RE: [xml-dev] Disk-based XPath Processing

["Michael Kay" <mike@saxonica.com>]

> On a related note: I noticed a reference (from Saxon-SA page, 
> I think) that implied that w3c schema specifies a n xpath 
> subset (perhaps for defining how key constraints can be 
> defined using subset of xpath?), like one that would be 
> useful for streaming access.
> Is this true?

This subset is specified at
http://www.w3.org/TR/xmlschema-1/#c-selector-xpath

(Actually it's two slightly different subsets, one for selectors and one for
fields. The general form of a uniqueness constraint in XML Schema is "within
every element E, each S must have a unique value for F": for example, within
every location, each employee must have a unique value for employee-number.
The "selector" is used to select S from E, the "field" is used to select F
from S.) 

The subsets are woefully small, unfortunately.

> Or are there other formal subset specifications, to make it 
> easier to know exactly what subset is (guaranteed to be) 
> supported by an implementation?

There are lots of academic papers on streaming XPath that have implemented
the parts of XPath that the researchers knew how to implement and left out
the bits they didn't (such is the privilege of academics). They aren't all
very good at revealing which bits they left out, but I don't think you'll
find one that can do preceding-sibling::*[3] or
following-sibling::*[last()-1].

Michael Kay
http://www.saxonica.com/