Re: zsh virsh completion

[Bart Schaefer <schaefer@brasslantern.com>]

On Jul 11,  2:52pm, Marko Myllynen wrote:
}
} - virsh help <command> output is pretty much like --help output of any
} other program, is there a trick to make the option descriptions also
} available when completing virsh command options?

I think you're asking about the call to _values in the cmdopts state
handler.  Each of the option specs (the words that you're caching in
_cache_virsh_cmdopts) can be in the form "-option[description]" just
like an option spec for _arguments.  So you would need to parse the
output of $(virsh help $cmd) a little more deeply than you presently
do to build those strings, and then do something a bit more clever
than ${=_cache_virsh_cmdopts} to cache and split them.

There is support in _arguments for parsing --help formatted output.
The basic pattern for having _arguments do this when "virsh --help"
doesn't produce that output, is to add a "command" style similar to:

  zstyle :completion::complete:virsh::options command 'virsh help'

The exact zstyle context to use depends on how you're making the call
to _arguments.  For your virsh completer you would need another call
to _arguments in the cmdopts state handler and corresponding change
to the context.  This can be tricky to get right.

} - would it make sense (or is it perhaps already possible somehow) to
} optionally enable "best guess" completion for all commands which have
} no command specific completion rules available yet? There are lots of
} commands for which completing just the options captured from --help
} output would already be hugely helpful.

I think you're looking for

    compdef -P '*' _gnu_generic

Which just says to try GNU-style options after all else fails, no
matter what the command word might be.  However, here's presently no
way to make sure this is called after other "compdef -P" patterns
that may be more specific.

(This leads me to wonder whether all those "#compdef -P" lines in
Completion/** files ought to instead be "#compdef -p".  I guess the
point is to allow the user to override with other compdefs and try
the supplied completions if none such is present?  But in that
case shouldn't _init_d also use -P rather than -p ?)