Re: [xsl] XSLT 3.0 and XML Schema version

Subject: Re: [xsl] XSLT 3.0 and XML Schema version
From: Michael Kay <mike@xxxxxxxxxxxx>
Date: Wed, 20 Nov 2013 11:16:00 +0000
On 20 Nov 2013, at 10:16, Sean B. Durkin <sean@xxxxxxxxxxxxxxxxx> wrote:

> If I feed the following stylesheet into an XSLT 3 processor, will the output
be implementation dependant?

No, it will be implementation-defined ;-)

The difference is that when things are implementation-defined, a conformant
processor has to document what it does.

Cross-dependencies between versions of specs are something that's typically
fixed only at a late stage of spec publication. However, it's very likely that
XSLT 3.0 will specify XPath 3.0, and XPath 3.0 is now a Proposed Rec and
therefore very unlikely to change, and XPath 3.0 says

* It is implementation-defined whether the type system is based on [XML Schema
1.0] or [XML Schema 1.1].

Date/time arithmetic is of course defined by XPath not by XSLT.

> If the XML processor under the XSLT processor respects XSD 1.0, then, by my
calculations, the output should be ...
> <aYearLater>0001-02-29</aYearLater>
> ... but if XSD 1.1, then it should be ...
>  <aYearLater>0000-03-01</aYearLater>
> On a side note, if you change the xsl:stylesheet/@version to 2.0 and feed
the stylesheet to Saxon HE 9.5, then you get output ...
>  <aYearLater>0001-03-01</aYearLater>

Saxon has a configuration-level switch allowing you to choose between XSD 1.0
and XSD 1.1 support. This is orthogonal to the choice of XSLT or XPath
version. If you choose XSD 1.1, then I think you will get 0000-03-01.

(For those unfamiliar with the detail, XSD 1.1 changed the definition of
non-positive year numbers to align with ISO 8601, which clarified that in the
"proleptic Gregorian calendar" there is a year 0. This differs from the
convention used by historians, where the year before 1AD is 1BC; but
historians do not generally use the proleptic Gregorian calendar anyway.)

One thing the specs don't really clarify is what format-date() should do with
negative years when the calendar is "AD". I think the right answer is probably
that -0001-01-01 should be output as 1 Jan 2BC.

Michael Kay

Current Thread