> On Nov 8, 2015, at 12:54 PM, Gregg Kellogg <[email protected]> wrote: > > From the PR <https://github.com/w3c/rdf-tests/pull/23>: > > SPARQL 1.1 tests functions/plus-1 and functions/plus-2 include ORDER BY. > Unfortunately, they sort strings against numbers. This ordering is undefined > by strict SPARQL. The results also don't identical query solutions adjacent > in the ordering which is quite odd. > > This PR corrects that by removing the ORDER BY. The data and results are > unchanged. ORDER BY is not a feature being tested. > > Any system passing the original tests will still pass these corrected tests. > > The queries and the references in the manifest are renamed as > plus-1-corrected and plus-2-corrected. > > It may well be that systems are not checking the ordering when the query has > ORDER BY in it. > > Background: Apache Jena puts a total ordering on any ORDER BY with permitted > extensions to compare literals with different datatypes (roughly, sort by > lexical form then by datatype) > > Adds queries plus-1-corrected.rq, plus-2-corrected. > Removes plus-1.rq and plus-2.rq. > Results remain the same.
This looks good to me. My Attean system passes the new tests (and I admit to being one of systems that doesn’t test ordering for ORDER BY queries). One thing I’m curious about is entirely removing the dawgt:approval triples in the manifests. In my previous SPARQL PR (#21) I used dawgt:approval dawgt:NotClassified for the new tests, but I wonder if CG approval should be indicated similarly (or identically) to the WG dawgt:approval dawgt:Approved indication. I think this is especially important if we’re not going to clean up the obsoleted or un-approved tests from the manifests, but think we should at least discuss the issue either way. thanks, .greg
