There's a bug in the HelpFormatter.add_argument method.  It does not take into account the extra indentation of subactions (sub parsers) when calculating self._action_max_length.

The corrected method and a test case is in the attached file.

class MyFormatter(argparse.HelpFormatter):
    """
    Corrected _max_action_length for the indenting of subactions
    """
    def add_argument(self, action):
        if action.help is not argparse.SUPPRESS:

            # find all invocations
            get_invocation = self._format_action_invocation
            invocations = [get_invocation(action)]
            current_indent = self._current_indent
            for subaction in self._iter_indented_subactions(action):
                # compensate for the indent that will be added
                indent_chg = self._current_indent - current_indent
                added_indent = 'x'*indent_chg
                invocations.append(added_indent + get_invocation( subaction) )
            # print('inv', invocations)

            # update the maximum item length
            invocation_length = max([len(s) for s in invocations])
            action_length = invocation_length + self._current_indent
            self._action_max_length = max(self._action_max_length,
                                          action_length)

            # add the item to the list
            self._add_item(self._format_action, [action])

Without this correction self._action_max_length is off by 2 if a subparser has the longest invocation.  Normally that wouldn't be an issue since normally subparser names are short and easy to use.  But with the 'aliases' that newer argparse allows, the invocation could be longer, and possibly longer than regular arguments.

I haven't passed this correction through test_argparse.py.  It also would need a test case or two.  I don't think it need a documentation change.

I don't know if this correction will work with the subparser grouping proposed in http://bugs.python.org/issue9341.