Re: wrong C++ warning "fix" last year

[Matt Austern <austern at apple dot com>]

On Tuesday, March 4, 2003, at 09:18 AM, Michael Matz wrote:

> Hi Matt,
>
> your patch from
> http://gcc.gnu.org/ml/gcc-patches/2002-10/msg00112.html
>
> Breaks the following testcase:
>
> --------- snip ----------
> class A {
> public:
>         char *const p;
> };
>
> class B {
> public:
>         B() { }
>         A a;
> };
> --------- snap ----------
>
> What you did, was removing a warning if the const-member-containing 
> type
> (M here) isn't an aggregate.  The problem now is, that this also 
> removes
> any warning in the case above.

I've looked at gcc 3.2's behavior.  My opinion:
 (a) My patch was correct.  gcc 3.2 did warn about this program, but 
gcc 3.2's warning was wrong.  gcc 3.2 gave a warning for class A (you 
can see that the warning is for line 3), saying that class A was 
broken.  That's false.  Class A is perfectly fine.  It's only class B, 
which uses class A incorrectly, that's broken.  gcc 3.2, like 3.3 and 
mainline, is silent about the error in class B.
 (b) A warning or error for class B, which really is broken, would be a 
good idea.  However, that's not related to my patch.  The warning/error 
would be for a class that inherits from an aggregate with const member 
variables and no constructors.

			--Matt