* 🧪 Switch macOS 13.2 to 12.0 in CI
The former revealed unexpected flakiness while the latter is the
previous value that was used to be stable. This is a temporary revert.
* Skip lookup_url integration test under macOS 12.0
* ✨ Add macOS 13.2 to `ansible-test`
* 🧪 Replace macOS 12.0 with 13.2 in the CI matrix
* Skip `lookup_url` under macOS 13.2
This is due to https://wefearchange.org/2018/11/forkmacos.rst.html
that manifests itself as follows:
TASK [lookup_url : Test that retrieving a url works] ***************************
objc[15394]: +[__NSCFConstantString initialize] may have been in progress in another thread when fork() was called.
objc[15394]: +[__NSCFConstantString initialize] may have been in progress in another thread when fork() was called. We cannot safely call it or ignore it in t
he fork() child process. Crashing instead. Set a breakpoint on objc_initializeAfterForkError to debug.
ERROR! A worker was found in a dead state
* 📝 Extend ansible-test change note w/ macOS 13.2
* 🐛 Make integration tests compatible w/ modern Git
This patch makes use of the `init.defaultBranch` setting to unify
the test across new and old Git versions since one defaults to
`master` and the other uses `main` for the default branch.
Where possible, it uses the `HEAD` committish to avoid having to
normalize the branch name.
The change fixes the following integration tests:
* `ansible-galaxy`
* `ansible-galaxy-collection-scm` (recursive collection)
* `git`
* 🐛Replace `git-symbolic-ref` with a repo template
This custom Git repository template emulates the `init.defaultBranch` setting
on Git versions below 2.28. Ref: https://superuser.com/a/1559582.
Other workarounds mentioned there, like invoking
`git symbolic-ref HEAD refs/heads/main` after each `git init` turned
out to have mysterious side effects that break the tests in surprising ways.
* 🎨 Make Git integration test non-destructive
This patch makes use of the `$HOME` environment variable to trick Git
into using a user-global config generated in the temporary directory.
Code snippets advertising the `wheel` dependency in the
`pyproject.toml`'s `[build-system].requires` setting were a historical
mistake. This has been corrected in
https://github.com/pypa/setuptools/commit/f7d30a95 but many pyprojects
still have `wheel` in their configs.
It is not needed for building sdists and the corresponding setuptools'
PEP 517 hook that provides requirements for building wheels already
esposes this dependency automatically.
* unarchive - properly handle relative path for dest
* Add integration test
* Return output of underlying commands with increased verbosity
* Revert "Return output of underlying commands with increased verbosity"
This reverts commit a2790c8275cdc5697b65670a0beffdc74b741bf6.
* Warn when a relative destination path was provided
* Create a queue per WorkerProcess to receive intra-task updates
* Update `pause` action to use the worker queue
* Deprecate ConnectionBase()._new_stdin
* Add new `Display` convenience method `prompt_until` to manage both controller- and worker-sourced prompting without cross-fork stdin sharing, in-worker mechanism to handle request-response over new worker queue.
This provides the same test coverage as the previous tests, without the dependency on git.
It also includes many more specific test cases with assertions, instead of simply relying on the code to not raise an exception.
* Add _convert_rst_in_template_to_manpage arg types
* 📦 Make manpage build dependencies conditional
Previously, said dependencies were declared as unconditionally
required even when manpages not needed to be built. This patch
Makes it so they are only required when needed.
* Correct _generate_rst_in_templates returned type
It was marked as Path before this patch but in fact, it's iterable of
paths.
* 🎨 Convert RST to manpage in-memory @ PEP 517
Previously, the automation was writing a temporary templated RST on
disk and calling a helper CLI script on that. But with this change, it
happens with less unnecessary I/O.
Co-Authored-By: Matt Davis <6775756+nitzmahone@users.noreply.github.com>
* 📦Expose sdist manpage build deps unconditionally
Due to a bug in pypa/build, the `get_requires_for_build_sdist()` hook
is always invoked with `config_settings=None`. This means that we
cannot conditionally extend build requirements in said hook.
As a workaround, this patch makes hook pretend that `--built-manpages`
is always passed.
Ref: https://github.com/pypa/build/issues/559.
---------
Co-authored-by: Matt Davis <6775756+nitzmahone@users.noreply.github.com>
This patch creates a thin wrapper around the `setuptools`' PEP 517
build backend in-tree. It features an ability to request generating
the manpage files in the process of building a source distribution.
This toggle is implemented using the `config_settings` mechanism of
PEP 517.
One must explicitly pass it a CLI option to the build front-end to
trigger said behavior. The packagers are expected to use the
following call:
python -m build --config-setting=--build-manpages
This option has no effect on building wheels.
🧪 The change includes integration tests
This test runs building and re-building sdists and wheels with and
without the `--build-manpages` config setting under the
oldest-supported and new `setuptools` pinned.
It is intended to preserve the interoperability of the packaging setup
across Python runtimes.
An extra smoke test also verifies that non PEP 517 interfaces remain functional.
PR #79606
Co-authored-by: Matt Clay <matt@mystile.com>
The current wording could be read as "tasks that are tagged *both* configuration and packages" while in fact it is "tasks that are tagged *either* configuration *or* packages"
Matt Clay