RE: Comments on the new RDF Test Cases draft

>  > -----Original Message-----
>>  From: Dan Brickley [mailto:danbri@w3.org]
>>  Sent: Wednesday, May 08, 2002 2:16 AM
>>  To: Massimo Marchiori
>>  Cc: Www-Rdf-Comments@W3. Org
>>  Subject: Re: Comments on the new RDF Test Cases draft
>>
>>
>>  On Wed, 8 May 2002, Massimo Marchiori wrote:
>>
>>  > [sent already, but it didn't seem it went thru... maybe just the thin
>>  > air of Hawaii... retrying now, sorry again for double postings]
>>  >
>>  > I just quickly read (yes, same flight... ;) the new RDF Test
>>  Cases as per
>>  > http://www.w3.org/TR/rdf-testcases/
>>  > In the main text (not time so far to read all the actual use cases... ;)
>>  > there is in Section 2:
>>  > <quote>
>>  > A parser is considered to pass the test if it produces a graph
>>  isomorphic
>>  > with the graph described by the N-triples output document.
>>  > </quote>
>>  >
>>  > This is wrong, according to the standard definition of graph isomorphism
>>  > (care when using words without accurate definitions...!).
>>
>>  Could you give a citation for the 'standard definition', and outline how
>>  we deviate from that concept?
>Citation: every book on graph theory, or just look on Google.
>Definition: a graph isomorphism is an adjacency-preserving bijective map
>between vertices
>(bijection is clear, adjacency-preserving means that two vertices are
>connected in one
>of the graphs if and only if the corresponding, via the map, ones in the
>other graph are
>connected).

That is the definition appropriate for unlabeled graphs. If there is 
a edge labelling defined (which is obviously the case for RDF graphs) 
then the appropriate notion of isomorphism would be a bijection 
between vertices preserving labels of nodes and arcs, not just 
adjacency, ie two nodes labelled with B and C are linked by an arc 
labelled with A iff the corresponding ones in the other graph are 
labelled by whatever B and C map into and connected by an arc 
labelled with whatever A maps into. If we assume that the isomorphism 
uses the identity mapping on labels, this definition will work fine: 
it will allow substitutions of blank nodes, and nothing else.

>  > > You'd define it using the RDF-MT semantical equivalence instead.
>>
>>  Hmm, not so sure. RDF parsers aren't expected to exhibit knowledge of all
>>  the semantic equivalencies implied by RDF's MT.
>Beep! (wrong button! ;)
>Quiz: Why are you doing the MT...?
>Answer to your specific question: correct, RDF parsers don't have to exhibit
>knowledge of
>the MT's semantic equivalence. But the above change just implies that a
>parser,
>to be considered RDF compliant, has to generate a graph that has the same
>semantics as the one you provide in the test case. Rephrased: each test case
>you put in the draft identifies one *equivalence class* of graphs, and a
>parser is free to choose whatever representative of this equivalence class
>it likes.
>The moment you will *check* for compliance, then yes, you will have to
>have knowledge of the MT semantics...

The trouble with using semantic criteria in the syntax is that you 
have to specify WHICH semantics you are using. Does it have to be 
simply equivalent, rdf-equivalent, or rdfs-equivalent?

Pat

Received on Monday, 13 May 2002 11:46:09 UTC