Re: Comments on GRDDL (transformation failure case)

This is an informal message concerning on-going consideration of your 
comment.

===

The WG asked me to construct a draft test case, to be considered for our 
test cases document

http://www.w3.org/TR/grddl-tests/

The draft is found at
http://www.w3.org/2001/sw/grddl-wg/td/pendinglist#error1
and shows that the document
http://www.w3.org/2001/sw/grddl-wg/td/withErrors
has a GRDDL result
http://www.w3.org/2001/sw/grddl-wg/td/one.rdf
(i.e. the merge of all possible GRDDL results for this document)

The document actually refers to six different transforms, but only the 
first works, and that alone gives the result specified - showing that no 
GRDDL result is produced by the other five transforms.

The errors in the other five transforms are various:
- 404
- an XSL document with a syntax error (a moderately catastrophic one!)
- an XSL document that produces XML that isn't RDF/XML
- two XSL document which terminate abnormally (using xsl:message 
terminate="yes") [one of these terminates before producing any output at 
all, the last after producing several triples that do not appear in the 
GRDDL result for this document]

The WG is likely to discuss whether we think this is an adequate 
response to your comment on Wednesday, the purpose of this message is to 
allow you to comment further, if you believe that would be helpful.

Note that in terms of:
[[
The extraneous value could be silently ignored, or the entire answer 
could be considered meaningless and the invoker informed of the error.
]]
this test case (and I believe the GRDDL specification) is deliberately 
silent. It shows the correct GRDDL result, but does not constrain 
implementations to either silently ignore errors, or to report them. It 
does show that erroneous input does not produce GRDDL results, and does 
not impact the processing of other GRDDL results.

Although it has yet to be confirmed by the WG, I believe that this 
choice of deliberate silence concerning the nature of the error handling 
is in-line with the consensus of the GRDDL WG.

Hope this helps

Jeremy Carroll

PS My own implementation is fairly noisy on this test - I think I'll add 
an option to silently ignore the errors.



-- 
Hewlett-Packard Limited
registered Office: Cain Road, Bracknell, Berks RG12 1HN
Registered No: 690597 England

Received on Friday, 25 May 2007 10:31:11 UTC