Re: conformance clause + extensibility

> If my implementation receives a schema-valid document containing either 
> an extension or an unimplemented feature, it simply prints "not 
> supported" along with explanatory details and exits. If it receives a 
> non-schema-valid document, something to that effect is printed (actually 
> the JAXB unmarshaller throws an exception) and the translator exits.

Okay, I guess we're okay here.  I think to do otherwise (to continue
silently) would be a failure to implement a "semantics-preserving
mapping", so it's already forbiden.

So, I'm okay with droping the text...

      A conformant RIF-BLD consumer must reject all inputs that do not
      match the syntax of BLD. If it implements extensions, it may do so
      under user control -- having a "strict BLD" mode and a
      "run-with-extensions" mode.

as well as the "at risk" warning.  I think that's what everyone else
wanted during the last meeting, and I can live with that.

    -- Sandro

> Sandro Hawke wrote:
> > According to our specs, what should a conformant RIF consumer do when
> > fed a RIF document which includes extensions it doesn't know about?
> >
> > My concern is that it might well ignore the extensions, trying to
> > silently remove them, parsing out the bits of the document it's
> > expecting to find and ignoring the rest.  That seems like a
> > dangerous-yet-tempting practice.
> >
> > A related question is what RIF consumers should do about features they
> > are required to implement but do not implement.  They can't be
> > conformant, but maybe we can still give some practical guidance.  For
> > instance, I think a consumer ought to give a warning or error on seeing
> > such features used, instead of perhaps passing them through in some
> > silent-failure mode.  Builtins might well just be treated as logic
> > functions, in BLD, in which case you'd quietly get the wrong answers.)
> >
> >    -- Sandro
> >
> >

Received on Monday, 15 March 2010 22:48:14 UTC