W3C home > Mailing lists > Public > xmlschema-dev@w3.org > September 2002

Re: How to reference attributes from list items

From: <noah_mendelsohn@us.ibm.com>
Date: Mon, 9 Sep 2002 11:47:01 -0400
To: "Eric Jain" <Eric.Jain@isb-sib.ch>
Cc: "Eddie Robertsson" <erobertsson@allette.com.au>, "xmlschema-dev" <xmlschema-dev@w3.org>
Message-ID: <OF59DA9F4F.2DE5AEF1-ON85256C2F.005594FB@lotus.com>

Just to give some perspective, among the reasons that the schema WG 
adopted a subset of XPath was that we (a) wanted to support streaming and 
(b) generally felt that having a small, predicatable set of capabilities 
for addressing keys would probably admit better optimizations and 
generally less implementation overhead in an already over-complex 
language.

Also:  I personally think that we probably made a mistake to do key/keyref 
at all in this round, or to do it in this way.   Some members of the 
database community told us early in the process of development that they 
had a strong requirement for an analog of primary-key/foreign-key, and we 
provided this facility to meet that need.  At the same time, we knew that 
there was a desire for more general co-occurrence constraints, but didn't 
find a way to express those constraints that generated concensus in the WG 
at the time.  Accordingly, we did key/keyref in 1.0 with an intention to 
revisit co-occurrence in another release.  I suspect we have the worst of 
both worlds:  my impression (only my impression) is that most database 
experts actually feel that you need primary-key/foreign-key across 
documents, not within a document.  At the same time we've intruduced a 
facility that feels temptingly close to giving you co-occurrence, but 
without meeting the general need, or (I think) being properly integrated 
into our type system.

Of course, I'm just giving my impressions of how we got here, not speaking 
officially for the workgroup.  Bottom line:  there were reasons for what 
was done, it was understood from the start that more general co-occurrence 
constraints would be need, but I'm not at all sure in retrospect that we 
did the right thing to do key/keyref first. 

------------------------------------------------------------------
Noah Mendelsohn                              Voice: 1-617-693-4036
IBM Corporation                                Fax: 1-617-693-8676
One Rogers Street
Cambridge, MA 02142
------------------------------------------------------------------







"Eric Jain" <Eric.Jain@isb-sib.ch>
Sent by: xmlschema-dev-request@w3.org
09/09/2002 07:59 AM

 
        To:     "Eddie Robertsson" <erobertsson@allette.com.au>
        cc:     "xmlschema-dev" <xmlschema-dev@w3.org>, (bcc: Noah 
Mendelsohn/Cambridge/IBM)
        Subject:        Re: How to reference attributes from list items



> Identity constraints and co-occurence constraints is on the
> table for a rework in coming versions of the language.

So it's only the path and nothing but the path you can use with keys...
Anyways, it's good to hear that some work is being done here. While this
isn't a critical feature and one can of course just include such a
constraint in the documentation, human or machine readable, it feels a bit
strange since the constraint is on a level you would naturally expect to
have expressed directly at the schema level.


--
Eric Jain
Received on Monday, 9 September 2002 11:49:40 GMT

This archive was generated by hypermail 2.2.0+W3C-0.50 : Tuesday, 11 January 2011 00:14:34 GMT