> Consider the following code:
> 
> <xsl:variable name="foo" select="nothing" as="xs:string?"/>
> 	
> <xsl:choose>
> 	<xsl:when test="$foo != ''">A</xsl:when>
> 	<xsl:when test="$foo = ''">B</xsl:when>
> 	<xsl:when test="not($foo != '')">C</xsl:when> </xsl:choose>
> 
> When there isn't a <nothing> element, the output is C.  That is:
> 
> $foo != '' is false
> 
> and
> 
> $foo = '' also is false 
> 
> Which is strange.  If I do "$foo is empty" then Saxon tells 
> me $foo is a string and not a nodeset.

The XPath 2.0 type system is a generalization of the 1.0 system. In 1.0,
strings, numbers and booleans could exist only in the singular, whereas
nodes could exist only in the form of a node-set (a single node was treated
as a set with one member). In 2.0, everything is a sequence. So the XPath
1.0 rule for a node-set $n that ($n = '') is true if there is at least one
node in $n whose string-value is $n, has been generalized so that for a
sequence of strings $s, ($s = '') is true if any string in $s.

$foo is a sequence of strings, which may contain either no strings or one
string. ($foo = '') is therefore true if $foo contains a string and that
string is equal to ''. ($foo != '') is trus if $foo contains a string and
that string is not equal to ''. 

It might be strange, but it's a logical extension of the XPath 1.0 rules.
And as it happens the results correspond very closely (though not exactly)
to the way null values work in SQL: in SQL, if a column foo of a table row
is null, then neither (SELECT ... WHERE foo='') nor (SELECT ... WHERE foo !=
'') will select that row.

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