This is an archived snapshot of W3C's public bugzilla bug tracker, decommissioned in April 2019. Please see the home page for more details.
http://www.w3.org/TR/xquery-operators/ says that it has IEEE float/double semantics as opposed to xml schema. i'm not clear on what the ordering/collation is of xs:float(0.0) and xs:float(-0.0) and how min()/max() treat xs:float(0.0) and xs:float(-0.0) (not sure based on the doc and not sure if this is specified in IEEE - there is no link to the IEEE standard provided). given that negative zero and positive zero seem to designate different numbers, one would expect that min({xs:float(0.0), xs:float(-0.0)}) => -0.0, but this doesn't appear to have been stated and thus one could argue that because xs:float(0.0) eq xs:float(-0.0), it is perfectly valid for min({xs:float(0.0), xs:float(-0.0)}) to return either -0.0 or 0.0 (or worse still, just return one or the other arbitrarily, as a typical impl will do by taking the first and comparing it with the others). it seems like sorting and min()/max() should be explicitly specified or clarified to remove any ambiguity.
There is no link given for [IEEE 754-1985] in our spec since it is not available online. If you review the material at http://grouper.ieee.org/groups/754/ you will see that the only pointer there is a mechanism for buying a printed copy. But the ieee.org page does provide several links to publically available material that describes both the IEEE 754 and current implementations. Hopefully you may find one or more of these useful. /paulc
The following are my attempts to answer your two questions. 1. min({xs:float(0.0), xs:float(-0.0)}) Section 6.3.1 op:numeric-equal states: "For xs:float and xs:double values, positive zero and negative zero compare equal." Note as well that this function is used to implement the operators eq, le, ge, and ne. Section 15.4.4 fn:min states that the following code is used to implement the function (after certain other cases are covered): "Otherwise, the result of the function is the result of the expression: if (every $v in $c satisfies $c[1] le $v) then $c[1] else fn:min(fn:subsequence($c, 2)) " Since fn:min()'s semantics depend on the "le" operator which is defined by the op:numeric-equal() function, this means that negative zero and positive zero (as floats or doubles) compare equal. I personally agree that the F&O spec could be clarified by giving the answer for the example: min({xs:float(0.0), xs:float(-0.0)}) My personal guess is that we will have to permit either postive zero or negative zero to be returned but the WGs will have to decide this. 2. ordering/collation is of xs:float(0.0) and xs:float(-0.0) To answer your sorting question you have to refer to XQuery/XSLT since the F&O spec itself does not define sorting. But Section 3.8.3 Order By and Return Clauses in XQuery 1.0 does sorting using the "gt" operator so positive zero and negative zero would again be treated as the same. This section then goes on to define what happens for equal values: "If neither V1 nor V2 is greater-than the other for any pair of orderspecs for tuples T1 and T2, then: - If stable is specified, the original order of T1 and T2 is preserved in the tuple stream. - If stable is not specified, the order of T1 and T2 in the tuple stream is implementation-dependent." I think this clearly defines how ordering of negative and positive zero is handled in XQuery. I assume it is as clear in XSLT 2.0. /paulc
ok, that makes sense and is consistent. thanks for the clarification. in a way the behavior is odd but once you go down the path that distinct values with distinct (albeit barely discernible) behaviors are equal as per IEEE things can't help but get a bit wierd. in any event, a clarification in the spec would be nice.
Just one extra observation: for sorting, -0 and +0 compare equal, and like all other equal sort keys, this can be resolved by adding a second sort key. So (using XSLT notation), <xsl:sort select="$n"/> <xsl:sort select="1 div $n"/> will cause -0 to be sorted before +0. Michael Kay
Added example: fn:min(xs:float(0.0E0), xs:float(-0.0Eo)) returns either positive or negative zero -- implementation dependent.
Closing bug because commenter has not objected to the resolution posted and more than two weeks have passed.