xs:NOTATION as a member type of a union

[michaelhkay]

Tests MS-SimpleType2006-07-15 stE056 through stE065 all declare a union type with xs:NOTATION as one of its member types.

On the face of it this violates the constraint (present in both XSD 1.0 and 1.1)

Schema Component Constraint: enumeration facet value required for NOTATION
It is an ·error· for NOTATION to be used directly in a schema. Only datatypes that are [·derived·] from NOTATION by specifying a value for ·enumeration· can be used in a schema.

The precise meaning of this constraint is unclear. Obviously xs:NOTATION can be used as the base type of a restriction with an enumeration facet. Whether it can be used anywhere else is open for debate, but a plain English reading of this constraint says that it can't be used as a member type of a union.

[michaelhkay]

The Saxonica tests over015 and over027 also illustrate the problem.

As regards over015, this was discussed at https://www.w3.org/Bugs/Public/show_bug.cgi?id=14388 . Sandy Gao suggested "the clarification that it's OK to declare an element/attribute to have type xs:NOTATION (not a schema error), but it's not OK to use NOTATION to validate a value (a runtime error)." This perhaps reflects that in XSD 1.1 the rule changed from

It is an ·error· for NOTATION to be used directly in a schema. Only datatypes that are [·derived·] from NOTATION by specifying a value for ·enumeration· can be used in a schema.

to

It is (with one exception) an [·error·] for [NOTATION] to be used directly to validate a literal as described in [Datatype Valid (§4.1.4)]: only datatypes [·derived·] from [NOTATION] by specifying a value for [·enumeration·] can be used to validate literals.

But it's not clear what kind of ·error· this is supposed to be, and anyway, it's still classified as a Schema Component Constraint. It also doesn't make much sense: if NOTATION is the last member type of a union, does that mean the error only occurs when the value fails to satisfy any other member type of the union? And why would the specification allow a schema to contain a construct that will inevitable cause an error at run-time if it's used?

[michaelhkay]

See also https://saxonica.plan.io/issues/7060