We also ran into this (ansible not loading variables in /path/to/inventory/group_vars/foo/main.yml) and had to revert to 1.9.4. This sort of thing should really be part of ansible's release testing.

Agreed, this is an absolute showstopper. Frankly I'm a little shocked that this kind of thing can get through unnoticed, it does not say good things about the release testing process. Also disappointing to see no activity on an issue like this for 17 hours.

Same on 2.0.1.0 : #14945.

@nikut : you say 17h. Try more than 10 days ...

Sad. It has blocked the development of my team deployment tasks.

As a temporary workaround, if you are able to reference the hosts file directly instead of referencing the directory that it is inside of, then it will work. Figuring out what the difference between those two cases is now.

$ ansible-playbook 15093.yml -i inventory/hosts       (10:45:34)

PLAY [localhost] ************************************************************** 

TASK: [debug ] **************************************************************** 
ok: [localhost] => {
    "var": {
        "hostvars['localhost']": {
            "group_names": [
                "test"
            ],
            "inventory_hostname": "localhost",
            "inventory_hostname_short": "localhost",
            "test_var_all": "This is defined in inventory/group_vars/all",
            "test_var_in_hosts_file": "defined in hosts file",
            "test_var_localhost": "This is defined in inventory/host_vars/localhost",
            "test_var_test": "This is defined in inventory/group_vars/test"
        }
    }
}

PLAY RECAP ******************************************************************** 
localhost                  : ok=1    changed=0    unreachable=0    failed=0

git bisected to this commit: ab18fd2

EDIT: pasted the wrong hash. This is the problem commit: e856ac2

@jbouzekri I just checked and this bug is not present on 2.0.1.0 or devel. So looks like these are separate bugs.

@abadger : but look at the referenced bug report #14945. It seems it is almost the same issue no ?

After comparing current code in stable-1.9 to code in devel I think that commit e856ac2 is wrong. The limit bug that was intended to fix is because something is not looking at runner.basedir (the equivalent of 2.x's vars['playbook_dir']). e856ac2 is modifying the basedir found for inventory rather than the one found for playbooks so it is doing the wrong thing. I'm going to revert that now in stable-1.9 and then look for where the runner.basedir is being omitted when limit is given.

Okay, this should be fixed by the revert in 81c4817 and that problem with the --limit should be fixed by #13556. So I'm closing this bug as fixed.

@abadger Confirmed that it works now in stable-1.9, both with my test case and with our production inventory. Thank you!

Hope there will be a test case for the "inventory is directory" case from now on. Will you be making a bugfix release considering the showstopper nature of this bug?

@nikut yes, there will be a 1.9.6... We're looking at one more bug on the 1.9 release and then we'll be putting out a release candidate.

If you'd like to make a pull request for an integration test of inventory is directory, that would be a great way for us to expand test coverage of this. Integration tests are ansible playbooks that use the assert module to show that the test conditions are met. They live here: https://github.com/ansible/ansible/tree/devel/test/integration