* Always allow "no-other-choice" pre-release dependencies when resolving collection dependency tree
PR #81606.
Prior to this patch, when `--pre` CLI flag was not passed, the
dependency resolver would treat concrete collection dependency
candidates (Git repositories, subdirs, tarball URLs, or local dirs or
files etc) as not meeting the requirements.
This patch makes it so pre-releases in any concrete artifact
references, and the ones being specifically pinned dependencies or
user requests, met anywhere in the dependency tree, are allowed
unconditionally.
This is achieved by moving the pre-release check from
`is_satisfied_by()` to the `find_matches()` hook, following the
Pip's example.
As a bonus, this change also fixes the situation when a collection
pre-releases weren't considered if it didn't have any stable releases.
This now works even if `--pre` wasn't requested explicitly.
Finally, this patch partially reverts commit
6f4b4c345b, except for the tests. And it
also improves the `--pre` hint warning to explain that it mostly
affects Galaxy/Automation Hub-hosted collection releases.
Ref #73416
Ref #79112Fixes#79168Fixes#80048Resolves#81605
Co-authored-by: Sloane Hertel <19572925+s-hertel@users.noreply.github.com>
(cherry picked from commit 7662a05085)
* Shorten the collection namespace and name @ tests
This is needed on the 2.15 branch which uses older galaxy containers
that have restricted FQCN size.
* ansible-galaxy fix scm dependency error
also changed usage of 'virtual colleciton' to actual type
avoid error by filtering out virtual collections that dont have
expected properties
simplified as per webknjaz
* Update lib/ansible/galaxy/collection/__init__.py
Co-authored-by: Sviatoslav Sydorenko <wk.cvs.github@sydorenko.org.ua>
(cherry picked from commit 2aef0406d4)
* If all tasks in the role are skipped or unreachable, the role is not marked as complete for the host.
* Only mark the role as complete if a task in the role succeeds or fails for the host.
(cherry picked from commit 8034651cd2)
* Python 3.11.4 introduces a new parameter 'filter' in extract and
extractall in tarfile. Handle deprecation warning message emitted
in Python 3.12.
* added probing mechanism in ansible-galaxy code to detect broken
data filter implementation in tarfile.
Fixes: #80832
(cherry picked from commit 1cc5efa)
Co-authored-by: Abhijeet Kasurde <akasurde@redhat.com>
For backport without the fix for multiple items clobbering previous settings
Co-authored-by: Sloane Hertel <19572925+s-hertel@users.noreply.github.com>
* vault/unvault filters, fixed id field to match documented.
fixes#81420
Co-authored-by: Felix Fontein <felix@fontein.de>
(cherry picked from commit f3a15a4a95)
* [stable-2.15] Revert logic to use Popen.communicate (#80874)
* Back out use of communicate, add better comments, add bufsize, and align with subprocess._communicate
* tests
* re-order logic slightly
* more comments
* loopty loop
* yet another comment
* Revert "yet another comment"
This reverts commit 96cd8ada5fa0441b92f2298bdaa6cb40594847d2.
* Revert "loopty loop"
This reverts commit 96ea066f6a7d18902c04a14f18dd79b38e56f5e7.
* ci_complete
* Copy in comment too
* Wording updates
Co-authored-by: Matt Davis <6775756+nitzmahone@users.noreply.github.com>
* Back out bufsize
---------
Co-authored-by: Matt Davis <6775756+nitzmahone@users.noreply.github.com>.
(cherry picked from commit 553f51e728)
Co-authored-by: Matt Martz <matt@sivel.net>
* Address merge conflict issues
* Enable mypy for the entire packaging directory
* Return CompletedProcess only when capturing output
This allows stdout/stderr on CompletedProcess to be `str` instead of `str | None`.
The unused args on CompletedProcess have been removed.
Overload type hints have been added to reflect these changes.
* Relax return type on ensure_venv
This improves consistency with its usage, since `run` accepts `env` of `dict[str, t.Any]`.
Also removed unnecssary `str()` usage when updating `env`.
* Fix type hint on suppress_when
* Fix callable annotation
* Add type hint for command_parser
PyCharm complains about using a protected member, and also that it can't find the type in the type stubs.
However, mypy properly recognizes the type.
* Avoid unnecessary TypeVar usage
(cherry picked from commit 47ab59753c)
* now uses class property instead of only relying on network_os variable
---------
Co-authored-by: Martin Krizek <martin.krizek@gmail.com>
(cherry picked from commit 6db006a5fa)
* Consider all configured collection paths when installing collections (#81243)
* Only install collections which can't be satisfied by a collection in any of the configured paths.
* Improve warning for unexpected collection install path
Fix warning when path is configured, but is a pip-managed path
Normalize the path before validating to fix warning consistency
(cherry picked from commit efbc00b6e4)
* Update warning patch for version 2.15 due to non-existence of the config GALAXY_COLLECTIONS_PATH_WARNING
Frozen requirements can now preserve any explicitly installed package that would normally be omitted, not just setuptools.
(cherry picked from commit dbb3feddaf)
* Clean up MANIFEST.in (#80688)
* Remove unnecessary MANIFEST.in commands
This eliminates more setuptools warnings during build.
* Sort MANIFEST.in
Now that only include commands are used, the order of the commands no longer effects the build output.
(cherry picked from commit 6e325d9e4d)
* Omit hacking directory from MANIFEST.in (#81245)
* Omit `hacking` directory from `MANIFEST.in`
* Update package-data sanity test
(cherry picked from commit b93a628aed)
Since man pages aren't accessible to users after a `pip install`, there's no need to include them in the sdist.
This change makes it trivial to build man pages from source, which makes them much easier to iterate on.
It also simplifies creation and testing of the sdist, since it no longer requires building man pages.
The new `packaging/cli-doc/build.py` script can generate both man pages and RST documentation.
This supports inclusion on the docs site without a dependency on `ansible-core` internals.
Having a single implementation for both simplifies keeping the two formats in sync..
(cherry picked from commit 691c8e8603)
This change fixes bugs in the manpage generator that existed since it
was first added.
It exposes CLI `ARGUMENTS` value to manpage templates.
Before this change, the code contained a typo, causing the `for`-loop
iterate over individual characters of the `'ARGUMENTS'` string rather
than iterating over a tuple. A missing comma was at fault.
The updated code gets rid of the `for`-loop and conditionals since it
seems to have been a premature complexity increase and no other things
than `'ARGUMENTS'` were ever added into the broken iterable.
The functional change is that `arguments` is now always present in the
Jinja2 context, unlike being missing sometimes because of the previous
design (not that it was ever present, because of the bug! sigh...)
The Jinja2 templates perform an `{% if arguments %}` check, letting
the template engine silently ignore the missing variable. The clause
was always falsy, meaning that the arguments section was not included
in the manpages for at least the last 6 years. With this fix, it will
be.
This patch also deduplicates calling `opt_doc_list` @ generate_man.
It was called late in the execution, more times than necessary. This
patch makes sure it happens once by putting it at the top of the scope.
It fixes rendering library and inventory in manpages.
The corresponding Jinja2 templates have blocks wrapped with
conditionals like `{% if inventory %}` and `{% if library %}` but said
variables were never injected into the context, nor were they even
deduced on the Python side of the generator. This means that the
conditional clauses were always falsy, never showing the portions of
the manpages.
The Python script has hints for how the `inventory` variable was to be
calculated, which is confirmed through the Git paleontology efforts.
The block of code that references to the `inventory` bit was
incorrectly checking a variable with a list of nested objects for the
presence of a string which was never going to work.
This patch fixes this check by verifying the CLI flag against the
correct variable containing a list of options and exposes it to the
Jinja2 templates.
It also exposes the `library` variable in a similar way.
The block displaying other binaries in Sphinx CLI docs has been
synchronized with the manpage template.
Previously, the current binary was displayed also. This patch gets rid
of the unwanted trailing comma there too.
Finally, the CLI executables list in the manpage template now reuses
the same variable as the RST template that doesn't need any
post-processing in Jinja2.
Before, it was already used in the RST template so this patch aligns
both templates to use the same logic as they got out-of-sync over time.
PR #80450..
(cherry picked from commit a84b3a4e72)
Co-authored-by: Sviatoslav Sydorenko <webknjaz@redhat.com>