- From: Steve Harris <steve.harris@garlik.com>
- Date: Tue, 17 May 2011 16:27:12 +0100
- To: public-rdf-dawg-comments@w3.org
- Cc: Jeen Broekstra <jeen.broekstra@gmail.com>
In response to http://lists.w3.org/Archives/Public/public-rdf-dawg-comments/2011Feb/0025.html > I do have some issues to report with two testcases, both dealing with > the AVG operator. > > Before I go into that: I have assumed that when the definition of AVG > (http://www.w3.org/TR/sparql11-query/#setFunctions) uses the '/' > division operator, this means it applies XPath's op:numeric-divide (as > defined in section 16.3). It would actually be good to clarify this in > the spec, I think (one way to do this would be to explicitly use > op:numeric-divide in the definition, rather than just '/'). > > Anyway, here goes: > > 1. testcase aggregates/agg-avg-01 ("AVG") > > This testcase currently assumes a particular precision in xsd:decimal > division, which is not mandated by the XPath definition. XPath defines > that in xsd:decimal operations, rounding is implementation dependent > (see http://www.w3.org/TR/xpath-functions/#op.numeric, near the end of > the section). I'm not sure that was the intention: "For xs:decimal values the number of digits of precision returned by the numeric operators is ·implementation-defined·. If the number of digits in the result exceeds the number of digits that the implementation supports, the result is truncated or rounded in an ·implementation-defined· manner." The minimum number of decimal places that must be supported my minimally conforming implementations is 18, so I don't think it's correct to round to one decimal place. > In the test, five numbers of type xsd:decimal are averaged (1.0, 2.0, > 3.5, 2.2, 2.2). The average is computed by executing a op:numeric-divide > on the sum of the five decimals: 11.1/5. The result of this is, as the > testcase defines, 2.22. However, this result increases precision to 2 > decimals (note that the precision of the input is 1 decimal). > > While this is not wrong, it is not the *only* correct answer: an > implementation that would round or truncate to 1 decimal and return 2.2 > as the answer would IMHO also perform the operation correctly. I don't believe the definition of decimal sanctions rounding to one decimal place. > To fix this, I would suggest that the testcase numbers are adapted so > that the result of the division when not rounded is of the same > precision as the input. Alternatively, change the test case to work on > floats or doubles, as division behavior is more precisely defined for > those datatypes. > > 2. testcase aggregates/agg-avg-02 ("AVG with GROUP BY") > > This testcase contains a typing error: the result of the AVG computation > where ?s = ex:ints should be of type xsd:decimal, not xsd:integer (XPath > defines the result of a op:numeric-divide on two integer arguments as > always being of type xsd:decimal). I believe this is now fixed. Please respond to this mail saying whether this response satisfies your comment. Regards, Steve, on behalf of the SPARQL Working Group. -- Steve Harris, CTO, Garlik Limited 1-3 Halford Road, Richmond, TW10 6AW, UK +44 20 8439 8203 http://www.garlik.com/ Registered in England and Wales 535 7233 VAT # 849 0517 11 Registered office: Thames House, Portsmouth Road, Esher, Surrey, KT10 9AD
Received on Tuesday, 17 May 2011 15:27:41 UTC