Yes, we can use profiles to define "minimum behaviour" within specific
application areas. But lets prevent the need for profiles as much as
possible when we can by improving the protocol.

We do not disagree on the fact that we will never be able to send the
same string to an arbitrary list of servers and get back the same
result. But we we could define a specific "optional or minimum
behaviour" in addition to a diagnostic to make things easier.

To give an example just to illustrate my way of thinking:
1) client searches for any=mercedes
2) server A return x hits (conventional), server B returns a diagnostic
saying not supprting "any" and server C returns additionally a list of
access points K,L,M with for each acces point the number of hits (assume
that this is part of SRU/SRW and the server's creaitivity)
3) client X will say to the user "A gives x hits, B gives a diagnistic
and C gives a dagnostic (neglecting the optional addition) but
client Y will say to the user "A gives x hits, B gives a diagnistic and
C says that fields K, L, M would return x, y, z hits and provides the
user with a clickable link to initiate such a search.

What would be wrong if we had specified in SRU/SRW such an optional
list of access points and allowing client Y behaviour as well as client
X behaviour?  There are only a few types of optional responses that are
meaningful, so we do not have to extrapolate this to nonsense
situations.

The alternative is that we send different queries to different servers
and inform the user on those differences. I expect most gateways
currently solve this problem by having some kind of knowledge database
in which they find the best combinations of queries for specific access
points but they do not inform the users about those differences to keep
interfaces simple. I don't believe this is the preferred way, but I
agree with Ralph that currently there is no alternative for
metaseachers.

I will keep silent now on this issue.

Theo


>>> [log in to unmask] 8/11/03 3:13:27 nm >>>
> -----Original Message-----
> From: Theo van Veen [mailto:[log in to unmask]]
> Sent: Sunday, August 10, 2003 6:17 PM
>
> But isn't that just what we are trying to solve? I get the
> feeling that we have different objectives.

I don't think it is a solvable problem.  We can do things to make it
easier.
One of those things is profiles.  But, you will never be able to send
the
same request to an arbitrary list of servers and get back useful
results
from all of them.

I agree with whoever it was that pointed out that if you have enough
influence among your database providers to get them to implement these
ideas, then you have enough influence to get them to adhere to a
single
profile of behavior and you won't need the implementation changes.

Ralph