Reference to pointer and cast - is this a bug ?

Reference to pointer and cast - is this a bug ?

[Philippe Combes]

Hi all,

Here is code that works exactly as I think it should with g++-2.95, but
not with
g++-4.0/4.1

#include <iostream>
using namespace std;

class parent
{
public:
         parent():i(0){}
         parent(int i_):i(i_){}
private:
         int i;
};

class child : public parent
{
public:
         child():parent() {}
         child(int i_):parent(i_){}
};

void trace( const parent* const& p_ptr )
{
         cout << "trace called on a reference tp pointer @"
              << (void*)&p_ptr << ": pointer = " << (void*)p_ptr << endl;
}

int main( int argc, char** argv )
{
         const child c(4);
         const child* c_ptr = &c;
         const child* const& c_ptr_r = c_ptr;
         cout << "Call trace on a reference to pointer @"
              << (void*)&c_ptr_r << ": pointer = " << (void*)c_ptr_r <<
endl;
         trace( c_ptr_r );
         return 0;
}


Output with g++-2.95 (which I think is right)
Call trace on a reference to pointer @0xbfb43a40: pointer = 0xbfb43a44
trace called on a reference tp pointer @0xbfb43a40: pointer = 0xbfb43a44

Output with g++-4.0 or 4.1
Call trace on a reference to pointer @0xbfac899c: pointer = 0xbfac89a0
trace called on a reference tp pointer @0xbfac89a4: pointer = 0xbfac89a0

which means that the pointer passed by reference to trace has been
copied, just
as if it was passed by value.

If this is a bug, I did not find it in the bug database.
If not, can anybody tell me why ?


Thanks in advance,

Philippe

Re: Reference to pointer and cast - is this a bug ?

[John Love-Jensen]

Hi Philippe,

> If this is a bug, I did not find it in the bug database.
> If not, can anybody tell me why ?

Not a bug.

A child is a parent.

A child pointer, itself, is a type.
A parent pointer, itself, is a type.

A child pointer is not a parent pointer.

Put ten more trace(c_ptr_r); in your source.  Notice that each one has a
different address.  Anonymous variables are being created - anonymous
variables that have a different type.

Take out the const part of the const& to trace, and you'll notice that the
compiler indicates that is an error - because a child pointer is not a
parent pointer.

Since you used a const&, the child pointer will be treated as if it is a
parent pointer (via the magic of the anonymous variable introduced and
implicit conversion).

The trace statement is printing the address of the anonymous variable, which
is why it differs.

May have something to do with the scaffolding for the SSA optimization, even
in non-optimized code (-O0).

HTH,
--Eljay