Saxon

See https://github.com/w3c/qt3tests/issues/28

Saxon is accepting map { 1:'A','x':'B' } as an instance of function(xs:anyAtomicType) as xs:string. But this is incorrect: if the map is treated as a function $F, then $F(23) returns () which is not an instance of the declared return type xs:string.

The logic in MapType.relationship() line 324 makes map(K, V) a subtype of F if function(K) as V is a subtype of F, which seems quite wrong; the rule is §2.5.6.2 rule 35, which says map(K, V), is a subtype of function(xs:anyAtomicType) as V?.

So both Saxon and the QT3 test appear to be wrong.