CI: Running tests removes the IRC package, and other local-install corruptions

[richteer]

All tests should be running in a totally different namespace, and NOT using the packages directory as part of the cloned repo. This is a severe bug, as it messes with the typical local Halibot development environment.

Furthermore, we shouldn't test fetch as part of the cli tests, since that depends on a repo server being accessible -- a failure on the side of the repo server SHOULD NOT lead to a test failure.

[richteer]

@sjrct Tagging you in this, since I think you are the one that wrote these test cases. Suggestions on how to proceed?

[richteer]

From main.py's implementation of unfetch:

        for pkgpath in bot.config["package-path"]:
            path = os.path.join(pkgpath, name)
            if os.path.exists(path):
                shutil.rmtree(path)
                print("Removed '{}' installed to '{}'.".format(name, path))
                success = True

We delete packages from all applicable package-paths. This should most certainly not be intended.

[richteer]

Opened #122 to track this portion of the bug.

Leaving this one open since there's still work to be fixed here: CLI tests should be totally self-contained, and not affect packages outside of its path (therefore, we should probably force package-path to be local-only), and should not depend on remote resources to pass tests.

I'd rather see coverage decrease in main.py than have a test that requires a remote server to pass.

[sjrct]

The solution for tests I think would be to limit the package-path of the test all configs to only point to those for the local install.

[sjrct]

I'll work on this.