RE: [xml-dev] Concerned about the increasing reliance on XPath

[Liam R E Quin <liam@w3.org>]

On Sat, 2011-05-07 at 15:53 -0400, David Lee wrote:
[...]
> I was presuming that Roger was asking about non-XML DB or more standard
> XPath implementations.

I wouldn't go so far as to say "standard" but I think his use case may
primarily be the constraints in XSD.

> The *vast majority* of *full* XPath implementations that I am aware of
> require all data to be in memory.

Hmm, I specifically said XPath 2... I'm not so sure I have the same
experience as you, although if you count by deployed seats rather than
by implementations, it might be a different matter - libxml, and various
implementation used by Web browsers, are all XPath 1; Saxon has XPath 2
support and until recent streaming work, required the document to be in
memory.  On the other hand db2, Oracle, Microsoft sql server, qizx,
basex, dbxml, marklogic, eXistDB, sedna,... are all examples where the
entire document doesn't need to be in memory, although this is generally
transparent to the person writing queries.

> And when they are not many XPath statements cannot be implemented in
> streaming mode without some very advanced coding more typical in XML DB's ..
> in which case indexes take the place of memory.

Yes - but "in memory" and "streaming" are of course very different.

And yes, indeed, it's easy to write XPath expressions that can't be
streamed, just as you can write C code that can't easily be
optimized :-)  This was a big issue for XQuery development - there were
streaming implementations, and there's a tension between having
everything streamable and having a language human coders are willing to
use :)  This is why the xpath axes are optional in XQuery, even though
in the end even the streaming implementations did support them.

Liam

-- 
Liam Quin - XML Activity Lead, W3C, http://www.w3.org/People/Quin/
Pictures from old books: http://fromoldbooks.org/