[XSL-LIST Mailing List Archive Home] [By Thread] [By Date]

RE: [xsl] XPath proximity position in predicates

---

---

I think it would certainly have been more natural if the spec had specified
rounding in this case, as it does for example with the arguments to
substring(). But it doesn't, and it's not a significant problem in practice.
In the first draft of XPath 2.0 rounding slipped in somehow (see
http://www.w3.org/TR/2001/WD-xpath20-20011220/#id-predicates) but we took it
out again to retain compatibility.

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

> -----Original Message-----
> From: David Landwehr [mailto:david.landwehr@xxxxxxxxxxxx] 
> Sent: 15 May 2006 12:05
> To: xsl-list@xxxxxxxxxxxxxxxxxxxxxx
> Subject: Re: [xsl] XPath proximity position in predicates
> 
> Hi Ken,
> 
> Notice that your situation is different from what I have 
> since you use the = operator. From a specification stand 
> point this would never change as the result of the test is a 
> boolean. The only situation where this "rounding" would apply 
> would be in the special case where you have a predicate 
> expression returning a number. I guess the specification 
> would be in its right to say it default should use round 
> (e.g. as it does for the arguments to the substring function) 
> it would still allow the author to change the behavior by 
> wrapping the calculation in a call to floor or ceiling. 
> However I'm not trying to get a change into the 
> specification, I simply found it interesting that it work in 
> this way and thought way not ask the people who know why ;)
> 
> Best regards,
> David
> 
> 
> G. Ken Holman wrote:
> > At 2006-05-15 12:30 +0200, David Landwehr wrote:
> >> It might be that a default of rounding the number expression isn't 
> >> always the solution an author would want. The reason I 
> took a guess 
> >> for rounding is that if you write select="element[(((7 div 
> 5) div 3)
> >> * 5) * 3]", you get from the constant expression 
> >> select="element[6.99999999999]" which will never select an element.
> >> In that particular case rounding would be what I expected.
> >
> > Yes, I understood that, and it seems intuitive but from a standards 
> > perspective the question would be would you want that to *always* 
> > happen, or should it be up to the user to decide what to do with 
> > inaccurate results as it might depend on the situation (as 
> I tried to 
> > highlight in my example)?
> >
> > Thanks for following up.
> >
> > . . . . . . . . . Ken
> >
> >> Best regards,
> >> David
> >>
> >>
> >> G. Ken Holman wrote:
> >>> At 2006-05-15 11:19 +0200, David Landwehr wrote:
> >>>> Reading XPath 1.0 it states that a predicate evaluating 
> to a number 
> >>>> will return true if equal to the proximity position of 
> the current 
> >>>> node. I was wondering if there is a reason the evaluated number 
> >>>> isn't rounded by the XPath engine? Because XPath is based on IEEE
> >>>> 754 the result of an evaluation might be inaccurate on the last 
> >>>> digit which will cause a predicate to return false where 
> it should 
> >>>> return true. I'm just asking this out of curiosity and 
> accepts that 
> >>>> an author has to call the round function if she uses arithmetic 
> >>>> which can give inaccurate results.
> >>>
> >>> As the typical use is merely ordinal position in a node set I've 
> >>> never had to worry about this in a predicate ... but I 
> did have to 
> >>> think about position() and rounding in a standalone <xsl:if> when 
> >>> doing a two-column display in XSL-FO and I wanted to introduce a 
> >>> column break in a standalone block ... I might have wanted to use
> >>> floor() instead of round() in the following:
> >>>
> >>>  <xsl:if test="position()=round(last() div 2)">
> >>>    <block break-before="column"/>
> >>>  </xsl:if>
> >>>
> >>> ... but I cannot readily extrapolate that into an XPath predicate.
> >>>
> >>> I think you summarized correctly that if the author is 
> doing things 
> >>> that might give inaccurate results they should do it 
> explicitly ...
> >>> as in my case they might want to make the decision 
> between floor() 
> >>> and ceiling() but in a predicate rather than just round(), so I 
> >>> don't believe it makes sense to just implement round() by default.
> >>>
> >>> I hope this helps.
> >>>
> >>> . . . . . . . . . Ken
> >
> >
> > --
> > Registration open for XSLT/XSL-FO training: Wash.,DC 2006-06-12/16
> > Also for XSLT/XSL-FO training:    Minneapolis, MN 2006-07-31/08-04
> > Also for XML/XSLT/XSL-FO training:Birmingham,England 2006-05-22/25 
> > World-wide on-site corporate, govt. & user group XML/XSL training.
> > G. Ken Holman                 mailto:gkholman@xxxxxxxxxxxxxxxxxxxx
> > Crane Softwrights Ltd.          http://www.CraneSoftwrights.com/s/
> > Box 266, Kars, Ontario CANADA K0A-2E0    +1(613)489-0999 (F:-0995)
> > Male Cancer Awareness Aug'05  http://www.CraneSoftwrights.com/s/bc
> > Legal business disclaimers:  http://www.CraneSoftwrights.com/legal
> >
> 
> 
> --
> --------------------------------------------
> David Landwehr (david.landwehr@xxxxxxxxxxxx) Chief Executive 
> Officer, SolidApp
> Web: http://www.solidapp.com
> Office: +45 48268212
> Mobile: +45 24275518
> --------------------------------------------

---