The problem is shown by the following code: basically, when using the +=
operator on a pointer which is the member of a class as in a line like below:
a.myptr += a.SideEffect();
if the a.myptr pointer is modified by the a.SideEffect() call, the += still
takes into account only the old pointer value to compute the += result. I was
expecting it to use the value modified by the a.SideEffect() call. See code
below to reproduce.
The behaviour I expected can be obtained if, in the code below, the class A is
removed and all variables/functions are changed to be global
variables/function.
Compiled under RHEL5, kernel 2.6.20-hardened-r2
gcc --version gives:
gcc (GCC) 4.1.2 20070626 (Red Hat 4.1.2-14)
---------- sample code -----------
#include <stdio.h>
class A{
public:
int SideEffect(void){
// prints that we are at position 0 in the buffer (correct)
printf("position in buffer when entering SideEffect: %d\n",
myptr-buffer);
myptr += 2;
// prints that we are at position 2 in the buffer (correct)
printf("position in buffer when leaving SideEffect: %d\n",
myptr-buffer);
return 10;
}
char buffer[256];
char * myptr;
};
int main(void){
A a;
a.myptr = a.buffer;
a.myptr += a.SideEffect();
// Likely bug: prints position 10, I was expecting 12
printf("position in buffer after calling SideEffect and += operator, expected
value is 12: %d\n", a.myptr-a.buffer);
return 0;
}
--
Summary: += operator on pointers with a function call having side
effects on the right hand side
Product: gcc
Version: 4.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: c++
AssignedTo: unassigned at gcc dot gnu dot org
ReportedBy: daniel dot fredouille at gmail dot com
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=34948
