Sometimes one needs to preserve the (relative order) of
mtimes on installed files. So it's better to just copy
over all the metadata from the source tree to the install
tree. If permissions need fixing, that will be done anyway
afterwards.
One major use case are resource()s:
They're unpacked in one place and then copied to their
final place using install_tree(). If the resource is a
source tree using autoconf/automake, resetting mtimes
uncorrectly might force unwanted autoconf/etc calls.
Fixes#9394Closes#13217.
## Background
Spack provides the ability to enable/disable parallel builds through two options: package `parallel` and configuration `build_jobs`. This PR changes the algorithm to allow multiple, simultaneous processes to coordinate the installation of the same spec (and specs with overlapping dependencies.).
The `parallel` (boolean) property sets the default for its package though the value can be overridden in the `install` method.
Spack's current parallel builds are limited to build tools supporting `jobs` arguments (e.g., `Makefiles`). The number of jobs actually used is calculated as`min(config:build_jobs, # cores, 16)`, which can be overridden in the package or on the command line (i.e., `spack install -j <# jobs>`).
This PR adds support for distributed (single- and multi-node) parallel builds. The goals of this work include improving the efficiency of installing packages with many dependencies and reducing the repetition associated with concurrent installations of (dependency) packages.
## Approach
### File System Locks
Coordination between concurrent installs of overlapping packages to a Spack instance is accomplished through bottom-up dependency DAG processing and file system locks. The runs can be a combination of interactive and batch processes affecting the same file system. Exclusive prefix locks are required to install a package while shared prefix locks are required to check if the package is installed.
Failures are communicated through a separate exclusive prefix failure lock, for concurrent processes, combined with a persistent store, for separate, related build processes. The resulting file contains the failing spec to facilitate manual debugging.
### Priority Queue
Management of dependency builds changed from reliance on recursion to use of a priority queue where the priority of a spec is based on the number of its remaining uninstalled dependencies.
Using a queue required a change to dependency build exception handling with the most visible issue being that the `install` method *must* install something in the prefix. Consequently, packages can no longer get away with an install method consisting of `pass`, for example.
## Caveats
- This still only parallelizes a single-rooted build. Multi-rooted installs (e.g., for environments) are TBD in a future PR.
Tasks:
- [x] Adjust package lock timeout to correspond to value used in the demo
- [x] Adjust database lock timeout to reduce contention on startup of concurrent
`spack install <spec>` calls
- [x] Replace (test) package's `install: pass` methods with file creation since post-install
`sanity_check_prefix` will otherwise error out with `Install failed .. Nothing was installed!`
- [x] Resolve remaining existing test failures
- [x] Respond to alalazo's initial feedback
- [x] Remove `bin/demo-locks.py`
- [x] Add new tests to address new coverage issues
- [x] Replace built-in package's `def install(..): pass` to "install" something
(i.e., only `apple-libunwind`)
- [x] Increase code coverage
VSX alitvec extensions are supported by PowerISA from v2.06 (Power7+), but might
not be listed in features.
FMA has been supported by PowerISA since Power1, but might not be listed in
features.
This commit adds these features to all the power ISA family sets.
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>
`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()`
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.
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()`
Vendors for ARM come out of `/proc/cpuinfo` as hex numbers instead of readable strings.
- Add support for associating vendor names with the hex numbers.
- Also move these mappings from Python code to `microarchitectures.json`
- Move darwin feature name mappings to `microarchitectures.json` as well
* CUDA HeaderList: Unit Test
* Spec Header Dirs: Only first include/
Avoid matching recurringly nested include paths that usually
refer to internally shipped libraries in packages.
Example in CUDA Toolkit, shipping a libc++ fork internally
with libcu++ since 10.2.89:
`<prefix>/include/cuda/some/more/details/include/` or
`<prefix>/include/cuda/std/detail/libcxx/include`
regex: non-greedy first match of include
Co-Authored-By: Massimiliano Culpo <massimiliano.culpo@gmail.com>
* CUDA: Re-Enable 10.2.89 as Default
* Fixed x86-64 optimization flags for clang
* Fixed expected results in unit tests
Before the flags used where the one for llc, the underlying compiler from LLVM IR to machine assembly. It turns out that the semantic of `-march`, `-mtune` and `-mcpu` changes from clang front-end to llc.
I found no definitive reference for the flags submitted in this PR, but I checked the assembly on a vectorizable function using Godbolt's web-site.
* Add process to determine aarch64 microarchitecture
* add microarchitectures for thunderx2 and a64fx
* Add optimize flags for gcc on aarch64 family processors thunderx2 and a64fx.
* Add optimize flags for clang on aarch64 family processors thunderx2 and a64fx
* Add testing for thunderx2 and a64fx microarchitectures
New entry for K10 microarchitecture.
Reorder Zen* microarchitectures to avoid triggering as k10.
Remove some desktop-specific flags that were preventing Opteron Bulldozer/Piledriver/Steamroller/Excavator CPUs from being recognized as such.
Remove one or two flags which weren't produced in /proc/cpuinfo on older OS (RHEL6 and friends).
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.
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
* 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.
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.
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
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.
The output of subprocess.check_output is a byte string in Python 3. This causes dictionary lookup to fail later on.
A try-except around this function prevented this error from being noticed. Removed this so that more errors can propagate out.
* microarchitectures: zen starts from x86_64, not from excavator
* Unit tests: fixed a test that is wrong with the new modeling
* microarchitectures: fixed features and inheritance for 15h family
bulldozer doesn't inherit from barcelona (10h) + added xop, lwp and tbm
instruction sets to the 15h family (it distinguish the family from 17h)
- This is needed to support Cray machines -- we need an architecture
mic_knl > x86_64
- We used Cray's naming scheme for this target to make it work seamlessly
with the module-based detection sccheme on Cray. mic_knl is pretty
much dead, so this will be the last succh target. We will need to work
wtih Cray and other vendors in the future.
Spack can now:
- label ppc64, ppc64le, x86_64, etc. builds with specific
microarchitecture-specific names, like 'haswell', 'skylake' or
'icelake'.
- detect the host architecture of a machine from /proc/cpuinfo or similar
tools.
- Understand which microarchitectures are compatible with which (for
binary reuse)
- Understand which compiler flags are needed (for GCC, so far) to build
binaries for particular microarchitectures.
All of this is managed through a JSON file (microarchitectures.json) that
contains detailed auto-detection, compiler flag, and compatibility
information for specific microarchitecture targets. The `llnl.util.cpu`
module implements a library that allows detection and comparison of
microarchitectures based on the data in this file.
The `target` part of Spack specs is now essentially a Microarchitecture
object, and Specs' targets can be compared for compatibility as well.
This allows us to label optimized binary packages at a granularity that
enables them to be reused on compatible machines. Previously, we only
knew that a package was built for x86_64, NOT which x86_64 machines it
was usable on.
Currently this feature supports Intel, Power, and AMD chips. Support for
ARM is forthcoming.
Specifics:
- Add microarchitectures.json with descriptions of architectures
- Relaxed semantic of compiler's "target" attribute. Before this change
the semantic to check if a compiler could be viable for a given target
was exact match. This made sense as the finest granularity of targets
was architecture families. As now we can target micro-architectures,
this commit changes the semantic by interpreting as the architecture
family what is stored in the compiler's "target" attribute. A compiler
is then a viable choice if the target being concretized belongs to the
same family. Similarly when a new compiler is detected the architecture
family is stored in the "target" attribute.
- Make Spack's `cc` compiler wrapper inject target-specific flags on the
command line
- Architecture concretization updated to use the same algorithm as
compiler concretization
- Micro-architecture features, vendor, generation etc. are included in
the package hash. Generic architectures, such as x86_64 or ppc64, are
still dumped using the name only.
- If the compiler for a target is not supported exit with an intelligible
error message. If the compiler support is unknown don't try to use
optimization flags.
- Support and define feature aliases (e.g., sse3 -> ssse3) in
microarchitectures.json and on Microarchitecture objects. Feature
aliases are defined in targets.json and map a name (the "alias") to a
list of rules that must be met for the test to be successful. The rules
that are available can be extended later using a decorator.
- Implement subset semantics for comparing microarchitectures (treat
microarchitectures as a partial order, i.e. (a < b), (a == b) and (b <
a) can all be false.
- Implement logic to automatically demote the default target if the
compiler being used is too old to optimize for it. Updated docs to make
this behavior explicit. This avoids surprising the user if the default
compiler is older than the host architecture.
This commit adds unit tests to verify the semantics of target ranges and
target lists in constraints. The implementation to allow target ranges
and lists is minimal and doesn't add any new type. A more careful
refactor that takes into account the type system might be due later.
Co-authored-by: Gregory Becker <becker33.llnl.gov>
Add llnl.util.cpu_name, with initial support for detecting different
microarchitectures on Linux. This also adds preliminary changes for
compiler support and variants to control the optimizatoin levels by
target.
This does not yet include translations of targets to particular
compilers; that is left to another PR.
Co-authored-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
* implicit_rpaths are now removed from compilers.yaml config and are always instantiated dynamically, this occurs one time in the build_environment module
* per-compiler list required libraries (e.g. libstdc++, libgfortran) and whitelist directories from rpaths including those libraries. Remove non-whitelisted implicit rpaths. Some libraries default for all compilers.
* reintroduce 'implicit_rpaths' as a config variable that can be used to disable Spack insertion of compiler RPATHs generated at build time.
- mkdirp now takes arguments to allow it to properly set permissions on created directories.
- Two arguments (group and mode) set permissions for the leaf directory.
- Intermediate directories can inherit permissions from either the topmost existing directory (the parent) or the leaf.
Fixes#11163
The goal of this work is to simplify stage directory structures by eliminating use of symbolic links. This means, among other things, that` $spack/var/spack/stage` will no longer be the core staging directory. Instead, the first accessible `config:build_stage` path will be used.
Spack will no longer automatically append `spack-stage` (or the like) to configured build stage directories so the onus of distinguishing the directory from other work -- so the other work is not automatically removed with a `spack clean` operation -- falls on the user.
* extends mkdirs with permissions for intermediate folders
Does not use os.makedirs mode parameter because its behavior is changed
with Python 3.7 (it ignores it for intermediate dirs), and moreover it
was not possible to set different modes for newly-created folders
and leaf folder.
reference:
- https://bugs.python.org/issue19930
- https://docs.python.org/3.7/library/os.html#os.makedirs
* comment mkdirp step easing code understanding
* revert mkdir to default for package metapath
since metapath is nested in package folder, there is no need
to specify permissions for intermediate folders because the prefix
already exists.
* comment create_install_directory package modes
The default library search for a package checks the lib/ and lib64/
directories for libraries before the root prefix, in order to save
time when searching for libraries provided by externals (which e.g.
may have '/usr/' as their root).
This moves that logic into the "find_libraries" utility method so
packages implementing their own custom library search logic can
benefit from it.
This also updates packages which appear to be replicating this logic
exactly, replacing it with a single call to "find_libraries".
Fixes#11782
Spack was not properly resolving relative paths to absolute paths
when a relative path was passed to "spack compiler add [PATH]".
Now, if provided a relative path, the absolute path is written to
compilers.yaml rather than the relative path.
For resources, it is desirable to use the expanded archive name of
the resource as the name of the directory when adding it to the root
staging area.
#11528 established 'spack-src' as the universal directory where
source files are placed, which also affected the behavior of
resources managed with Stages.
This adds a new property ('srcdir') to Stage to remember the name of
the expanded source directory, and uses this as the default name when
placing a resource directory in the root staging area.
This also:
* Ensures that downloaded sources are archived using the expanded
archive name (otherwise Spack will not be able to determine the
original directory name when using a cached archive).
* Updates working_dir context manager to guarantee restoration of
original working directory when an exception occurs
* Adds a "temp_cwd" context manager which creates a temporary
directory and sets it as the working directory
