Quote things, SC2268, SC2143, SC2181, SC2162, SC2016, SC2013, &c.
Rewrite globbing per `shellcheck`’s SC2013, and alsö s/typeset/local/g. Eliminate `compgen` where possible.
Alsö: use the existing utility functions `_bash-it-get-component-type-from-path` and `_bash-it-get-component-name-from-path`, which just use parameter substitution anyway. Why was `sed` here?
Alsö, don't add not-existing directories to `$PATH` in `pathmunge()`.
Finally, merge PR #1865 from NoahGorny...and clean it a bit...
Test was failing only on Mac OS X for some reason, so refactor a little. It turned out to be due to explicitly setting `$HOME` *after* loading the plugin.
For testing non-core files, just `load ../test_helper_libs` after `load ../test_helper` instead of trying to guess which libs must be loaded, in which order, for testing to succeed.
`_command_exists` spews log messages, so our test log is littered with `_log_debug not found`. This fixes that too.
Removed the copy/paste code, moved it into a reusable function that can
be called from the `local_setup` function as needed. Also simplified the
handling of the $HOME variable since @rico-chet provided a better way of
doing this.
Can't use the official v0.4.0 release, since our tests rely on a change
that was made after the release of v0.4.0.
This code currently uses an unversioned commit of the bats-core repo, we
will switch to a supported version once it has been released, e.g.
v0.4.1.