Betr.: Retrieving XML

The subject gradually changed from searching XML to retrieving XML.
 
As far as it concerns retrieving XML why not do it the SRU/SRW way. There we also will have a limited number of recordSchemas and it would be preferable to keep the Z39.50-XML record schemas identical to the SRU/SRW record schemas. 

As far as it concerns sending stylesheets: when a client gets the XML it can do the XSLT translation itself. I do not see the need of sending the stylesheet. We have a SRU client that is a complete portal for distributed searching written in a few lines of XSL with some Javascript. It allows you to specify the stylesheet locally. With our SRU server you can specify the xsl as URI as parameter and do a local transformation with stylesheet from somewhere else.
  
As far as it concerns searching. We have found that it is extremely powerful to index all end-nodes of XML trees in records. Each word is indexed twice: once as  "any" and once with the tag-name of the end-node as access point. So regardless of the amount of nesting of tags <author>van veen</author> is always found as author. By this we loose information on the context and can not use XPATH for searching but I have a strong feeling that this is more an advantage than a disadvantage. We can however when records are retrieved.

Theo

>>> Sebastian Hammer <quinn@indexdata.dk> 22-04-02 13:49 >>>
At 12:31 22-04-2002 +0100, Robert Sanderson wrote:

> > Yes, this is a good notion. Actually, given a known initial schema, it
> > makes sense for the client to submit an XSLT stylesheet to request a
> > transformation of each result set record prior to shipment. In my mind,
> > this is exactly equivalent to the use of Espec-1, only more powerful, and
> > standardised to boot.
>
>Can't we do that now by giving a URI in elementSetNames which refers to an
>XSLT style sheet, rather than dumping potentially very long stylesheets
>dynamically at the server for every request?  For retrieval purposes,
>specifying a style sheet and specifying a schema for the record to be
>returned in result in the same outcome -- you get the record in a certain
>format.

Giving the URI as opposed to shipping the StyleSheet is arguably a 
simplification, although it does force the Z39.50 server to act as a HTTP 
client, and it requires the stylesheet to be available somewhere public.

But again, it makes no sense to give the server a stylesheet unless you 
also have a way of telling the server what schema you expect the data to be 
in *at the beginning* of the transformation.

You gotta say: "I would like this data in Schema X and *THEN* apply 
stylesheet Y to it. Otherwise, you're always relying on the server to make 
data available in one specific schema."

I think we need CompSpec for this... that allows you to give the schema. 
Then we just need a convention for using a stylesheet as an alternative to 
Espec-1.

--Sebastian
--
Sebastian Hammer, Index Data <http://www.indexdata.dk/>
Ph: +45 3341 0100, Fax: +45 3341 0101

Received on Monday, 22 April 2002 08:39:26 UTC