W3C home > Mailing lists > Public > public-qt-comments@w3.org > April 2004

[XQuery] IBM-XQ-028 NaN Comparison

From: Andrew Eisenberg <andrew.eisenberg@us.ibm.com>
Date: Fri, 9 Apr 2004 14:59:21 -0400
To: public-qt-comments@w3.org
Message-ID: <OFC8CE6F00.00C6DA31-ON85256E71.00680D47-85256E71.00684F9F@us.ibm.com>
I believe that "le" and "ge" return the wrong values when NaN values are 
provided for either of the two operands.

Consider the following query:

        let $x := 1.0e0
        let $nan := xs:double('NaN')
        return
                $x eq $nan,     (: will return false :)
                $x lt $nan,     (: will return false :)
                $x gt $nan,     (: will return false :)
                $x le $nan,     (: will return true :)
                $x ge $nan,     (: will return true :)
                $x ne $nan      (: will return true :)

I believe that "le" and "ge" should return false to be consistent with 
IEEE 754. They return true because:

1) The definition of op:numeric-less-than and  op:numeric-greater-than in 
F&O both contain the following:

        "If $arg1 or $arg2 is NaN, the function returns false."

2) The Binary Operator table in XQuery, B.2 Operator Mapping, contains the 
following:

        A ge B  numeric         numeric fn:not(op:numeric-less-than(A, B)) 
xs:boolean
        A le B  numeric         numeric fn:not(op:numeric-greater-than(A, 
B))     xs:boolean


To effect the change I suggest, change the that two rows of the Binary 
Operator table to the following:

        A ge B  numeric         numeric         op:numeric-greater-than(A, 
B) or op:numeric-equal(A,B)     xs:boolean
        A le B  numeric         numeric         op:numeric-less-than(A, B) 
or op:numeric-equal(A,B)        xs:boolean

                                                -- Andrew

--------------------
Andrew Eisenberg
IBM
4 Technology Park Drive
Westford, MA  01886

andrew.eisenberg@us.ibm.com
Phone: 978-399-5158    Fax: 978-399-7012
Received on Friday, 9 April 2004 14:59:55 UTC

This archive was generated by hypermail 2.3.1 : Wednesday, 7 January 2015 15:45:19 UTC