Re: [xsl] General rule for designing XPath expressions to return items in document order?

Subject: Re: [xsl] General rule for designing XPath expressions to return items in document order?
From: Michael Kay <mike@xxxxxxxxxxxx>
Date: Fri, 10 Jan 2014 12:01:23 +0000
On 10 Jan 2014, at 10:56, Costello, Roger L. <costello@xxxxxxxxx> wrote:

> David wrote:
>
> 	I haven't checked the streaming rules in detail
> 	but I would expect //head to be streamable.
> 	(You might not be able to access any of the
> 	child nodes in a streamable way but for example
> 	count(//head) ought to be able to count all the
> 	head in the document in a single pass.
>
> Michael responded:
>
> 	That's a fair summary. //head is "consuming"
> 	(it reads the input stream), and "crawling"
> 	(it accesses all the nodes in the subtree). When
> 	you get an expression that is consuming and
> 	crawling, you are allowed to do "inspection"
> 	operations on the result, for example count()
>
> But Michael doesn't that contradict section 19.1 of the XSLT 3.0
specification:
>
> 	For example <xsl:value-of select="//head"/> will
> 	still fail the streamability tests, because of the
> 	possibility that one head element is a child of
> 	another. This problem can be remedied by
> 	writing <xsl:value-of select="//head/text()"/>.
>
>
No, there's no contradiction:

(1) <xsl:value-of select="//head"/> isn't allowed for the same reasons that
//head/data() isn't allowed: xsl:value-of is an "absorption" operation.

(2) The fact that <xsl:value-of select="//head/text()"/> is allowed doesn't
contradict anything I said. It's allowed because

a) Under section 19.1, it gets turned into //text()[parent::head] or
equivalent

b) Unlike //head, //text() is "striding" - the nodes it selects are
non-overlapping (because text nodes never overlap). So absorption operations
like data(), string(), and <xsl:value-of> are allowed.

I suspect you're making the mistake of thinking that //head/text() is very
similar to //head/string(). But that's only true when every <head> element has
a single text node child.

Michael Kay
Saxonica

Current Thread