`ansible-galaxy init --offline ...` can create a role without
talking to the galaxy api server
`ansible-galaxy install ...` only needs to talk to the galaxy api
server for galaxy roles, not tar files or scm archives
Fixed a bug in command line role installation
Older git archive commands create tar archives even with a tar.gz
extension. So change it to always create tar archives and have
the install_role method cope.
Removed ssh roles from the test case as they don't work unless
you can connect to bitbucket via ssh and have your key there.
Corrected a minor typo in error messages
* Roles can now be given a friendly name as third field in role spec csv
* Roles can be installed from URL (not just from archived SCMs)
* Integration tests to demonstrate this
* Unit tests to ensure that role spec parsing works as expected
Ensure that if meta/main.yml has dependencies containing URLs
these are correctly resolved at playbook runtime.
Update role_name during run time so that output of e.g. task names
look sensible
When trying to create a directory relative to the current working
directory, a directory is created at the root of the filesystem
instead. This patch ensures that directories specified with relative
paths will be created in the current working directory. Fully
qualified paths aren't affected
Signed-off-by: Major Hayden <major@mhtx.net>
This addresses a bug in ansible-pull where running ansible-pull
with an existing inventory causes the ansible job that does
the SCM checkout to run twice - once for localhost and once
for the fully qualified hostname.
This can cause a race condition, and usually results in one
of the ansible checkouts failing because one of the scm checkouts
has its references updated underneath it. Although the SCM checkout
actually succeeds, ansible fails with non-zero exit status, and
so ansible-pull does not continue.
Now that localhost is implicit for ansible runs, the ansible
scm checkout can be done using just localhost as a target.