth avoiding.
* This is also possibly controversial because Spack YAML strings can
include `:`. My reckoning is that this only occurs for version
specifications, so it is safe to check for `endswith(':') and not
('@' in string)`
* As a consequence, this PR ends up reserving spack yaml functions
load_config/dump_config exclusively for the purpose of storing spack
config
`test_envoronment_status()` was printing extra output during tests.
- [x] disable output only for `env('status')` calls instead of disabling
it for the whole test.
This PR ensures that environment activation sets all environment variables set by the equivalent `module load` operations, except that the spec prefixes are "rebased" to the view associated with the environment.
Currently, Spack blindly adds paths relative to the environment view root to the user environment on activation. Issue #12731 points out ways in which this behavior is insufficient.
This PR changes that behavior to use the `setup_run_environment` logic for each package to augment the prefix inspections (as in Spack's modulefile generation logic) to ensure that all necessary variables are set to make use of the packages in the environment.
See #12731 for details on the previous problems in behavior.
This PR also updates the `ViewDescriptor` object in `spack.environment` to have a `__contains__` method. This allows for checks like `if spec in self.default_view`. The `__contains__` operator for `ViewDescriptor` objects checks whether the spec satisfies the filters of the View descriptor, not whether the spec is already linked into the underlying `FilesystemView` object.
This PR ensures that on Darwin we always append /sbin and /usr/sbin to PATH, if they are not already present, when looking for sysctl.
* Make sure we look into /sbin and /usr/sbin for sysctl
* Refactor sysctl for better readability
* Remove marker to make test pass
These changes update our gcc microarchitecture descriptions based on manuals found here https://gcc.gnu.org/onlinedocs/ and assuming that new architectures are not added during patch releases.
This extends Spack functionality so that it can fetch sources and binaries from-, push sources and binaries to-, and index the contents of- mirrors hosted on an S3 bucket.
High level to-do list:
- [x] Extend mirrors configuration to add support for `file://`, and `s3://` URLs.
- [x] Ensure all fetching, pushing, and indexing operations work for `file://` URLs.
- [x] Implement S3 source fetching
- [x] Implement S3 binary mirror indexing
- [x] Implement S3 binary package fetching
- [x] Implement S3 source pushing
- [x] Implement S3 binary package pushing
Important details:
* refactor URL handling to handle S3 URLs and mirror URLs more gracefully.
- updated parse() to accept already-parsed URL objects. an equivalent object
is returned with any extra s3-related attributes intact. Objects created with
urllib can also be passed, and the additional s3 handling logic will still be applied.
* update mirror schema/parsing (mirror can have separate fetch/push URLs)
* implement s3_fetch_strategy/several utility changes
* provide more feature-complete S3 fetching
* update buildcache create command to support S3
* Move the core logic for reading data from S3 out of the s3 fetch strategy and into
the s3 URL handler. The s3 fetch strategy now calls into `read_from_url()` Since
read_from_url can now handle S3 URLs, the S3 fetch strategy is redundant. It's
not clear whether the ideal design is to have S3 fetching functionality in a fetch
strategy, directly implemented in read_from_url, or both.
* expanded what can be passed to `spack buildcache` via the -d flag: In addition
to a directory on the local filesystem, the name of a configured mirror can be
passed, or a push URL can be passed directly.
fixes#13073
Since #3206 was merged bootstrapping environment-modules was using the architecture of the current host or the best match supported by the default compiler. The former case is an issue since shell integration was looking for a spec targeted at the host microarchitecture.
1. Bootstrap an env modules targeted at generic architectures
2. Look for generic targets in shell integration scripts
3. Add a new entry in Travis to test shell integration
Custom string versions for compilers were raising a ValueError on
conversion to int. This commit fixes the behavior by trying to detect
the underlying compiler version when in presence of a custom string
version.
* Refactor code that deals with custom versions for better readability
* Partition version components with a regex
* Fix semantic of custom compiler versions with a suffix
* clang@x.y-apple has been special-cased
* Add unit tests
We've been doing this for quite a while now, and it does not seem to
cause issues.
- [x] Switch the noisy warning to a debug to make Spack a bit quieter
while building.
* Added architecture specific optimization flags for Clang / LLVM
* Disallow compiler optimizations for mixed toolchains
* We emit a warning when building for a mixed toolchain
* Fixed issues with suffixed versions of compilers; Apple's Clang will,
for the time being, fall back on x86-64 for every compilation.
* Methods setting the environment now do it separately for build and run
Before this commit the `*_environment` methods were setting
modifications to both the build-time and run-time environment
simultaneously. This might cause issues as the two environments
inherently rely on different preconditions:
1. The build-time environment is set before building a package, thus
the package prefix doesn't exist and can't be inspected
2. The run-time environment instead is set assuming the target package
has been already installed
Here we split each of these functions into two: one setting the
build-time environment, one the run-time.
We also adopt a fallback strategy that inspects for old methods and
executes them as before, but prints a deprecation warning to tty. This
permits to port packages to use the new methods in a distributed way,
rather than having to modify all the packages at once.
* Added a test that fails if any package uses the old API
Marked the test xfail for now as we have a lot of packages in that
state.
* Added a test to check that a package modified by a PR is up to date
This test can be used any time we deprecate a method call to ensure
that during the first modification of the package we update also
the deprecated calls.
* Updated documentation
Python 3 metaclasses have a `__prepare__` method that lets us save the
class's dictionary before it is constructed. In Python 2 we had to walk
up the stack using our `caller_locals()` method to get at this. Using
`__prepare__` is much faster as it doesn't require us to use `inspect`.
This makes multimethods use the faster `__prepare__` method in Python3,
while still using `caller_locals()` in Python 2. We try to reduce the
use of caller locals using caching to speed up Python 2 a little bit.
Our importer was always parsing from source (which is considerably
slower) because the source size recorded in the .pyc file differed from
the size of the input file.
Override path_stats in the prepending importer to fool it into thinking
that the source size is the size *with* the prepended code.
Since the backup file is only created on the first invocation, it will
contain the original file without any modifications. Further invocations
will then read the backup file, effectively reverting prior invocations.
This can be reproduced easily by trying to install likwid, which will
try to install into /usr/local. Work around this by creating a temporary
file to read from.
* This updates stage names to use "spack-stage-" as a prefix.
This avoids removing non-Spack directories in "spack clean" as
c141e99 did (in this case so long as they don't contain the
prefix "spack-stage-"), and also addresses a follow-up issue
where Spack stage directories were not removed.
* Spack now does more-stringent checking of expected permissions for
staging directories. For a given stage root that includes a user
component, all directories before the user component that are
created by Spack are expected to match the permissions of their
parent; the user component and all deeper directories are expected
to be accessible to the user (read/write/execute).
This feature generates a verification manifest for each installed
package and provides a command, "spack verify", which can be used to
compare the current file checksums/permissions with those calculated
at installed time.
Verification includes
* Checksums of files
* File permissions
* Modification time
* File size
Packages installed before this PR will be skipped during verification.
To verify such a package you must reinstall it.
The spack verify command has three modes.
* With the -a,--all option it will check every installed package.
* With the -f,--files option, it will check some specific files,
determine which package they belong to, and confirm that they have
not been changed.
* With the -s,--specs option or by default, it will check some
specific packages that no files havae changed.
fixes#13005
This commit fixes an issue with the name of the root directory for
module file hierarchies. Since #3206 the root folder was named after
the microarchitecture used for the spec, which is too specific and
not backward compatible for lmod hierarchies. Here we compute the
root folder name using the target family instead of the target name
itself and we add target information in the 'whatis' portion of the
module file.
From Python docs:
--
'surrogateescape' will represent any incorrect bytes as code points in
the Unicode Private Use Area ranging from U+DC80 to U+DCFF. These
private code points will then be turned back into the same bytes when
the surrogateescape error handler is used when writing data. This is
useful for processing files in an unknown encoding.
--
This will allow us to process files with unknown encodings.
To accommodate the case of self-extracting bash scripts, filter_file
can now stop filtering text input if a certain marker is found. The
marker must be passed at call time via the "stop_at" function argument.
At that point the file will be reopened in binary mode and copied
verbatim.
* use "surrogateescape" error handling to ignore unknown chars
* permit to stop filtering if a marker is found
* add unit tests for non-ASCII and mixed text/binary files
- Add a test that verifies checksums on all packages
- Also add an attribute to packages that indicates whether they need a
manual download or not, and add an exception in the tests for these
packages until we can verify them.
Both floating-point and NEON are required in all standard ARMv8
implementations. Theoretically though specialized markets can support
no NEON or floating-point at all. Source:
https://developer.arm.com/docs/den0024/latest/aarch64-floating-point-and-neon
On the other hand the base procedure call standard for Aarch64
"assumes the availability of the vector registers for passing
floating-point and SIMD arguments". Further "the Arm 64-bit
architecture defines two mandatory register banks: a general-purpose
register bank which can be used for scalar integer processing and
pointer arithmetic; and a SIMD and Floating-Point register bank".
Source:
https://developer.arm.com/docs/ihi0055/latest/procedure-call-standard-for-the-arm-64-bit-architecture
This makes customization of Aarch64 with no NEON instruction set
available so unlikely that we can consider them a feature of the
generic family.
This PR adds a 'concretize' entry to an environment's spec.yaml file
which controls how user specs are concretized. By default it is
set to 'separately' which means that each spec added by the user is
concretized separately (the behavior of environments before this PR).
If set to 'together', the environment will concretize all of the
added user specs together; this means that all specs and their
dependencies will be consistent with each other (for example, a
user could develop code linked against the set of libraries in the
environment without conflicts).
If the environment was previously concretized, this will re-concretize
all specs, in which case previously-installed specs may no longer be
used by the environment (in this sense, adding a new spec to an
environment with 'concretize: together' can be significantly more
expensive).
The 'concretize: together' setting is not compatible with Spec
matrices; this PR adds a check to look for multiple instances of the
same package added to the environment and fails early when
'concretize: together' is set (to avoid confusing messages about
conflicts later on).
While the build environment already takes share/pkgconfig into account,
the generated module files etc. only consider lib/pkgconfig and
lib64/pkgconfig.
When removing support for dotkit in #11986 the code trying to set the
paths of the various module files was not updated to skip it. This
results in a failure because of a key error after the deprecation
warning is displayed to user.
This commit fixes the issue and adds a unit test for regression.
Note that code for Spack chains has been updated accordingly but
no unit test has been added for that case.
Dotkit is being used only at a few sites and has been deprecated on new
machines. This commit removes all the code that provide support for the
generation of dotkit module files.
A new validator named "deprecatedProperties" has been added to the
jsonschema validators. It permits to prompt a warning message or exit
with an error if a property that has been marked as deprecated is
encountered.
* Removed references to dotkit in the docs
* Removed references to dotkit in setup-env-test.sh
* Added a unit test for the 'deprecatedProperties' schema validator
fixes#12915closes#12916
Since Spack has support for specific targets it might happen that
software is built for targets that are not exactly the host because
it was either an explicit user request or the compiler being used is
too old to support the host.
Modules for different targets are written into different directories
and by default Spack was adding to MODULEPATH only the directory
corresponding to the current host. This PR modifies this behavior to
add all the directories that are **compatible** with the current host.
Sometimes when remove_file is called on a link, that link is missing
(perhaps ctrl-C happened halfway through a previous action). As
removing a non-existent file is no problem, this patch changes the
behavior so Spack continues rather than stopping with an error.
Currently you would see
ValueError: /path/to/dir is not a link tree!
and now it continues with a warning.
bin/spack now needs to have a "-*- python -*-" line after the shebang, so
that emacs will interpret it as a python file instead of as a shell
script. Add one line to the license check limit to accommodate this.
