* fix coverage output from synthetic packages
* synthetic packages (eg, implicit collection packages without `__init__.py`) were always created at runtime with empty string source, which was compiled to a code object and exec'd during the package load. When run with code coverage, it created a bogus coverage entry (since the `__synthetic__`-suffixed `__file__` entry didn't exist on disk).
* modified collection loader `get_code` to preserve the distinction between `None` (eg synthetic package) and empty string (eg empty `__init__.py`) values from `get_source`, and to return `None` when the source is `None`. This allows the package loader to skip `exec`ing things that truly have no source file on disk, thus not creating bogus coverage entries, while preserving behavior and coverage reporting for empty package inits that actually exist.
* add unit test
* Validate salt when using crypt. Respect salt_size in password lookup. Repair salt for bcrypt. Fixes#71107. Fixes#53750. Fixes#36129.
* Handle algorithms we don't know about, and make sure to return the salt
* Account for old passlib
* Add tests for salt constraints
* Add changelog fragment
* Add test for #36129
* config: singular ANSIBLE_COLLECTIONS_PATH
Every other *_PATH setting in ansible is singular, and the traditional
$PATH variable is also singular despite containing a list of
directories. Let's be consistent both internally and with POSIX
tradition.
* update all ANSIBLE_COLLECTIONS_PATHS env references to be singular
* deprecate plural ANSIBLE_COLLECTIONS_PATHS setting
* find_module can't pop ImportError- we need to just translate to `None` since this is a normal condition with files on sys.path (eg `/usr/lib/python36.zip`)
* added test
* `meta/` directory in collections
* runtime metadata for redirection/deprecation/removal of plugin loads
* a compatibility layer to keep existing content working on ansible-base + collections
* a Python import redirection layer to keep collections-hosted (and otherwise moved) content importable by things that don't know better
* supported Ansible version validation on collection loads
* Support pre-releases via new SemanticVersion. Fixes#64905
* Don't treat buildmeta as prerelease
* Don't inherit from str and int
* Add helper method to try and construct a SemanticVersion from a LooseVersion
* Don't count major 0 as pre-release, it's different
* Guard against invalid or no version in LooseVersion
* return a bool
* Add integration tests for pre-release
* Fix up lingering issues with comparisons
* typo fix
* Always allow pre-releases in verify
* Move pre-release filtering into CollectionRequirement, add messaging when a collection only contains pre-releases
* Update changelog
* If explicit requirement allow pre releases
* Enable pre-releases for tar installs, and collections already installed when they are pre-releases
* Drop --pre-release alias, make arg name more clear
* Simplify code into a single line
* Remove build metadata precedence, add some comments, and is_stable helper
* Improve from_loose_version
* Increase test coverage
* linting fix
* Update changelog
* Use is_sequence, and Mapping throughout, add support for tuples. Fixes#65722
* Address tests
* Remove unused import
* Add changelog
* Add docstring for clarity
* Argh, linting fix
* Not chasing this rabbit
* wrap_var doesn't return a ref to the original item
* no ref tests
* Remove unused import
* Remove sanity/ignore.txt entries
* replace use of "_" as a variable name
* Cleanup boilerplate
* Remove default values from mandatory parameters
* Sanity test documentation updates
* Remove unused imports from iam_role
* add subdir support to collection loading
* collections may now load plugins from subdirs under a plugin type or roles dir, eg `ns.coll.subdir1.subdir2.myrole`->ns.coll's roles/subdir1/subdir2/myrole, `ns.coll.subdir1.mymodule`->ns.coll's plugins/modules/subdir1/mymodule.py
* centralize parsing/validation in AnsibleCollectionRef class
* fix issues loading Jinja2 plugins from multiple sources
* resolves#59462, #59890,
* sanity test fixes
* string fixes
* add changelog entry
* Remove UnsafeProxy
Move the work from UnsafeProxy to wrap_var and add support for bytes.
Where wrap_var is not needed, use AnsibleUnsafeBytes/AnsibleUnsafeText
directly.
Fixes#59606
* item is not always text
* Address issues from reviews
* ci_complete
- Support more import statements:
from ansible_collections.ansible.builtin.plugins.module_utils import basic
from ansible_collections.ansible.builtin.plugins.module_utils.basic import AnsibleModule
- Add unit tests for more import statements.
- Raise ImportError instead of returning None if load_module fails.
* Use `compile` before `eval` in collection loader.
This fixes two issues:
1. File names are available when tracing execution, such as with code coverage.
2. Future statements are not inherited from the collection loader.
* Add unit tests for collection loading.
These tests verify several things:
1. That unit tests can import code from collections when the collection loader is installed.
2. That tracing reports the correct file and line numbers (to support code coverage).
3. That collection code does not inherit __future__ statements from the collection loader.
* Update unit test handling of the collection loader.
Since the collection loader is installed simply by importing ansible.plugins.loader,
we may already have a collection loader installed when the test runs. This occurs if
any other tests are collected which use that import during collection. Until that code
is moved into an initialization function to avoid loading during import, the unit tests
will need to replace any existing collection loaders so that they reflect the desired
configuration.
* Insert into sys.modules before calling exec.
This is a requirement of PEP 302.
It will prevent recursion errors when importing the current module or using a relative import.
* Use the correct value for __package__ in modules.
This allows using relative imports in collections.
* Add warning about modifying code for trace test.
* Add test for relative import in collection.
* Add __init__.py to collection to satisfy pylint.
The relative-beyond-top-level rule in pylint may not be appropriate for collections.
However, until that rule is disabled for collections this will keep tests passing.
* Skip gitlab tests if dependencies aren't met
* Skip certain unittests if passlib is not installed
* Fix tests with deps on paramiko to skip if paramiko is not installed
* Use pytest to skip for cloudstack
If either on Python-2.6 or the cs library is not installed we cannot run
this test so skip it
* Add tests for KubeAPIVersion
* Legibility improvements for KubevirtVM tests
* Create units.utils.kubevirt with common stuff
* Add some VMIRS unit tests
* Note: Python2 is not as intelligent at detecting false import loops as
Python3. context_objects.py cannot be added to cli/arguments because it
would set up an import loop between cli/__init__.py,
cli/arguments/context_objects.py, and context.py on Python2.
ci_complete
* Move ansible.compat.tests to test/units/compat/.
* Fix unit test references to ansible.compat.tests.
* Move builtins compat to separate file.
* Fix classification of test/units/compat/ dir.
* Share the implementation of hashing for both vars_prompt and password_hash.
* vars_prompt with encrypt does not require passlib for the algorithms
supported by crypt.
* Additional checks ensure that there is always a result.
This works around issues in the crypt.crypt python function that returns
None for algorithms it does not know.
Some modules (like user module) interprets None as no password at all,
which is misleading.
* The password_hash filter supports all parameters of passlib.
This allows users to provide a rounds parameter, fixing #15326.
* password_hash is not restricted to the subset provided by crypt.crypt,
fixing one half of #17266.
* Updated documentation fixes other half of #17266.
* password_hash does not hard-code the salt-length, which fixes bcrypt
in connection with passlib.
bcrypt requires a salt with length 22, which fixes#25347
* Salts are only generated by ansible when using crypt.crypt.
Otherwise passlib generates them.
* Avoids deprecated functionality of passlib with newer library versions.
* When no rounds are specified for sha256/sha256_crypt and sha512/sha512_crypt
always uses the default values used by crypt, i.e. 5000 rounds.
Before when installed passlibs' defaults were used.
passlib changes its defaults with newer library versions, leading to non
idempotent behavior.
NOTE: This will lead to the recalculation of existing hashes generated
with passlib and without a rounds parameter.
Yet henceforth the hashes will remain the same.
No matter the installed passlib version.
Making these hashes idempotent.
Fixes#15326Fixes#17266Fixes#25347 except bcrypt still uses 2a, instead of the suggested 2b.
* random_salt is solely handled by encrypt.py.
There is no _random_salt function there anymore.
Also the test moved to test_encrypt.py.
* Uses pytest.skip when passlib is not available, instead of a silent return.
* More checks are executed when passlib is not available.
* Moves tests that require passlib into their own test-function.
* Uses the six library to reraise the exception.
* Fixes integration test.
When no rounds are provided the defaults of crypt are used.
In that case the rounds are not part of the resulting MCF output.
* test/: PEP8 compliancy
- Make PEP8 compliant
* Python3 chokes on casting int to bytes (#24952)
But if we tell the formatter that the var is a number, it works
This feature changes the scalar value of `serial:` to a list, which
allows users to specify a list of values, so batches can be ramped
up (commonly called "canary" setups):
- hosts: all
serial: [1, 5, 10, "100%"]
tasks:
...
On Python 2, shlex.split() raises if you pass it a unicode object with
non-ASCII characters in it. The Ansible codebase copes by explicitly
converting the string using to_bytes() before passing it to
shlex.split().
On Python 3, shlex.split() raises ('bytes' object has no attribute 'read')
if you pass a bytes object. Oops.
This commit introduces a new wrapper function, shlex_split, that
transparently performs the to_bytes/to_unicode conversions only on
Python 2.
Currently I've only converted one call site (the one that was causing a
unit test to fail on Python 3). If this approach is deemed suitable,
I'll convert them all.
Matt Davis