Lists Home |
Date Index |
> > And I'd rather have a spec written in prose than a schema any
> day of the week
> > .
> Having all three is best (formal spec, prose spec, and sample
> instances), if they're all available. But I think I'd still
> pick the formal spec if I were forced to choose just one.
Ditto. You can to some extent generate a formal spec from a prose spec and
vice versa, but where the formal-> prose may be lacking in higher level
semantics it's less likely to be *wrong*.
> There's an issue of quality, too. A prose spec written by, say,
> James Clark, is often sufficient, whereas one written by, say,
> Dave Winer may not be any more informative than the collection
> of sample instances.
I've recently being trying to deal with data that uses one of Mr. Winer's
specs . Wayne Steele has generated a DTD from the prose, which is linked
from the spec. The DTD contains the comment:
"You can add whatever other attributes on the outline element you want."
> With a formal notation, you can usually count on at least _some_
> degree of precision.