At the ftf of May 2005 the WG recognized the need to ensure that all
parts of the Datatypes spec provide a consistent account of the nature
of lexical mappings and the value and lexical spaces.

The invariants we wish to express are:

   For all types, the lexical space is the domain of the lexical
   mapping, and the value space its domain.  There are no ineffable
   values, there are no meaningless lexical forms.

   For all primitive types and all types derived from them by
   restriction or constructed from them by list, the lexical
   mapping is a function:  each lexical form uniquely determines
   a value.

   For the special types, however, and for union types, the
   lexical mapping is not (necessarily) a function.  When these
   are used as the types of elements, xsi:type can used
   to specify which mapping to use.  

(It is also true that the context in which our type system 
is used may provide other mechanisms for addressing this 
problem.  The untypedAtomic mechanism of QT is one such; 
we should write nothing that appears to conflict with the 
existence of such a mechanism.)

The descriptions of the special types, of lexical mappings in
general, and of unions should be reviewed and if necessary
revised to ensure that the invariants identified above are
stated clearly and that nothing is stated which contradicts
them.

At the ftf in August 2005 the WG asked the editors to provide
wording for this topic to be considered at our meeting of
2 September 2005.