Validity & Must Not/Should Not/May Not

Robert Burns writes:

> On Jul 13, 2007, at 2:18 PM, Smylers wrote:
> > Robert Burns writes
> > 
> > > However even it ifs invalid it can still be may not/ should not /
> > > must not.
> > Oh, I had't thought of that, sorry.  What things are we classifying
> > as invalid yet still only telling authors that they "may not" or
> > "should not", rather than "must not"?

Sorry to repeat the question, but I'm still struggling with this and
think it would be clearer with an example.  Please can you give an
example of something which would be invalid yet not a "must not" item.

> > What do those states represent?
> I'm suggesting its not as simple as saying something is valid or
> invalid. We have a much bigger toolbox than that and we should make
> use of it. Let's not always use the hammer. In other words, we do not
> have to insist that conformance checkers flag [whatever] as invalid.
> We can ask the question, how  should the conformance checker respond
> ...? It can tell the author there is an error (i.e., "must  not"...).
> The conformance checker could issue a warning (i.e., should not). Or
> there could be some sort of comment ("may") ...

If the spec has a list of things that an author must not do then surely
it's invalid for an author to do any of those things?  And that the
point of a conformance checker is to check that the author conforms with
the requirements, to fail a document that does invalid things?

Or, to put it the other way round, if something is only a "should not"
then it's permitted to do it, so how can it be invalid?

> Its a lot to keep track of all of the options for conformance criteria
> (must not, should not, may not, may, should, must) and the different
> conformance audiences such as authors, and all sorts of UAs, general
> UAs, authoring UAs,  conformance checking UAs, conversion UAs, visual,
> aural, tactile, etc.  However, these options  also  provide us  with a
> lot of fine-grained control over our  recommendations. We should make
> use of that power.

Yes.  Though I suspect that in many situations it isn't necessary to
consider so many categories of user-agents:

* Programs that generate HTML are bound by the author requirements for
  HTML they generate.  That applies whether the content is being
  provided by a human or a document in some other format is being

* Conformance checkers are checking that HTML meets the authoring
  requirements, so obviously that's what they need to implement.


Received on Sunday, 15 July 2007 20:42:11 UTC