>>>>> "Cantor," == Cantor, Scott E <[log in to unmask]> writes:

    Cantor,> On 7/1/11 11:44 AM, "Maria Turk" <[log in to unmask]> wrote:
    >> 
    >> Sam H doesn't like the idea of presenting the user with a dialog to
    >> create the association.  He thinks it's a better idea to create the
    >> association automatically.  My issue with this solution is the use case
    >> where the user has many IDs for a given service and doesn't want to
    >> associate any ID with a service.  Basically always wants to be asked.

    Cantor,> I don't know if it's analagous, but the philosophy on IdP discovery on the
    Cantor,> web side now is to never automatically follow cookies to bypass the IdP
    Cantor,> selection process. Users find the lack of consistency (sometimes they get
    Cantor,> asked, sometimes not) more confusing than an extra click.

    Cantor,> That's in addition to the arguments about multiple identities.

A huge difference is that  you may  very well get 20-30 requests in a
    row to the same service.
I think using GSS without associations is incredibly impractical.

    Cantor,> Another consideration is whether it's always the case that a "failure" in
    Cantor,> the use of an identity at a service is going to be reflected in a manner
    Cantor,> that will be detected by this componentry so it can break the association.
    Cantor,> That may be true here; it's not true in the web case, which is another
    Cantor,> reason for not auto-following.

It is not always possible to detect.

My argument is roughly that you need to have some mechanism to put the
selector into always-reconfirm mode that is well understood.  Nothing
we're talking about here seems to avoid that requirement.  Once you have
something that your users understand to put things into a mode where it
is always reconfirming associations, it seems that you can be more
aggressive about making the associations.

I'll note that we have currently seen no proposals to convey a good UI
for enabling the always reconfirm mode.  I'd be very interested in a
credible approach that didn't need that.  However, while we've had
proposals for reducing the number of erronious associations, none that
get it low enough that I think it will be usable without the average
user understanding how to trigger such a mode.

--Sam