[help-GIFT] Re: Understanding MRML: undo

[Wolfgang Müller]

> As Wolfgang explained to me, the idea with this was to let the
> client track the session state and if the user requests an
> undo-step, the server just gets notified by a message
>
> <mrml>
>       <user-data command="history-backward" />
> </mrml>
>
> say.

Hmm. I did not express myself well (I think you understood what I wrote, but 
not what I think ;-) ). I guess what was rather thought of at the time to 
have the client maintain the sequence of queries. The history back would have 
meant for the client that it reposts the old query, and that the server is 
(if it wants to know about that) that there this query was issued because of 
history-backward. This corresponds a lot to what web browsers do.

> I wonder how this could work, since in general the session has
> a state on the server site either and this state has to be maintained
> by the server. Thus, I can imagine this to be useful only in a setting
> where the session on the server site is stateless.

Yes, and it's something that is (IMHO) rather algorithm-specific, so I do not 
think it should be configured server-wide, but rather algorithm-wide. q-p-l?? 
I think (and Ralf thinks that, too) we should make a table about the 
properties an algorithm has and in what q-p-l attributes we want to pack 
these values. Then maybe we should get a vote if and how to rename the 
query-paradigm-list such that the name reflects the thing. This probably goes 
a bit into the direction of Steph's thoughts on MRML evolution.

> To conclude: I propose to introduce a new elemenent
> "undo-query-step" to request an undo of the last query step from the
> server.

...together with the results of the previous step?

Any comments on that explicitly sollicited. If you do not have the time to 
deal with that currently, please tell this list, so Ralf can go ahead with 
what he thinks is OK.

Cheers,
Wolfgang