RE: datatyping is not needed

OK, I can buy that argument.

But I'm am a little worried about the cases where users are willing to pay
the cost of non-indexed searches on properties. Such as,  the scope is known
to be limited (a collection of few tens of resources) or the actual server
load is very low (say a small work group server).

Or how about a case where the "famous dead" are used filter most of the
records, but then a little further refinement happens using "obscure dead"
properties. SQL servers already do a process like this when dynamically
deciding how to process complex queries.

And I agree that data typing does open a particularly nasty can of worms.

The two questions to ask are... what percentage of queries/datatypes can be
covered by just defining say "string", "numeric" and "data/time"? what
percentage of servers/backends can easily test "obscure dead" properties
using those query/datatypes?

Marcus

> -----Original Message-----
> From:	Jim Davis [SMTP:jdavis@parc.xerox.com]
> Sent:	Monday, July 13, 1998 12:20
> To:	Marcus Jager; www-webdav-dasl@w3.org
> Subject:	RE: datatyping is not needed
> 
> At 11:33 AM 7/13/98 PDT, Marcus Jager wrote:
> >Hi,
> >
> >Ummm, Did I miss something or didn't we need the datatypes so the server
> >knows how to perform the relational operators (ge,le,... ) on the dead
> >properties.
> 
> You may have identified a bad assumption on my part.  I could be wrong,
> but
> here is what I was thinking when I asserted the lack of need.
> 
> It seems to me that there are two categories of dead properties, which I
> might call the "obscure dead" and the "famous dead".  The obscure dead are
> those the server knows nothing about, except that they are stored.  The
> famous dead are those that are indexed by the server.  It seems to me that
> to build an index, (e.g. a SQL table) the server must know at least a
> little abou the dead property, e.g. the datatype.  How it knows this is
> out
> of band, e.g. a server admin builds the table.
> 
> I was assuming that DASL implementations would only allow searching of the
> famous dead, because it's too expensive to search the obscure ones.  For
> the famous dead, the server already knows (and does not want the query to
> specify a different datatype than what it knows).  The obscure are not
> indexed and thus not searchable.
> 
> I could certainly imagine you arguing that this whole line of reasoning is
> invalid because it is too influenced by current implementations.  But I am
> strongly influenced by the design principle that one should try to
> standardize current practise.
>  
> A second argument against datatyping in the query itself is that to be
> meaningful, it would require DASL itself to define the semantics of
> comparisons among the various datatypes, and this seems like a large
> topic.
>  As it is, we not only leave the semantics undefined (is "10" less than
> "9"?  As strings in alphabetic order, yes, as integers, "no") we don't
> even
> provide a place to where the semantics would go.  Here, I think, doing
> less
> is a good idea.
> 
> I'd be grateful for counter arguments.
> 
>

Received on Monday, 13 July 1998 18:14:30 UTC