Nested types can only be declared in primary declaration.
- Simplifies qualified name lookup because extensions cannot define new nested types.
- Makes sense in the context of scoped conformances because the definition of a type in an extension may break someone else's code. (not so compelling because of properties?)
Make a distinction between extensions and givens.
- An extension reopens the scope of a type and adds new members.
- A given exposes a conformance.
Let a.m be an expression where a has type T, we may resolve m to an entity introduced in the primary declaration of a or in an extension of a, but not to an entity introduced in a given.
If m denotes a trait requirement, then name resolution should bind it to the entity introduced in the trait, qualified by a witness of a's conformance to that trait.
- Store the contents of "small" type trees in the inline storage of their identities.
- Use a separate array to store the tag of each syntax tree rather than calling
tag(of:).
- Is it desirable to write extensions of context functions?