This reverts the previous commit (d71834d) and instead fixes the problem
by making sure that options used by ConnectionInformation do not have
defaults set in the playbook objects so they're properly inherited from
the CLI options object if not otherwise specified in the play
Moving this above the play setting means that any default values set
in play (like connection) override any corresponding CLI option, which
is wrong. Generally CLI options should override things set in playbooks
Before this change if a variable was of type int or bool and the variable was referenced
by another variable, the type would change to string.
eg. defaults/main.yml
```
PORT: 4567
OTHER_CONFIG:
secret1: "so_secret"
secret2: "even_more_secret"
CONFIG:
hostname: "some_hostname"
port: "{{ PORT }}"
secrets: "{{ OTHER_CONFIG }}"
```
If you output `CONFIG` to json or yaml, the port would get represented in the output as a
string instead of as a number, but secrets would get represented as a dictionary. This is
a mis-match in behaviour where some "types" are retained and others are not. This change
should fix the issue.
Update template test to also test var retainment.
Make the template changes in v2.
Update to only short-circuit for booleans and numbers.
Added an entry to the changelog.
* Elminates a lot of logic in the AnsibleComposer class.
* Update tests with new column offsets. The rule should now be
consistently: Column is the start of the entry's value (so for
strings, the first non-space after the entry beginning, for dicts, the
first character of the first key)
* Changes AnsibleConstructor so that only unicode strings are returned
(no str type)
* Tracks line, column numbers for strings
* Adds unittests for AnsibleLoader (generic for all the yaml parsing)
The rest of ansible uses validate_certs, so make that the main
documented parameter. However, leave verify as an alias since that's the
passthrough value to the underlying libraries.
* Do backslash escape parsing in parse_kv() [was being done in the copy
module purely for newlines in the copy module's content param before]
* Make parse_kv always return unicode
* Add bandaid to transform args to unicode until we can fix things
calling parse_kv to always send it unicode.
* Make split_args deal with unicode internally. Warning, no bandaid for
things calling split_args without giving it unicode (shouldn't matter
as dealt with str internally before)
* Fix copy and unarchive action plugins to not use setdefaultencoding
* Remove escaping from copy (it was broken and made content into latin-1
sometimes). escaping is now in parse_kv.
* Expect that content is now a unicode string so transform to bytes just
before writing to the file.
* Add initial unittests for split_args and parse_kv. 4 failing
tests.because split_args is injecting extra newlines.
- fixed issue with previous commit with bad constants vs C ref on become
- added list-tags
- rearranged common options to utils/cli.py
- added generic validate for both vault and become conflicts
- removed dupes and conflicting options
Note: In v1 we fix this by transforming into unicode just before we use
it (when we send it to jinja2) because jinja2 cannot handle non-ascii
characters in str.
In v2 our model is that all text values need to be stored as unicode
type internally. So we transform this to unicode when we read it from
the inventory file and save it into the internal dict instead.
- become constants inherit existing sudo/su ones
- become command line options, marked sudo/su as deprecated and moved sudo/su passwords to runas group
- changed method signatures as privlege escalation is collapsed to become
- added tests for su and become, diabled su for lack of support in local.py
- updated playbook,play and task objects to become
- added become to runner
- added whoami test for become/sudo/su
- added home override dir for plugins
- removed useless method from ask pass
- forced become pass to always be string also uses to_bytes
- fixed fakerunner for tests
- corrected reference in synchronize action plugin
- added pfexec (needs testing)
- removed unused sudo/su in runner init
- removed deprecated info
- updated pe tests to allow to run under sudo and not need root
- normalized become options into a funciton to avoid duplication and inconsistencies
- pushed suppored list to connection classs property
- updated all connection plugins to latest 'become' pe
- includes fixes from feedback (including typos)
- added draft docs
- stub of become_exe, leaving for future v2 fixes
There is an old PR that shows a great use case for having a different
set of states for the server module. Before the other modules start
being in real use, pull this out so that we don't get ourselves into a
pickle.
Prevents an exception from raising in some corner cases where SSH might
be misconfigured. It seems not to fix a specific problem, although it
makes the method a little bit more solid.
We have been caching the paths to the plugins but before we would only
cache the specific plugin that we were looking for. This meant that we
might search through all of the plugin directories before finding the
specific module we were interested in. The next plugin we needed we
might again search through all the plugin directories before finding the
plugin we wanted.
This new code will cache all the potential plugins in each directory
searched. For a play that only uses one plugin we'll only search
through directories until we encounter the directory that has the proper
plugin. For a large playbook with many plugins used we'll search
through each directory at most once each.
From v1: 68e86de26f730c2ba403
Dynamic task includes still need some work, this is a rough first version.
* doesn't work with handler sections of playbooks yet
* when using include + with*, the insertion order is backwards
* fix potential for task lists to be unsynchronized when using the linear
strategy, as the include conditional could be predicated on an inventory
variable
