I am trying to improve the completion for the gradle command, and have run
into something I cannot explain. To simplify things as much as possible, I
have constructed this example file, calling it _myfoo:

#compdef myfoo

local curcontext="$curcontext" ret=1 state

_arguments -C \
    '--option1' \
    '--option2' \
    '*:a subcommand:->subcommand' \
    && ret=0

if [[ -n $state ]]; then
   print state: $state > /dev/pty2
fi

return ret

On the command line, I write: "myfoo --<tab>", and the completion kicks in
and gives me the options "--option1" and "--option2", which is what I
expected. However, on /dev/pty2 I see "state: subcommand", which I did NOT
expect. I would have thought that since only "--option1" or "--option2"
could possibly match after "--", state would not have been set*, *but it is.

The trouble with this is that in the gradle completion, the possible
subcommands are generated inside the if statement in this case, and thatis
time consuming. I would thus like to avoid going into it when the "--" on
the command line indicates that an option should be completed.

I would have thought that the call to _arguments did not set state since
found some matching options, but obviously I am using it incorrectly. What
should I do instead?