Optimization bug

[Carlo Wood]

Hi,

at least gcc version 2.95.1 and 2.96-20000402 have an optimization bug
as described below.  I did not test other versions, but I seems likely
that all versions of gcc have this bug.

The problem is gcc is confused near `return' with its local variables
on the stack: returning from a function exits the scope of the local
variables, so that when a recursive call to the same function is
performed, this call is replaced by a jmp to the start of the function
and the same stack space is used for the local variables.

This is incorrect however when a pointer to a local variable is
passed in the recursive call.  This particular bug breaks a part of
binutils (I found it in objdump) in supposedly rare circumstances
(and they refuse to write a work around when it is a compiler bug ;).

I wrote a little test case to demonstrate the problem:

-----------------------------------------------------
int foo(int* p)
{
  int i = 1;
  *p += 2;
  if (*p == 3)
    return i;
  return foo(&i);
}

int main(void)
{
  int i = 0;
  return foo(&i);
}
-----------------------------------------------------

>gcc -v
Reading specs from /usr/lib/gcc-lib/i686-redhat-linux/2.95.1/specs
gcc version 2.95.1 19990816/Linux (release)

>gcc-cvs -v
Reading specs from /usr/local/egcs/lib/gcc-lib/i686-pc-linux-gnu/2.96/specs
gcc version 2.96 20000402 (experimental)

>gcc -O0 20000812.c
>a.out; echo $?
1
>gcc -O1 20000812.c
>a.out; echo $?
3

>gcc-cvs -O0 20000812.c
>a.out; echo $?
1
>gcc-cvs -O1 20000812.c
>a.out; echo $?
1
>gcc-cvs -O2 20000812.c
>a.out; echo $?
3


Regards,

-- 
Carlo Wood <carlo@alinoe.com>                        -=- Jesus Loves you -=-