* Allow a collection role to call a standalone role by default. Fixes#69101
* tweaked changelog text
* Guard against NoneType
Co-authored-by: Matt Davis <nitzmahone@users.noreply.github.com>.
(cherry picked from commit da98fc267a)
Co-authored-by: Matt Martz <matt@sivel.net>
* Fix fileglob when using 'file*' vs 'stuff/file*'
when not having dir in glob, files/ subdir was being ignored.
* tests for fileglob
(cherry picked from commit d3cab602a5)
Change:
We were only testing dnf on RHEL previously.
Test on CentOS 8 as well.
Test Plan:
Ran locally in docker.
Signed-off-by: Rick Elrod <rick@elrod.me>
Change:
Extend the logic for custom error handling in the dnf module, so that on
newer DNF (such as DNF that ships with modern Fedora 31 container
images, and ships with RHEL 8.2) we report errors consistently with
older DNF.
Test Plan:
Ran dnf integration tests against an old Fedora 31 container image and a
brand new Fedora 32 container image; tess passed on both.
Signed-off-by: Rick Elrod <rick@elrod.me>
* fixed fetch traversal from slurp
* ignore slurp result for dest
* fixed naming when source is relative
* fixed bug in local connection plugin
* added tests with fake slurp
* moved existing role tests into runme.sh
* normalized on action excepts
* moved dest transform down to when needed
* added is_subpath check
* fixed bug in local connection
fixes#67793
CVE-2019-3828
(cherry picked from commit ba87c225cd)
- also only replace when needed
- switched from replace to index
- added test to verify bogus_facts are not accepted
CVE-2020-10684
(cherry picked from commit a9d2ceafe4)
* fix vault tmpe file handling
* use local temp dir instead of system temp
* ensure each worker clears dataloader temp files
* added test for dangling temp files
* added notes to data loader
CVE-2020-10685
(cherry picked from commit 6452a82452)
* subversion module - provide password securely with svn command line option --password-from-stdin when possible, and provide a warning otherwise.
* Update lib/ansible/modules/source_control/subversion.py.
* Add a test.
Co-authored-by: Sam Doran <sdoran@redhat.com>
(cherry picked from commit d91658ec0c)
* ansible-galaxy - Fix tar path traversal issue during install - CVE-2020-10691 (#68596)
(cherry picked from commit a20a527014)
* Remove extra tests missing from rebase
* Updated nxos_lacp tests to handle platforms not supporting lacp system mac command
* nxos_lacp: addressing comments
* nxos_lacp: Updating image tag search to include more tags
(cherry picked from commit 00193f27eb)
Add changelog for nxos_lacp fix
* Add CentOS 8 to the test matrix (#63649)
(cherry picked from commit 2a7623dd5c)
* Skip PostgreSQL tests on CentOS 8
The tests in devel have diverged significantly from what is in
stable-2.9. It is easiest to skip these test for CentOS 8 in this branch
since they are still being run in devel.
* Actually inspect the paths and prevent escape
* Add integration tests
* Generate zip files for use in integration test
* Adjust error message
(cherry picked from commit d30c57ab22)
* Update tests to use RHEL 7.8.
Keeping support for RHEL 7.6 since collections are still using it.
* Fix tests for RHEL 7.7+ due to extras repo name change..
(cherry picked from commit 04edd77c42)
Co-authored-by: Matt Clay <mclay@redhat.com>
* add changelog fragment
Signed-off-by: Rick Elrod <rick@elrod.me>
* Update changelogs/fragments/ansible-test-opensuse-15.1.yml
Co-Authored-By: Matt Clay <matt@mystile.com>
* Update docker.txt to use the OpenSUSE 15.1 container image
Signed-off-by: Rick Elrod <rick@elrod.me>
* handle installing mysql on suse
Signed-off-by: Rick Elrod <rick@elrod.me>
* attempt to get tests passing again
Signed-off-by: Rick Elrod <rick@elrod.me>
Co-authored-by: Matt Clay <matt@mystile.com>
* ansible-galaxy - optimise some paths and use fake galaxy int tests (#67685)
* ansible-galaxy - optimise some paths and use fake galaxy int tests
* Added init, built, and publish tests
* Test against both mocked Galaxy and AH server
* Finish off writing the install tests
* Fix up broken tests
* Rename test target and add migrated tests
* Use cloud provider for Galaxy implementation
* Added blank static config
* Use correct alias group
* Set release version and fix copy typo
* Remove reset step as it is no longer needed
* Use sane env var names for test container name
(cherry picked from commit 26129fcb80)
* Use --api-key and not --token
* Set fallaxy tests as a smoketest
(cherry picked from commit b241c021b7)
* [stable-2.9] Remove Tower module tests from CI. (#67368)
* Remove Tower module tests from CI.
The required AMIs are no longer available.
* Mark Tower tests as unsupported..
(cherry picked from commit b041d96762)
Co-authored-by: Matt Clay <mclay@redhat.com>
* Mark tower_common unsupported.
* nxos_interfaces: RMB state fixes
* shippable fixes
* Add add'l comments per review
* fix long line
* Fix mode/enabled system defaults handling
* fix N3L test skips
* lint
* test updates for titanium images
* doc fix
(cherry picked from commit d72025be75)
Add changelog for nxos_interfaces fix
Co-authored-by: Chris Van Heuveln <cvanheuv@cisco.com>
* Make banner detection non-greedy in ios_banner (#63092)
* Make banner detection non-greedy in ios_banner
* Added ios_banner unit test to detect bug #63091
* Corrected PEP8 errors
* Added integration test
* Corrected typo in integration test
(cherry picked from commit 01a92f0191)
* Added changelog fragment
* Update 66274-ios_banner_nongreedy.yml
Co-authored-by: kalimsshar <34369784+kalimsshar@users.noreply.github.com>
Co-authored-by: Matt Davis <nitzmahone@users.noreply.github.com>
Add integration test
There are a number of other parameters that result in stack traces as well when this module is used ad-hoc. I'm not sure if we're interested in fixing them all since this module isn't meant to be run ad-hoc.
(cherry picked from commit cc2376b782)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* throttle tests: fix detection of parallel execution
The test wasn't able to detect if too many workers were running.
On my laptop:
- without this change, the 'throttle' target takes ~20 seconds
- with this change, the 'throttle' target takes ~70 seconds
- 1 second isn't long enough to encounter the issue
* Fix throttle test when strategy is 'free' based
'free' strategy allows multiple tasks to be executed in parallel: use
one 'throttledir' per task.
Use 'linear' strategy with a dedicated play for cleanup/setup tasks
* throttle: reset worker idx before queuing a new task
* TestStrategyBase: define task.throttle
otherwise '1' will be used instead of the default value due to the
following expression being equal to '1':
int(templar.template(task_mock.throttle))
Co-authored-by: James Cammarata <jimi@sngx.net>
(cherry picked from commit bbbdc1c25c)
* Fix fact gathering for sensor-groups (#66439)
(cherry picked from commit 5c158755fc)
Add changelog for nxos_telemetry fix
* Update 66439_fix_nxos_telemetry_facts.yaml
Co-authored-by: Mike Wiebe <mwiebe@cisco.com>
Co-authored-by: Matt Davis <nitzmahone@users.noreply.github.com>
* Add integration tests
* Handle error in _get_diff_data()
* Change to warning rather than error
* Also change failure to warning in assemble action plugin
(cherry picked from commit 9b7198d25e)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* Add constraint for Jinja2 on Python 2.6.
* Fix constraint in inventory_aws_conformance test.
* Add constrraints for template_jinja2_latest test.
(cherry picked from commit 965854fbd2)
Co-authored-by: Matt Clay <matt@mystile.com>
Fixes#66549
The inefficiency improvement
https://github.com/ansible/ansible/pull/63713 introduced a bug where
`enablerepo` was not being honored if combined with
`disablerepo="*"`. This fixes that issue.
Signed-off-by: Adam Miller <admiller@redhat.com>
(cherry picked from commit 22fe22f796)
Co-authored-by: Adam Miller <admiller@redhat.com>
Co-authored-by: Adam Miller <admiller@redhat.com>
* Add test constraint for setuptools.
* Update pip test to work on centos6 container.
(cherry picked from commit 51e5b714e0)
Co-authored-by: Matt Clay <matt@mystile.com>
* Allow updating of ec2_group rules with EC2 classic ELB targets
Fix regression introduced in #45296 with EC2 Classic SGs
Fixes: #57247
Also add (unsupported) ec2 classic test suite with test case for this scenario
* move ec2 classic tests to conditional within ec2_group target
* clean up ec2_classic tests
* ec2_classic account can't run most ec2_group tests
* Use correct var, move cleanup for async
* Add changelog and tests. Fixes#65393. Fixes#65277.
* Kill off all long running async tasks from listen_ports_facts
* Update task to work with older jinja2
(cherry picked from commit 03a4edb)
Co-authored-by: Matt Martz <matt@sivel.net>
* Simply sorting of Windows files below other plugin types
Using the sort method with a custom key function uses less memory than creating multiple lists then joining them.
This seemed to be an acceptable use of a lamdba, even though I geneally try to avoid them.
* Fix sorting of plugins inside of collections
Explicitly sort Windows files below others, mimicking what we do in plugin/loader.py
* Add documentation about ansible.builtin and ansible.legacy
Also document to the two different methods used for searching based on the candidate type.
* Add changelog
* Add integration test
* Update comment with expected sort order
(cherry picked from commit 6f76a48f59)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* Add aws elb multiple host header support (#65021)
* Add support for aws elbv2 multiple host header actions
* Add example of multiple host header
* Add test for multiple host headers
* Add missing quote
* Update test/integration/targets/elb_application_lb/tasks/test_modifying_alb_listeners.yml
Co-Authored-By: Jill R <4121322+jillr@users.noreply.github.com>
* Update test/integration/targets/elb_application_lb/tasks/test_modifying_alb_listeners.yml
Co-Authored-By: Jill R <4121322+jillr@users.noreply.github.com>
* Remove debug statements
(cherry picked from commit d52af75c68)
* elbv2 - print() statement results an task failure, even if the task itself was successfull (#65183)
(cherry picked from commit cba9abcbec)
* Add changelog fragment.
* Update changelogs/fragments/65021-65183-elb-bugs.yml
* Force values to be sorted when compared. (#65315)
* Force values to be sorted when compared.
This avoids problems where lists of values could be generated in
indeterminate order.
* Update lib/ansible/module_utils/aws/elbv2.py
Thank you felixfontein, sorry I missed this detail.
Co-Authored-By: Felix Fontein <felix@fontein.de>
(cherry picked from commit 777b6d3e30)
* Update changelog.
Co-authored-by: Michael Mayer <mjmayer@gmail.com>
Co-authored-by: Markus Bergholz <markuman@gmail.com>
Co-authored-by: roberbri <roberbri@cisco.com>
self._get_user_property returns a string, so when doing a comparison
using this value, cast the second variable to a string so that the
comparison behaves correctly
* Add changelog
* Add to_text import
* Add integration test.
(cherry picked from commit c73288ad53)
Co-authored-by: John Chen <kryptonite303@users.noreply.github.com>
* Ensure `allow_duplicates: true` enables to run single role multiple times(#64902)
* Changed return value in `_load_roles` . Fixes#64902
* Add changelog fragment
* Add an integration test for the issue
* Fix changelog generation error and integration test.
* Fix yaml syntax error in changelog fragment
(cherry picked from commit daecbb9bf0)
* Fix multiple subnet (of same IP version) idempotence for docker_network.
* Add changelog.
* Unit tests no longer make sense, since the part of the code they test has been removed.
* Re-add CIDR validation. Move it to better position (module setup instead of idempotence check).
* Update changelog.
* Only run new tests on VM test images.
* Actually do what is documented. Especially since an empty object is a valid value for aux_addresses.
(cherry picked from commit 17ef253ad1)
The check for this parameter was missing from BusyBox.modify_user(), resulting in unexpected password changes to existing user accounts.
(cherry picked from commit 18130e1419)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* Parse Healthcheck.StartPeriod properly
* Add changelog fragment
* Use proper markup in changelog
Co-Authored-By: Felix Fontein <felix@fontein.de>
Co-authored-by: Felix Fontein <felix@fontein.de>
(cherry picked from commit f31b8e08b2)
* Fix nxos_file_copy option value path validation (#65423)
* Fix nxos_file_copy option value path validation
* Modify `local_file`, `local_file_directory` and
`remote_file` option type from `str` to `path`
so that the option value is validated in Ansible
for a legitimate path value
* Fix review comments
(cherry picked from commit 88008badb1)
* Fix review comments
* DNF Handle Empty AppStream stream definition
Fixes#63683
Signed-off-by: Adam Miller <admiller@redhat.com>
* Switch Fedora dnf test target modularity to stratis
In Fedora 29, the metadata was not properly set for a default stream
for ripgrep even though there is a profile called "default", however
that's an arbitrary string and the module maintainer must set the
default stream (which it never was for the ripgrep module, thus
failing the "empty stream" install test)
Signed-off-by: Adam Miller <admiller@redhat.com>
(cherry picked from commit ed86907587)
Co-authored-by: Adam Miller <admiller@redhat.com>
This avoids installing dnf-plugins-core, which breaks the yum and dnf modules
when uninstalling packages using a wildcard after they have already been removed.
This should resolve issues with the yum integration tests failing after docker tests run.
(cherry picked from commit a5c36eedd8)
Co-authored-by: Matt Clay <matt@mystile.com>
podman 1.4 now installs docker man pages. Add handler to remove docker packages to avoid test failure due to this conflict.
Adjust inventory_docker_swarm integration test
Add conditional to cleanup handlers to allow use of the setup_docker role without removing packages at the end of the play. The inventory_docker_swarm integration tests does its own cleanup..
(cherry picked from commit 6b2b876d43)
Co-authored-by: Sam Doran <sdoran@redhat.com>
Lookup 'first_found' returns empty list which results in
raw_params checking. Check NoneType for 'raw_params' before
proceeding.
Fixes: #64939
Signed-off-by: Abhijeet Kasurde <akasurde@redhat.com>
(cherry picked from commit 8d0c2cd4d5)
* docker_swarm_service: Sort lists when checking for changes
When two lists are checked for changes in this module, the lists are
reported changed when the order of the items is different. This PR
resolves this issue.
* docker_swarm_service: Minor typo fix
* docker_swarm_service: Another minor typo
* docker_swarm_service: Should use sorted(), not sort()
* docker_swarm_service: Sort lists of dictionaries
* docker_swarm_service: Fix style issues in tests
* docker_swarm_service: Updates to integration tests
* docker_swarm_service: Casting string types within lists when comparing
* docker_swarm_service: Special handling of unordered networks with ordered aliases
* docker_swarm_service: Sorting network lists
* docker_swarm_serivce: Better unit test code coverage for lists and networks
* docker_swarm_service: Fixed coding style for sanity tests
* docker_swarm_service: More coding style fixes
* docker_swarm_service: Ignoring test for Python < 3
* docker_swarm_service: Update to version info check for backwards compatibility
* docker_swarm_service: Added change fragment #63887
* docker_swarm_service: Better handling of missing sort key for dictionary of lists
* docker_swarm_service: Preventing sorts from modifying in-place
Co-Authored-By: Felix Fontein <felix@fontein.de>
* docker_swarm_service: Removed spurious import in test
* docker_swarm_service: Preventing sorts from modifying more data in-place
Co-Authored-By: Felix Fontein <felix@fontein.de>
(cherry picked from commit a096cd08c5)
Refactor tests to run the same tasks with CentOS and Fedora using different variables.
(cherry picked from commit 05a7ce798d)
Co-authored-by: Sam Doran <sdoran@redhat.com>
This is a fix for a regression introduced by Perfy. Since then we mainly
operate on host.name instead of the Host object. In a call to
set_nonpersistent_facts where we set ansible_failed_task and
ansible_failed_result variables we were still passing the object which
led to those vars being undefined.
Fixes#64789
(cherry picked from commit 5c0b2c151c)
- use setup_epel role rather than task which contains correct URL now since it has changed for EL8
- add conditional to setup_epel to prevent attempted installation on non-RHEL distros
(cherry picked from commit f2e605c856)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* win_domain_computer module: Minor documentation error
* Fix idempotence when name != samaccountname
* Added changelog
* Added unsupported alias for CI check
(cherry picked from commit 04b8f75ffa)
Since https://github.com/ansible/ansible/pull/56733, we were not able to apply
firewall rules with no `allowed_hosts` key.
closes: #61332
In addition, this patch ensures the `allowed_hosts` key accepts a dict,
instead of a dict in a single entry list.
```yaml
vmware_host_firewall_manager:
esxi_hostname: "{{ esxi1 }}"
rules:
- name: NFC
enabled: True
allowed_hosts:
- all_ip: False
ip_address:
- "1.2.3.4"
```
Should be written:
```yaml
vmware_host_firewall_manager:
esxi_hostname: "{{ esxi1 }}"
rules:
- name: NFC
enabled: True
allowed_hosts:
all_ip: False
ip_address:
- "1.2.3.4"
```
(cherry picked from commit ab2aaca61d)
Add handlers to setup_mysql_db and setup_mysql8 to remove installed packages..
(cherry picked from commit 3e4ae42256)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* [stable-2.9] Eos vlan override (#63639)
* Fix overridden & deleted in eos_vlans
* Fix vlan creation in overridden
* Right, Python 2.6
(cherry picked from commit 741d529)
Co-authored-by: Nathaniel Case <ncase@redhat.com>
* eos_vlans: Never try to set vlan_id as a property (#63689)
(cherry picked from commit d98482c294)
* Add changelog
* Move tests with docker registry into own target.
* Add docker_login tests.
* Add step which makes sure hello-world:latest is around.
* Make work inside docker container.
* Add dependency.
* Use plaintext password.
* Forgot check_mode.
* Add no_log to avoid double log output in verbose mode.
(cherry picked from commit a79f7e575a)
Improve tests
- add more unit test cases
- add specific integration test with more cases
Testing shows no major downside to calling .strip() twice in a comprehension vs. using a regular for loop and only calling .strip() once. Going with the comprehension for ease of maintenance and because comprehensions are optimized in CPython.
(cherry picked from commit 987265a6ef)
Co-authored-by: Sam Doran <sdoran@redhat.com>
This fixes a regression that was caused by switching from copy() to
deepcopy() when 'saving' variables before templating. Since HostVars
did not implement the __deepcopy__() method, deepcopy returned incorrect
results when host vars were present in the variables.
Fixes#63940
(cherry picked from commit cd8ce16d48)
* [stable-2.9] Set alter_sys=True instead of False to address backwards incompat (#64670)
* Set alter_sys=True instead of False to address backwards incompat
* ci_complete
* Add integration test
* ci_complete
* sanity
* ci_complete
* Changelog fragment
* Update import test and validate-modules to match.
(cherry picked from commit b93d92ef9a)
Co-authored-by: Matt Martz <matt@sivel.net>
* Rebase and add alter_sys to validate-modules
* Move X25519, X448, Ed25519 and Ed448 feature tests to module_utils.
* Correctly sign with Ed25519 and Ed448 keys.
* Fix public key comparison. Ed25519 and Ed448 do not have public_numbers().
* Add tests.
* Add changelog.
* Give better errors for cryptography 2.6.x and 2.7.x.
* Test for new errors.
* Forgot one.
* Used wrong private key.
* Use private key password for CA key. Add more stuff to its certificate.
(cherry picked from commit fed267df03)
Add integration test for copy: deep recursive with remote_src=True
(cherry picked from commit b7e38dfa52)
Co-authored-by: Alexander Korsunsky <A.Korsunsky@gmail.com>
* Fix#63919: don't run os.makedirs on empty dir path
* integration test for lineinfile create: yes without path (Sam Doran <sdoran@redhat.com>)
(cherry picked from commit 3c978a3225)
* Make acl module to work with whitespaces in path
* Added a changelog fragment
* Add quotes to changelog fragment
(cherry picked from commit 504d76e956)
* Fix issue when setting an empty pass to no_log param (#62804)
* Fix issue when setting an empty pass to no_log param
* Fix typo
(cherry picked from commit 322e225830)
* Fix up actual get for older versions
Use a separate variable for the boolean test rather than having the same variable sometimes be a boolean and sometimes be a regular expression match object
Add integration tests to cover this scenario
(cherry picked from commit 29d4d318a5)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* clean "changed" after it has been processed
without this change, a loop of `debug` tasks with `changed_when`
causes the "changed" status to get lost before output
* runme.sh tests for debug loop status
(cherry picked from commit bfd32c9b00)
* Specifying IP addresses needs API version 1.22 or newer.
* Simplify code.
* Use IPAMConfig.IPv*Address instead of IPAddress and GlobalIPv6Address.
* Add changelog.
* Fix syntax errors.
* Add integration test.
* Don't rely on netaddr.
* Normalize IPv6 addresses before comparison.
* Install netaddr, and use it.
(cherry picked from commit 62c0cae29a)
* [stable-2.9] Added changelog fragment
(cherry picked from commit 7da37e5)
Co-authored-by: Andrea Tartaglia <andrea@braingap.uk>
* Fixes --version in ansible-galaxy cli
(cherry picked from commit 202ad4f89a)
* Get no_log parameters from subspec
* Add changelog and unit tests
* Handle list of dicts in suboptions
Add fancy error message (this will probably haunt me)
* Update unit tests to test for list of dicts in suboptions
* Add integration tests
* Validate parameters in dict and list
In case it comes in as a string
* Make changes based on feedback, fix tests
* Simplify validators since we only need to validate dicts
Add test for suboptions passed in as strings to ensure they get validated properly and turned into a dictionary.
ci_complete
* Add a few more integration tests
(cherry picked from commit e9d29b1fe4)
Co-authored-by: Sam Doran <sdoran@redhat.com>
* isa string should rewrap as unsafe in get_validated_value
* _is_unsafe shouldn't be concerned with underlying types
* Start with passwords as text, instead of bytes
* Remove unused imports
* Add changelog fragment
* Update changelog with CVE
* fix default collection resolution in adhoc
* if an adhoc command is run with a playbook-dir under a configured collection, default collection resolution is used to resolve unqualified module/action names
* Set ANSIBLE_PLAYBOOK_DIR in integration tests.
* Fix config conflict in ansible integration test.
* add adhoc default collection test
* text-ify warning string
(cherry picked from commit 6d52bdf4db)
* add ANSIBLE_PLAYBOOK_DIR envvar support
* allows `ANSIBLE_PLAYBOOK_DIR` envvar as a fallback on CLI types that support `--playbook-dir`. This should have been implemented with #59464, but was missed due to an oversight.
* added basic integration test
* make first-class PLAYBOOK_DIR config entry
* update changelog
(cherry picked from commit fd229dcbb5)
The `pids` module returns the list of the PID in a `pids` key.
This change ensures we correctly wait for the end of the previous mongod
instances before we start the next ones.
In addition, we remove an unnecessary `ignore_errors`.
See: #61938
Original PR: https://github.com/ansible/ansible/pull/62627
(cherry picked from commit cee55ab718)
* Ensure k8s apply works with check mode
Update the new predicted object with fields from the previous object
before applying in check mode
Don't log output of `file` with `state: absent` on huge virtualenvs!
Fixes#60510
* Use openshift client fix to improve apply for check mode
Use new apply_object method to get a better approximation
of the expected object in check mode.
Requires released upgrade to openshift
* Add changelog fragment for k8s apply check mode fix
* Update changelogs/fragments/60510-k8s-apply-check-mode.yml
Co-Authored-By: Felix Fontein <felix@fontein.de>
(cherry picked from commit a684bb9f5b)
* Fix plugin names for collection plugins.
Add an integration test to verify plugin __name__ is correct for collection plugins.
* Fix collection loader PEP 302 compliance.
The `find_module` function now returns `None` if the module cannot be found. Previously it would return `self` for modules which did not exist.
Returning a loader from `find_module` which cannot find the module will result in import errors on Python 2.x when using implicit relative imports.
* add changelog
* sanity/units/merge fixes
(cherry picked from commit 1c64dba3c9)
* Add integration tests for ansible-doc.
* Enable tests that now pass
* Cleanup processing of plugin docs
* Mostly separate the steps of processing plugin docs
1) Acquire source data
2) Transform and calculate additonal data
3) Format data for output
4) Output data
format_plugin_doc() is still mixing transformation and formatting but
that should be fixed in a devel-only change
* Raise exceptions in _get_plugin_doc() on errors.
* Remove check to exclude on blacklisted extensions. We already request
only .py files
* If there is no DOCUMENTATION entry in the plugin, raise an exception
from _get_plugin_doc(). Everywhere we use _get_plugin_doc(), this is
treated as an error
* If there is no ANSIBLE_METADATA raise an exception as well as
displaying of docs assumes that this has been set.
* If there is neither DOCUMENTATION nor ANSIBLE_METADATA, warn about the
lack of METADATA and error on the lack of DOCUMENTATION. Lack of
DOCUMENTATION is more important so it is what the user should see.
* Add a few special cases for backwards compat. These should probably
be made errors in 2.10:
* no docs but has metadata shows no documentation rather than an error
* empty plugin file shows no doumentation rather than an error
* Simplify backwards compatibility logic.
(cherry picked from commit 3b86dc3e12)
Matt Martz