Is there any reason for §3.3.7/1.2 to be considered an error?

Question

This code was obtained from the example in §3.3.7/1.5:

enum { i = 1 };
class X {
    char v[i]; // error: i refers to ::i
    // but when reevaluated is X::i
    enum { i = 2 };
};

GCC emits an error because of §3.3.7/1.2

However, if we apply §3.4.1/7, the lookup for the name i in the declaration char v[i]; will find enum{ i = 1 }; in global scope. What's the problem with the redeclaration enum{ i = 2 };?

Related to [Is rule 3 in section 3.3.7/1 from N3936 redundant?](http://stackoverflow.com/q/26761385/1708801) — Shafik Yaghmour, Mar 11 '15 at 21:05

score 3 · Accepted Answer · answered Mar 11 '15 at 21:02

3

The problem is that in member function scope (and other class scopes) the class version of the enum shadows the global value.

If it wasn't an error, the compiler would have to pick one of two surprising behaviors:

1) Use the class enum's value always, which means it would have to fully parse the class to determine if a valid size is available.

2) Have the value of i within class members be completely different from the actual length of the class member array declared, notionally, to be the same length.

answered Mar 11 '15 at 21:02

Mark B

95,107
10
109
188

I understand this. But I don't see a reason for an error. – Ayrosa Mar 11 '15 at 21:03
@Ayrosa what should `sizeof X{}.v` yield? – M.M Mar 11 '15 at 21:20
Your number (2) justifies it. Thanks. – Ayrosa Mar 11 '15 at 21:27

Is there any reason for §3.3.7/1.2 to be considered an error?

1 Answers1