[Bug c++/14958] New: incorrect compile of ?:, specifically ptr = flag? class_obj: 0;

[Bug c++/14958] New: incorrect compile of ?:, specifically ptr = flag? class_obj: 0;

[gsmith at alumni dot uwaterloo dot ca]

I have a class tmparray<T> which converts to T*.

When I use

short * foop = flag? foo : 0;

... where foo is a tmparray<short>, I expected this
to act as

short * foop = flag? (short*)foo : (short*)0;

In fact, when flag is false, a non-zero value
is placed in foop which is different from what you
get when flag is true, but close to that.

If *either* of the casts are added, it works properly.
This seems to be independent of -ansi and -O, which
makes me wonder if there's some bizarre reason this
*should* actually happen, but it doesn't seem to make
a lot of sense. It also happens on 3.2

Example attached. It prints this for me
----------------------
address of foo = 0xbffff810, data at 0x8049998
foop = 0x80499b8
address of foo = 0xbffff810, data at 0x8049998
foop = 0x8049998
----------------------
I expected this (this is what you get when a cast is added)
----------------------
address of foo = 0xbffff810, data at 0x8049950
foop = (nil)
address of foo = 0xbffff810, data at 0x8049950
foop = 0x8049950
----------------------
Here is the code
======================

template <class T>
class tmparray {
 protected:
    T * m_p;
 public:
    tmparray( int n ){
        m_p = new T [n];
    }
    operator T* () const { return m_p; }
	void discard() {
		if( m_p ){
			delete [] m_p;
			m_p = 0;
		}
	}
	~tmparray() { discard(); }
};
#include <stdio.h>
	
//
// the ?: in the function below is not compiled
// correctly (I think). When the flag is false,
// the result stored into foop is non-zero and
// rather strange. When *either* expression alongside
// the ':' is cast to (short*), it works properly

void buggyfunc(bool flag)
{
	tmparray<short> foo(12);
	printf("address of foo = %p, data at %p\n", &foo, (short*)foo);
	short * foop = flag? foo : 0;
	printf("foop = %p\n", foop );
}


main()
{
	buggyfunc(false);		// foop should be (nil)!!
	buggyfunc(true);		// foop should print data address
}

-- 
           Summary: incorrect compile of ?:, specifically ptr = flag?
                    class_obj: 0;
           Product: gcc
           Version: 3.3.3
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: c++
        AssignedTo: unassigned at gcc dot gnu dot org
        ReportedBy: gsmith at alumni dot uwaterloo dot ca
                CC: gcc-bugs at gcc dot gnu dot org


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14958

[Bug c++/14958] incorrect compile of ?:, specifically ptr = flag? class_obj: 0;

[pinskia at gcc dot gnu dot org]

------- Additional Comments From pinskia at gcc dot gnu dot org  2004-04-14 18:35 -------
No this is invalid as it is really doing the following:

(short*) (flag? typeof(foo)(foo) : typeof(foo) (0))

so that it is creating a tempary variable to store the tmparray<T> array.
you need to explicantly do the follow to get the behavior you want:
short * foop = flag? (short*)foo : (short*)0;

-- 
           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|UNCONFIRMED                 |RESOLVED
         Resolution|                            |INVALID


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14958

[Bug c++/14958] incorrect compile of ?:, specifically ptr = flag? class_obj: 0;

[bangerth at dealii dot org]

------- Additional Comments From bangerth at dealii dot org  2004-04-14 18:54 -------
Indeed. First, the result 
type of any expression is independent of the type of the variable 
it is assigned to, so the conversion you proposed can't happen. 
Second, the type of one of the 2nd/3rd parameters to the ?: operator 
is converted to the type of the other one if possible. In your case, 
  flag? foo : 0 
is equivalent to 
  flag? foo : (foo)0 
with foo=tmparray<short>. 
 
W. 

-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14958

[Bug c++/14958] incorrect compile of ?:, specifically ptr = flag? class_obj: 0;

[gsmith at alumni dot uwaterloo dot ca]

------- Additional Comments From gsmith at alumni dot uwaterloo dot ca  2004-04-14 19:28 -------
Thank you! I made the constructor 'explicit' and now I get an error
on that line, which is much better. I couldn't see any way that this
could be valid *and* produce these particular results, sorry to
clutter up the buglist.
This explains why both of these work:

short * foop = flag? foo : (short *)0;
short * foop = flag? (short*)foo : 0;




-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=14958