lint silently ignores content outside current directory even when passed as argument

[cidrblock]

Given the following dir structure

.
├── other_dir
└── stuff
    └── lint
        ├── borked
        │   ├── playbook.yml
        │   └── roles
        │       ├── uh-huh
        │       │   └── tasks
        │       │       └── main.yml
        │       └── yep
        │           └── tasks
        │               └── main.yml
        └── no_errors
            ├── playbook.yml
            └── roles
                └── no-errors-here
                    └── tasks
                        └── main.yml

Running ansible-lint from other_dir reports no errors

other_dir pwd && ansible-lint /home/bthornto/github/lint_bug/stuff 
/home/bthornto/github/lint_bug/other_dir

Passed with production profile: 0 failure(s), 0 warning(s) on 0 files.

Running from /tmp reports no errors

(venv) m910x ➜  /tmp pwd && ansible-lint /home/bthornto/github/lint_bug/stuff
/tmp

Passed with production profile: 0 failure(s), 0 warning(s) on 0 files.

Running from $HOME is fine (in a parent of the dest to linted)

pwd && ansible-lint /home/bthornto/github/lint_bug/stuff
/home/bthornto
WARNING  Listing 12 violation(s) that are fatal

Tested with volume mounts in creator ee as well

[ssbarnea]

While it looks like a bug, it appears to be just a side effect of current logic.

In absence of a config file or git repository, which do defined the location of the project_dir, the linter will assume CWD as being the current project.

This means that if your are inside user home and try to lint a file from outside like /tmp the linter will not do anything as it will skip the file.

The logic about ignoring any files outside the current project makes sense as we never wanted to lint code that the user cannot fix.

Still passing a single folder to linter might make people think that this should only chdir to it and run from there.

What if user is passing more than one argument, ansible-lint /tmp/foo /var/foo? We cannot have more than one project_dir.

As a safety measure, we should probably make the linter return an error code if it did finish without linting any files (everything was skipped). At least it should prevent accidents. Still, this will not fix this issue.

I could add some code that would detect if only one argument was given that is a folder outside current directory and assume that this is a project_dir and chdir to it. I should mention that we already have a --project-dir argument in linter, but almost nobody is using it. For example the current bug can be avoided if linter would be called with ansible-lint --project-dir=/tmp/foo /tmp/foo/playbook.yml.

Any suggestions?