g++ regression - or not ?

[Dirk Mueller]

Hi, 

consider this code: 

class Bar
{
public:
    static void blah() { printf( "static method blah called\n" ); }
};

class Foo
{
public:
    static Bar *bar() { printf( "Foo::bar() called!\n" ); return new Bar; }
};

int main()
{
    Foo::bar()->blah();
}

price question: do you think Foo::bar() will be called, or not ?
with gcc 2.x and quite a few other compilers it won't. with gcc 3 it is 
called. I think this is a bit more consistent, still I consider it a 
optimisation-regression. Am I wrong ?

I'm pretty surprised that there is no warning for the Foo::bar()->blah() 
call. I'd expected one, because this is certainly very bad and misleading 
coding style. Its not obvious that most compilers will optimize the 
Foo::bar() call away. I'd even be okay with an error. 

Opinions ? Comments from the C++ standard experts? :-)

Please keep me in CC, thanks. 


Dirk