Typedecl::membership() should be optional because of 'local classes'

[d-winsor]

Typedecl will not have a membership if it is declared locally in a function.

int foo()
{
    struct Local { int x; };
}

[GabrielDosReis]

It should instead return a const Expr& instead of const Type&: in the above case we want to designate foo as its owner, or the nested block if the definition appears in deeper. Alternatively, it could just systematically return const Region&.

[d-winsor]

Would the owner be ipr::Fundecl (foo) or the ipr::Block. Consider the case below:

void foo()
{
    {
        struct SameName{ int blue; };
    }
    {
        struct SameName{ int red; };
    }
}

[GabrielDosReis]

Yes, that is exactly why I suggested foo, or the enclosing block if nested. foo is for the case where it is the outermost compound statement that makes up the body of the function -- so no indirection is needed to find the function that owns the block that owns the type.

[d-winsor]

Sorry I just miss read your reply,