My rather late opinion on this subject (I had to catch up with one
month's mail):

I agree with Xpath as an option for specifying an SRU/W response, but I
would like to consider the following also.

- In general we have to do with 1) metadata describing objects, 2)
captions derived from these metadata records and 3) the actual objects
being full text or other media types or services. In many cases the
metadata will only point to the actual objects rather then being
contained in the metadata.
- The server may or may not be able to deal with the Xpath request
- The server may or may not be able to make a distinction between full
records or captions.
- I expect the distinction between captions and complete records to be
already available in most implementations of indexing software.

I would like to make a distinction between record captions and
responses that are specified by Xpath. In the first case it is up to the
server what is contained in the record caption (I do expect it to be
available with a default and meaningful contents). In the second case
the server will need to retrieve the complete record to run an xslt
transformation. The response contains a recordSchema but in case the
response is specified by by xPath the returned recordSchema does not
have to have a name or identifier.

Lets consider the following situations:
1) The client may want captions only, just to show to the user a list
of short records without the danger of getting back 40 Mb records.
2) The client wants specific fields for specific purposes and sends an
Xpath spec (or a reference to it)
3) The client wants just the whole record for a full display or to
provide full functionality.

In case the client is not able to meet request 2) and 3) I would prefer
it to respond with captions and I would prefer the recordSchema of these
captions to be DCX, containing dc:title and other fields that are
considered to be relevant for users to select records. Additionally it
would make sense to have an indication in the recordSchema  for captions
so the client will know that there also is a full record.

Theo


>>> [log in to unmask] 7/24/03 2:14:44 nm >>>
> Date: Thu, 24 Jul 2003 12:18:36 +0100
> From: "Matthew J. Dovey" <[log in to unmask]>
>
> General consensus seems to be for Xpath so I'll withdraw my
> objections.

Cool.

> But...
>
> In general (and can we leave it mute whether Xpath is an example or
> not) I'd still like to follow the principle of adding stuff to SRW
> to solve existing requirements rather than following the Z39.05
> model of producing a highly versatile spec which can address new
> requirements before they are formed but which doesn't get used since
> it is too complex!

I am sure _everyone_ would agree that this is a good general
principle.  I would, anyway.

BTW., as a point of historical interest, I don't think much, if any,
of Z39.50's perceived complexity went in speculatively.  Most features
went in as a result of someone's real need; the problem is that there
was less take-up of the more esoteric features than anticipated.  I'd
guess that _someone_ uses _every_ part of the spec ... but probably
no one implementation uses more than 50% of it.

 _/|_
_______________________________________________________________
/o ) \/  Mike Taylor  <[log in to unmask]>
http://www.miketaylor.org.uk
)_v__/\  "Leopards break into the temple and drink the sacrifical
         chalices dry; this occurs repeatedly, again and again:
         finally it can be reckoned upon beforehand and becomes a
         part of the ceremony" -- Franz Kafka.

--
Listen to my wife's new CD of kids' music, _Child's Play_, at
        http://www.pipedreaming.org.uk/childsplay/