--hosts=... semantics ?

Harlan bumped into another thing that has worried me a bit,
so let me ask for your thoughts.

When you do, say,

    ark package install foo--1.2.3

there is an implicit --hosts=ALL argument.  (This is our
much-loved "site-at-a-time" thinking: by default, you always
do something for the whole site.)

Now, when you're trying things out, experimenting, etc., you
often want to use a limited set of hosts; for example,

    ark package reveal --hosts=. password-stuff

just tries it on the host you're on.  This has the nice
property that you will trash just one host, not your whole
site :-)

The problem arises when a <constraint> forces an action that
must take place on some other host.  Let's say 'reveal' for
host foo depends on a 'compile' taking place on host bar.

If you've specified --hosts=foo, should the compile take
place on bar, or should it die?  (Now, it dies: you've got
to specify --hosts=foo,bar.)

The problem with

   ark package reveal --hosts=foo,bar blah-package

is that it will do the 'reveal' on bar as well, which wasn't
the idea.

I am inclined to have *two* options --hosts=... and
--constraint-hosts=...  --hosts=... would specify the hosts
where the specified action ('reveal' in the example just
past...) is to happen; --constraint-hosts=... would be other
hosts that could be used (only) to satisfy constraints.

Anyone with any better ideas?

Will