Refining the fix made in #5885.
Merging this quickly after PR as further testing of of earlier fix demonstrated flaws. Those flaws are now removed and tested to be removed.
It turns out that some of the assumptions in #5885 were slightly off. The previous fix relied on a call to the module to creat a tmp_path. This is insufficent as there are few cases that we need to have the tmp directory before we make the module call. If we don't have a tmp_path before we do a recursive call or when we find a file that does not match the remote md5 hash we need to create a tmp directory. Also we are not more percise when we will need to clean up the remote tmp_path.
There is a subtle bug in how the git module currently works. If the
version you request is a tag name, and you've already got the repo
cloned, and the tag name is a new tag, but refers to the already checked
out working copy, the git module would exit early without change. This
is bad as it means the new tag ref was not fetched and could not be used
in later tasks.
This change will check if the version is a remote tag, and if the tag
doesn't exist locally. If that is true, it'll do a fetch.
The activity could still be seen as not a change, because the working
copy won't be updated, if the new tag refers to the already checked out
copy, but that's not different than before and can be fixed as a more
comprehensive overhaul of tracking change in the git module.
A fix for uri module regarding following redirects. The old behavior would follow redirects either way. This change clarifies the functionality and makes it a bit more explicit. Comparing the old behavior to the new 'yes' == 'all', 'no' == 'safe' and now 'no' will not follow any redirects. Historic behavior is still supported and documented with a push to the new values.
This doesn't account for boto configs where e.g. RDS has one
default region and EC2 another - all will default to `ec2_region_name`.
However, this is just handy to allow an easy site wide default
region if existing configuration already relies on it.
Modules can be improved to mention this in the documentation and
turn off required=True where needed. But it works with `ec2`
and `ec2_vol` without change.
Refactor the currently well-factored ec2 modules (i.e. those that already use ec2_connect) to
have a common argument spec. The idea is that new modules can use this spec without duplication
of code, and that new functionality can be added to the ec2 connection code (e.g. security
token argument)