Connection objects are Python version, platform and multiprocessing
start method independent, so better to use those than a mix of plain
file descriptors and inadequate guesses in the child process whether it
was forked or not.
This also allows us to delete the now redundant MultiProcessFd class,
hopefully making things a bit easier to follow.
There was a bit of mystery surrounding the arguments for `_setup_pkg_and_run`. It passes
two file descriptors for handling the `gmake`'s job server in child processes, but they are
unsed in the method.
It turns out that there are good reasons to do this -- depending on the multiprocessing
backend, these file descriptors may be closed in the child if they're not passed
directly to it.
- [x] Document all args to `_setup_pkg_and_run`.
- [x] Document all arguments to `_setup_pkg_and_run`.
- [x] Add type hints for `_setup_pkg_and_run`.
- [x] Refactor exception handling in `_setup_pkg_and_run` so it's easier to add type
hints. `exc_info()` was problematic because it *can* return `None` (just not
in the context where it's used). `mypy` was too dumb to notice this.
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
Change the installer to take `([pkg], args)` in the constructor instead
of `[(pkg, args)]`. The reason is that certain arguments are global
settings, and the new API ensures that those arguments cannot be
different across different "build requests".
The `explicit` install arg is now a list of hashes, and the installer is
no longer responsible for determining what package is installed
explicitly. This way environment installs can simply pass the list of
environment roots, without them necessarily being explicit build
requests. For example an env with two roots [a, b], where b depends on
a, would not always cause spack install to mark b as explicit.
Notice that `overwrite` already took a list of hashes, this makes
`explicit` consistent.
`package.do_install(explicit=True)` continues to take a boolean.
fixes#47101
The bug was introduced in #33495, where `spack find was not updated,
and wasn't caught by unit tests.
Now a Database can accept a custom predicate to select the installation
records. A unit test is added to prevent regressions. The weird convention
of having `any` as a default value has been replaced by the more commonly
used `None`.
Signed-off-by: Massimiliano Culpo <massimiliano.culpo@gmail.com>
Both `multiprocessing.connection.Connection.__del__` and `io.IOBase.__del__` called `os.close` on the same file descriptor. As of Python 3.13, this is an explicit warning. Ensure we close once by usef `os.fdopen(..., closefd=False)`
`setup-env.sh` is meant to be sourced, not executed directly.
By revoking execution permissions, users who accidentally execute
the script will receive an error instead of seeing no effect.
* Remove execution permission from `setup-env.sh` and friends
* Don't make output file executable in `spack commands --update-completion`
---------
Co-authored-by: Todd Gamblin <tgamblin@llnl.gov>
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
in case of inheritance the static tags prop may be updated multiple
times, and it turns out builder classes magically inherit from
traditional package classes
The "use_store" context manager is used to swap the value
of a global variable (spack.store.STORE), while keeping
another global variable consistent (spack.config.CONFIG).
When doing that it tries to evaluate the previous value
of the store, if that was not done already. This is wrong,
since the configuration might be in an "intermediate" state
that was never meant to trigger side effects.
Remove that operation, and add a unit test to
prevent regressions.
`Spec.__getitem__` queries dependent edges, which almost always point to
nodes outside the sub-dag considered. It should only ever look at edges
being traversed.
This modifies heuristic to decay to clingo default
over time. The hope is that this helps with specs
that have an optimal solution with a high penalty.
Let target and compiler heuristic decay too, do not
guess compiler
When we changed how to deal with errors in November,
we didn't realize that for an unconstrained choice
rule it is more important in the heuristic to guess
what is NOT in the answer set, since it will be the
majority of options.
Previously this was following automatically from what
was in the answer set, via `1 { ... } 1` cardinality
constraints.
Here we improve the heuristic and the solve time for specs.
#40773 introduced python-venv, which improved build isolation and avoids issues with,
e.g., `ubuntu`'s system python modifying `sysconfig` to include a (very unwanted)
`local` directory within the default install layout.
This addresses a few cases where #40773 removed functionality, without harming the
default cases where we use `python-venv`.
Traditionally, *every* view with `python` in it was essentially a virtual environment,
because we would copy the `python` interpreter and `os.py` into every view when linking.
We now rely on `python-venv` to do that, but only when it's used (i.e. new builds) and
only for packages that have an `extends("python")` directive.
This again makes every view with `python` in it a virtual environment, but only
if we're not already using a package like `python-venv`. This uses a different
mechanism from before -- instead of using the `virtualenv` trick of copying `python`
into the prefix, we instead create a `pyvenv.cfg` like `venv` (the more modern way
to do it).
This fixes two things:
1. If you already had an environment before Spack `v0.22` that worked, it would
stop working without a reconcretize and rebuild in `v0.22`, because we no longer
copy the python interpreter on link. Adding `pyvenv.cfg` fixes this in a more
modern way, so old views will keep working.
2. If you have an env that only includes python packages that use `depends_on("python")`
instead of `extends("python")`, those packages will now be importable as before,
though they won't have the same level of build isolation you'd get with `extends`
and `python-venv`.
* views: avoid making client code deal with link functions
Users of views and ViewDescriptors shouldn't have to deal with link functions -- they
should just say what type of linking they want.
- [x] views take a link_type, not a link function
- [x] views work out the link function from the link type
- [x] view descriptors and commands now just tell the view what they want.
* python: simplify logic for avoiding pyvenv.cfg in copy views
Signed-off-by: Todd Gamblin <tgamblin@llnl.gov>
