Re: Problems with exception handling in 2.96 snapshots

[Nathan Sidwell <nathan at codesourcery dot com>]

Max Weninger wrote:
> 
> Hi,
> 
> I just discovered the following behaviour with the latetet 2.96 snapshot
> 
> class A {
> public:
>         A( ){;}
>         ~A( ){;}
>         void test( ) throw( A );
> };
> 
> When i try to compile this code the following errors came up
> 
> x.C:5: invalid use of undefined type `class A'
> x.C:1: forward declaration of `class A'
> 
> The same worked with all older versions e.g. 2.95.2
> 
> Whats wrong ?
Your code :-) Arguably c++.

[9.2]/2 says that within the definition of a class, the class is considered
complete within function bodies, default args & ctor initializers. Elsewhere
it is incomplete. Therefore it is incomplete in a member function prototype, and
the exception specifier.

[8.3.5]/6 says that the type of a parameter or the return type for a function
declaration that is not a definition may be an incomplete type. (I can't find
where it says a declaration which is a definition must have complete type,
but I infer that from this part of the standard.)

The body of a function is the bit between { ... } ([8.4]/2

So, your throw specifier is a declaration and A is incomplete at that point.
This is annoying. However, there is a bigger problem, which looks like a defect
in the standard. Here it is

struct A {
  void baz (A arg);
  void foo (A arg) {}
};
void A::baz (A arg) {}

Here A::foo is a definition, so its parameters may not have incomplete type.
A::baz in A is not a definition, so its parameters may have incomplete type.
The definition of A::baz is after the definition of A, so A is complete.

I don't think compilers actually implement that, but it's what the words say.

nathan

-- 
Dr Nathan Sidwell   ::   http://www.codesourcery.com   ::   CodeSourcery LLC
nathan@codesourcery.com : http://www.cs.bris.ac.uk/~nathan/ : nathan@acm.org