You're right of course, but I had a devil of a job
discovering why! This is one example of the spec being
terribly unhelpful: it says quite clearly (in section
3.3.1): "Values of type NMTOKEN must match the Nmtoken
production...", and then qualifies this in section 3.3.3, by
saying in effect: "the value I was talking about in section
3.3.1 isn't the "attribute value" defined in section 3.1 as
you might have thought, rather it's the "normalized
attribute value" defined as follows..."
Next time round could we please have a bit more rigour in
the spec!?
Mike Kay