[expr.typeid] glvalue of incomplete type vs typeid

[xmh0511]

struct C;
int storage;
C* ptr = (C*)&storage;
typeid(*ptr);  //#1

All compilers report #1 is an error, with the diagnosis invalid use of incomplete type. However, there's no relevant rule in the standard that restricts the type of the glvalue operand in [expr.typeid]. Is it an omission?

[jwakely]

Since it's unknown whether S is a polymorphic type or not, the behaviour here seems undefined. So giving an error is fine. Making it explicitly ill-formed would probably be better though.

[jensmaurer]

dynamic_cast consistently requires complete types for both source and target.

We have "If the type of the type-id is a class type or a reference to a class type, the class shall be completely-defined."

It seems plausible that this rule should also apply to the "expression" case.

[xmh0511]

Maybe, we could say that

When typeid is applied to an expression, If the static type of the expression is a (possibly cv-qualified) class type, the class shall be completely-defined.

Which is before [expr.typeid] p2.

[xmh0511]

@jensmaurer For [expr.typeid] p3

If the expression is a prvalue, the temporary materialization conversion is applied. The expression is an unevaluated operand.

How about the following case?

typeid(static_cast<void>(0));

The operand is a prvalue of type void as per [expr.static.cast] p1 and [expr.static.cast] p6, which means the temporary materialization conversion is not possible to be applied to that operand. It is a bit contradictory.