The user may need the help to solve any fatal error that appears while todo.sh isn't properly set up. As the help actions do not depend on any setting that the fatal errors check, we can still invoke them.
Factor out dieWithHelp() and use that for printing the fatal errors.
POSIX sed doesn't support regex branches; I don't want to switch to modern regexp (via -E), since that makes more characters special. Rather, expand the alternatives into two sed commands.
Need to make match of preceding priority optional.
Also tighten patterns for priority and done marker by anchoring them to the front, as the leading task number has already been substituted away.
When the command-under-test fails with a non-zero exit code, neither the different exit code nor any output differences are printed. Also, there's a lot of duplication between test_expect_output and test_expect_code_and_output. So, let's fix both by delegating the first to the latter.
Usually (i.e. when a pager is available), the help() output is piped through the pager, and then the return code will be the one of the pager. So the exit 1 at the end of help() doesn't make much sense, and should be abolished for consistency.
Currently, the only way to get usage help for a custom action is to use "todo.sh help" and scroll / search for the action name. (Or try to call the action without / with invalid parameters to hopefully get a one-line syntax summary.)
This extends the help command to print the usage of optionally passed custom action names.
For ls / _list(), custom add-ons can use post_filter_command for additional filtering before the highlighting is applied. Let's allow that for listall and listpri, too, although they employ post_filter_command themselves.
This applies e24777fa2c to todo.cfg, too, and adds a verification to the existing test.
Note: Because todo.cfg is preprocessed by test-lib, we use the original config from the working copy, not the copy inside the test fixture.
An empty line is used to delimit test commands in test_todo_session. (This wasn't properly reflected in the usage comment, fixed that, too.) This prevents us from testing commands that include empty lines. (Currently, there is no such output in todo.sh, but custom add-ons may have this.)
Instead of forcing the test to filter out the empty lines, extend the syntax to allow empty lines in the expected output by escaping them with a single backslash. (When redirecting via <<EOF, the backslash itself must be escaped = doubled; better use <<'EOF'.)
Currently, todo.sh produces no indented output, so this hasn't affected tests so far. I found this while testing one of my custom actions with the test-lib.
By unsetting IFS, the read command won't strip off leading and trailing whitespace while parsing the test session definition.
The todo_completion had the problem that it didn't consider the todo.cfg itself for the location of TODO_ACTIONS_DIR, it just tried the default location or a globally exported config value. With the injection of custom configuration now in place, we can actually delegate the listing to todo.sh itself.
The added built-in "listaddons" command is used for that; it may also be helpful for troubleshooting or to find out about available add-ons. (But the help / shorthelp commands offer more information.)
Additionally, completion is now more precise; only executable actions and no subdirs are listed now; this is also covered by the enhanced test.
The last bonus: The custom add-on actions are now only determined when there's actually completion of commands.
