TREX-Q: A query language based on XML Schema

Brad Penoff, CIS, The Ohio State
University, and Sun Microsystems Ltd.

Chris Brew, Linguistics, The Ohio State University

 
James Clark's TREX (http://www.thaiopensource.com) is a clean, simple
and powerful schema language for XML that is designed to complement a
datatyping language such as that described in XML Schema Part 2:
Datatypes (http://www.w3.org/TR/xmlschema-2/).  We have designed and
partially implemented a query language (called TREX-Q) based on TREX,
but borrowing some of the features of McKelvie's XMLQUERY
(http://www.ltg.ed.ac.uk/~dmck/xmlstuff/xmlquery/index.html). TREX has
spawned a closely-related successor: RELAX-NG
(http://www.oasis-open.org/committees/relax-ng/).  In this abstract we
sometimes use the term TREX to stand for both RELAX and TREX: though
our actual implementation work was based on TREX proper.


TREX is implemented, but the provided application is just a validator,
returning a single bit of information.  For corpus exploration mere
validation is insufficient: we also require the ability to return
structured results.  In other words, we require a query language. We
took several ideas from XMLQUERY.McKelvie's contribution is
threefold. Firstly,he provides a concise and usable syntax for
queries.  Secondly, XMLQUERY is the distillation of several years'
experience of implementing the things that linguists and dialogue
annotators who have been exposed to XML ask for next.  Finally,
McKelvie specifies a non-deterministic virtual machine for corpus
query which shows how to reconcile the language's power with efficient
execution over corpora too large to hold in memory, for which streamed
access is the only practical option.

Our implementation strategy has been to use the code for James Clark's
TREX validator (http://www.thaiopensource.com/trex/jtrex.html) as the
basis for extensions in the direction of XMLQUERY. The input to our
program is a standard TREX schema augmented with special elements that
control what is effectively a TREX-Q interpreter.  Other approaches
are possible. For example, Okajima's RELAX-NGCC
(http://homepage2.nifty.com/okajima/relaxngcc/index_en.htm) which is a
Yacc-like compiler from annotated RELAX-NG to efficient standalone
Java source code.

TREX is far from concise, so we do not aspire to reproduce the
syntactic neatness of XMLQUERY in TREX-Q. In actual use we anticipate
that many users will generate queries either by the use of a graphical
"query by example" interface or by using a more concise user-level
syntax.

At the workshop we will present the thinking behind our design and
demonstrate the current state of TREX-Q. Our hope is that TREX-Q
combines the elegance of McKelvie's query language design with the
clarity and precision of the TREX formalism.