Re: "wait for" patch for gnome-session

[Tom Tromey <tromey cygnus com>]

> : I'd rather have the Gnome ssh-add GUI know to do this (and to set
> : restartanyway) automatically.

Adrian> Do you think having a checkbox in the capplet is more
Adrian> generically applicable, though? I may not have explained it
Adrian> very well, but WaitFor works for any program started by
Adrian> gnome-session... although right now I only have the GUI for
Adrian> manually started programs. I was just using ssh-add as an
Adrian> example.

I think there are two issues:

1. Having a waitfor property that does what you say
2. Exposing this property to the user via the control-center

The first is an implementation technique.  It probably makes sense as
a solution for the general problem of acquiring passwords at login
time.

The second is a UI issue.  I don't have a strong opinion.  Are there
other cases where this option would be useful?  If not, I'd say leave
it out and let it be up to the individual programs to request it.

What I mean by that last statement is that I want something like
gnome-ssh-askpass (a program I have installed here which seems to
almost do what I want) to set the restartanyway hint as well as the
waitfor hint.

Likewise if I had run a (hypothetical) gnome-kerberos-askpass program.

(Ideally the password-asking programs would have a cancel button or
something that could be used to disable the asking either for a single
session or permanently.)

> : My understanding is that right now you have to create a .Xclients that
> : says "exec ssh-agent gnome-session" in order to have this work.  I'd
> : much prefer to see this done via some more user-friendly mechanism,
> : like a checkbox in the control center.

Adrian> Yeah, but how? Just have control-center write out an
Adrian> .Xclients? What kind of interface are you imagining?

As far as user interface, maybe something like a "security" capplet
that has checks for enabling ssh and ssh-askpass, as well as kinit.  I
dunno.  Does PAM get involved here?  Does it know about "single sign
on" type things?

For the implementation, well, I don't know that either.  Writing
.Xclients would work I guess.  Maybe there are other plausible
approaches, like hacking gnome-session to know about ssh-agent.  In
the end this doesn't matter so much to me (since I'm not doing the
work :-) as long as it is simple for the user to set up and relatively
robust.

Hope this helps.  Are we still misunderstanding each other?

Tom