Grey test cases

Brian: (on test case status)
>      rdfms-abouteach (to show whether its illegal or just a warning)

Graham: (on xml:base error1)
>     maybe it's better to duck this particular issue?

WG @ feb f2f: (on rdfms-rdf-names-use see issues list)
> an RDF processor SHOULD emit a warning

Jeremy on charmod literal
> Hence I feel happier with SHOULD language for that part.


I believe there are a number of cases where we should be non-commital.

That is, we should mark certain test cases as resulting in implementation
dependent behaviour.

Possible variation includes:
- differences in actual output
  (so far we have seen:
    a few bagID test cases
    xml literal
  )
- emission or omission of a warning message
  rdfms-rdf-names-use seems a good example
- some implementations reject others accept input
  (but those that accept are obliged to produce a particular triple)


I think we should:
- agree that not all test cases are black and white
- agree on how to handle the grey cases within the test cases document



I note that M&S was, in parts, deliberately vague, deliberately
underspecified, and that this as well as having disadvantages also had
advantages. I think it is clear that some aspects of the whole problem space
are much clearer now than they were four years ago.

While we have mainly been clarifying those vague and underspecified parts of
M&S I do not believe that we should get so carried away with that task as to
forget that there are good reasons not to specify some things, and to be
open to that possibility.

As long as we clearly demark the implementation specific areas, then  most
implementations can give clear error messages when such implementation
specific features are used in a way that could impact interoperability and I
believe we will have done our job.

Jeremy

Received on Tuesday, 2 April 2002 09:46:09 UTC