Proposed syntax for anonymous union types

union-types.md

@@ -0,0 +1,74 @@
+# Union Types
+
+**Author**: Michael Kay, Saxonica
+
+Proposal to add anonymous union types as a new kind of ItemType. 
+
+
+## Description
+
+If you want to write a function that accepts instances of `xs:date`, `xs:time`, and `xs:dateTime`, there
+are several approaches available:
+
+* You can define three separate functions (as the standard function library does, for example for `format-dateTime` and `adjust-dateTime-to-timezone`).
+* You can define the function signature to accept `xs:anyAtomicType`, and then check that the supplied value
+is one of the accepted types at run-time.
+* You can define a union type in a schema, import the schema, and then reference the union type in the function's type signature.
+
+Using a union type clearly has attractions over the other two approaches, but the downside is the paraphernalia
+of creating a schema and importing it. Even in XSLT, where `xsl:import-schema` allows inline schemas to
+be defined, this is a lot of boilerplate.
+
+This proposal extends the `ItemType` syntax with an option to define a new anonymous union type, for example:
+
+```
+union(xs:dateTime, xs:date, xs:time)
+```
+
+This can then be used in function signatures, for example:
+
+```
+declare function f:format($value as union(xs:dateTime, xs:date, xs:time)) as xs:string {...};
+```
+
+Specifically, the grammar for `ItemType` is extended with the option `LocalUnionType`, where the
+grammar for `LocalUnionType` is defined as:
+
+```
+LocalUnionType ::= "union" "(" QName ("," QName)* ")"
+```
+
+*Note: one could argue for the minimum number of member types in a union type being zero, one, or two. I've made
+it one for a number of reasons: it gives the simplest grammar, the case of an empty union serves almost no practical purpose,
+while restricting it to two or more is unnecessarily paternalistic.*
+
+Each of the QNames must identify a **generalized atomic type** as defined in the XPath/XQuery specification,
+which must be present in the in-scope schema types.
+
+The production `SingleType` (used in `cast as` and `castable as` expressions) is also extended to 
+allow a `LocalUnionType` as an alternative to a `QName`.
+
+The semantics of a `LocalUnionType` are exactly the same as the semantics of a union type defined in a schema 
+with the same member types, in the same order. For example, the rules for type subsumption 
+mean that `union(A, B)` is a subtype of `union(A, B, C)`. The type defined meets the definition of
+a *pure union type*.
+
+A union type can appear wherever an `ItemType` or `SingleType` is permitted, for example as part of a `SequenceType`
+in variable and function declarations, or as the operand of a `cast as` or `castable as` expression.
+
+
+## Use Cases
+
+The primary use case, given in the introduction, is to make it easier to define functions that are polymorphic in 
+the sense of accepting several different atomic types as an argument, without the administrative machinery of 
+creating the type in a schema and importing the schema.
+
+The ability to cast to a union type can also be convenient, for example casting to `union(xs:integer, xs:decimal, xs:double)` 
+will give different results depending on the lexical form of the supplied operand, which would otherwise require
+a complex sequence of `castable as` tests.
+
+Defining the result of a function as a union type can also give more precise typing.
+
+## Grammar
+
+TBA.