The check for short paths under Windows via os.path.samefile, introduced in #11936, also found similar tests in symlinked tests in the GH Actions CI.
Fixes#12039.
Co-authored-by: Bruno Oliveira <bruno@soliv.dev>
Previously, if more than one fixture finalizer raised, only the first
was reported, and the other errors were lost.
Use an exception group to report them all. This is similar to the change
we made in node teardowns (in `SetupState`).
As detailed in
https://github.com/pytest-dev/pytest/issues/11475#issuecomment-1937043670,
currently with `--import-mode=importlib` pytest will try to import every
file by using a unique module name, regardless if that module could be
imported using the normal import mechanism without touching `sys.path`.
This has the consequence that non-test modules available in `sys.path`
(via other mechanism, such as being installed into a virtualenv,
PYTHONPATH, etc) would end up being imported as standalone modules,
instead of imported with their expected module names.
To illustrate:
```
.env/
lib/
site-packages/
anndata/
core.py
```
Given `anndata` is installed into the virtual environment, `python -c
"import anndata.core"` works, but pytest with `importlib` mode would
import that module as a standalone module named
`".env.lib.site-packages.anndata.core"`, because importlib module was
designed to import test files which are not reachable from `sys.path`,
but now it is clear that normal modules should be imported using the
standard mechanisms if possible.
Now `imporlib` mode will first try to import the module normally,
without changing `sys.path`, and if that fails it falls back to
importing the module as a standalone module.
This also makes `importlib` respect namespace packages.
This supersedes #11931.
Fix#11475Close#11931
(diff better viewed ignoring whitespace)
In pytest<8, the collection tree for `pyargs` arguments in an invocation
like this:
pytest --collect-only --pyargs pyflakes.test.test_undefined_names
looked like this:
```
<Package test>
<Module test_undefined_names.py>
<UnitTestCase Test>
<TestCaseFunction test_annotationUndefined>
... snipped ...
```
The pytest 8 collection improvements changed it to this:
```
<Dir pytest>
<Dir .tox>
<Dir venv>
<Dir lib>
<Dir python3.11>
<Dir site-packages>
<Package pyflakes>
<Package test>
<Module test_undefined_names.py>
<UnitTestCase Test>
<TestCaseFunction test_annotationUndefined>
... snipped ...
```
Besides being egregious (and potentially even worse than the above,
going all the way to the root, for system-installed packages, as is
apparently common in CI), this also caused permission errors when trying
to probe some of those intermediate directories.
This change makes `--pyargs` arguments no longer try to add parent
directories to the collection tree according to the `--confcutdir` like
they're regular arguments. Instead, only add the parents that are in the
import path. This now looks like this:
```
<Package .tox/venv/lib/python3.11/site-packages/pyflakes>
<Package test>
<Module test_undefined_names.py>
<UnitTestCase Test>
<TestCaseFunction test_annotationUndefined>
... snipped ...
```
Fix#11904.
pytest allows a fixture to request its own name (directly or
indirectly), in which case the fixture with the same name but one level
up is used.
To know which fixture should be used next, pytest keeps a mutable
item-global dict `_arg2index` which maintains this state. This is not
great:
- Mutable state like this is hard to understand and reason about.
- It is conceptually buggy; the indexing is global (e.g. if requesting
`fix1` and `fix2`, the indexing is shared between them), but actually
different branches of the subrequest tree should not affect each
other.
This is not an issue in practice because pytest keeps a cache of the
fixturedefs it resolved anyway (`_fixture_defs`), but if the cache is
removed it becomes evident.
Instead of the `_arg2index` state, count how many `argname`s deep we are
in the subrequest tree ("the fixture stack") and use that for the index.
This way, no global mutable state and the logic is very localized and
easier to understand.
This is slower, however fixture stacks should not be so deep that this
matters much, I hope.
According to my understanding, this code, which handles obtaining the
relevant fixturedefs when a dynamic `getfixturevalue` is used, has an
optimization where it only grabs fixturedefs that are visible to the
*parent* of the item, instead of the item itself, under the assumption
that a fixturedef can't be visible to a single item, only to a
collector.
Remove this optimization for the following reasons:
- It doesn't save much (one loop iteration in `matchfactories`)
- It slightly complicates the complex fixtures code
- If some plugin wants to make a fixture visible only to a single item,
why not let it?
- In the static case (`getfixtureclosure`), this optimization is not
done (despite the confusing name `parentnode`, it is *not* the parent
node). This is inconsistent.
Bruno Oliveira
Bruno Oliveira
Ronny Pfannschmidt
dependabot[bot]
Ran Benita
github-actions[bot]
jakkdl
mrbean-bremen
donghui
pytest bot