[cfe-commits] Unique'ing VariableArrayType

[Argiris Kirtzidis]

After the subtle bug that was uncovered here:
 http://lists.cs.uiuc.edu/pipermail/cfe-commits/Week-of-Mon-20081006/008095.html
I think that VariableArrayTypes should be unique'd based on the 
expression that represents the size.

The comments of over VariableArrayType class mention:

> /// Note: VariableArrayType's aren't uniqued (since the expressions 
> aren't) and
> /// should not be: two lexically equivalent variable array types could 
> mean
> /// different things, for example, these variables do not have the 
> same type
> /// dynamically:
> ///
> /// void foo(int x) {
> ///   int Y[x];
> ///   ++x;
> ///   int Z[x];
> /// }

But those two '[x]' expressions are different DeclRefExpr objects.
Not unique'ing the VariableArrayTypes doesn't seem to have any benefit, 
since it's a bug to create two VariableArrayTypes with the same Expr* 
(they will both try to destroy it).
And in general it's unintuitive to have Sema::GetTypeForDeclarator 
create different Type* objects each time it gets called with the exact 
same declarator.

Any thoughts ?

-Argiris