Re: [www-qa] After the Workshop, still plenty of options to consider

In starting this thread, I emntioned that a WG could do this:
>4. Beyond a few normative examples, WG aims to have numerous normative
>    cases in the body of the Rec.

Lofton Henderson replied:
>About #4 below, I'm puzzled by "normative examples". In my
>experience, it is common that examples are informative. Else, you have
>a normative description and a normative example of the same
>functionality, and they could be inconsistent.

I'm not too comfortable with #4 myself, but I think that it is at
least worth mentioning that a Recommendation could say "given this
input [example...], a conforming processor must produce this output
[example...]" with the usual verbiage about equivalence of InfoSets
rather than character-wise matching. The WG would only do this if
they thought they could produce 10-50 cases when a total suite would
have many more. Moving up to #5 (having enough cases to call it an
incomplete test suite) is clearly better.

As for the inconsistency issue: I think that Recommendations can be
internally inconsistent even if they have no examples, because there
are other ways in which the same assertion may be made more than once
in a spec. I can show you several such situations in the XSLT spec,
but I think that they're all consistent ones (since the errata came
out, at least). If we've managed to achieve consistency when stating
something in the body text more than once, then I think there is hope
that the same could happen with normative examples.

If someone wishes to develop a documentation style where every
assertion is made absolutely once only, I'd be interested in looking at
it. A spec written that way might need to be supplemented by text that
a human can read sequentially, though the text would be non-normative.
.................David Marston

Received on Thursday, 12 April 2001 16:15:28 UTC