[xquery-talk] XQuery and Web 2.0

John D. Mitchell jdmitchell at gmail.com
Thu Apr 24 17:38:43 PDT 2008


On Thu, Apr 24, 2008 at 3:58 AM, John Snelson <john.snelson at oracle.com> wrote:
[...]
>  Simple and complete REST support is important, I agree.

Are you saying that should be in the XQuery language itself or in the
standard libraries?  Seems overkill for the former but I'd love to see
it in the latter.

Of course, there's lots of things I'd like to see fixed/added to the
standard libraries -- don't even get me started on the brain damaged
date/time stuff. :-)

>  However, I think XQuery is already way ahead of the map-reduce phenomenon.

Indeed, they are orthogonal.

> It seems to me that since XQuery is declarative the compiler ought to be
> able to work out how and where to parallelize your program, rendering
> explicit parallel algorithms obsolete.

Well, in terms of the XQuery fundamentals of doing queries against the
underlying data store, there are existing implementations that do that
quite nicely (but, ObDisclosure, I work for Mark Logic so I'm probably
biased).  However, that parallelism is a quite restricted notion of
automatic parallelism in that it is implicit and hidden inside the
implementation.

In terms of XQuery in competition with things like JavaScript, there's
nothing per se in XQuery that makes it particularly suitable as a
general purpose programming language -- it's geared for what it's good
at and that's just fine, IMHO.  It certainly can expand out of it's
current, niche usage but given decisions of the committee in the past,
I'm not sure how far it will go in that.

In terms of the general sentiment that functional languages obviate
the need for explicit parallelism... Well that's akin to the beliefs
that non-trivial distributed systems can be created that completely
hide their distributed nature (e.g., the realities of the underlying
network).  In general, automatic parallelism seem to be able to get up
into the 4-8 way parallel range and that's it for all but the
embarasingly parallel (that also fit the particular parallelizer's
model).

[...]
>  I think what your saying is that XQuery needs a killer app - and I agree
> with that.

I'm way biased but check out http://markmail.org/.  Pure XQuery
backend with the latest Ajax hotness for a UI.

Take care,
John

P.S. These are, of course, just my personal opinions. YMMV.


More information about the talk mailing list