The exception bug in egcs-970929

The exception bug in egcs-970929

[H.J. Lu]

This code is miscompiled with -O and dwarf based exception on x86.

# gcc -O des.cc
# a.out
zsh: 26333 segmentation fault  ./a.out

If dwarf based exception is not used, everything is fine. Somehow
g++ adjusts stack twice when calling destructor.

-- 
H.J. Lu (hjl@gnu.ai.mit.edu)
---
class Int 
{
protected:
  int          rep;
public: 
  Int () :rep(0) {}
  Int (const Int& b) :rep(b.Int::val()) {}
  ~Int();
  operator int() const { return val(); }
  int   val() const { return rep; }
  void  operator  = (const int b)
  { rep = b; }
  void  operator ++ ()
  { ++rep; }
};
Int::~Int() {}
void Proc0();
void Proc7(Int IntParI1, Int IntParI2, Int *IntParOut);
main()
{
	Proc0();
	exit(0);
}
void Proc0()
{
	Int		IntLoc1;
	Int		IntLoc2;
	Int		IntLoc3;
	IntLoc1 = 2;
	IntLoc2 = 3;
	while (IntLoc1 < IntLoc2)
	{
		IntLoc3 = 5 * IntLoc1 - IntLoc2;
		Proc7(IntLoc1, IntLoc2, &IntLoc3);
		++IntLoc1;
	}
}
void Proc7(Int IntParI1, Int IntParI2, Int *IntParOut)
{
	Int	IntLoc;
	IntLoc = IntParI1 + 2;
	*IntParOut = IntParI2 + IntLoc;
}

Re: The exception bug in egcs-970929

[Andreas Schwab]

H J Lu <hjl@lucon.org> writes:

|> This code is miscompiled with -O and dwarf based exception on x86.
|> # gcc -O des.cc
|> # a.out
|> zsh: 26333 segmentation fault  ./a.out

|> If dwarf based exception is not used, everything is fine. Somehow
|> g++ adjusts stack twice when calling destructor.

I have just sent this to Kenner for gcc2:

The FDE insns that belong to a specific pc value always come after a
DW_CFA_advance_loc that advances to that pc value.  Thus when
__frame_state_for executes the FDE insns it must still do that if the
current pc value equals the target pc value.

1997-10-04  Andreas Schwab  <schwab@issan.informatik.uni-dortmund.de>

	* frame.c (__frame_state_for): Execute the FDE insns until the
	current pc value is strictly bigger than the target pc value.

*** testgcc-970926/frame.c.~1~	Wed Sep 24 23:05:26 1997
--- testgcc-970926/frame.c	Sat Oct  4 18:43:39 1997
*************** __frame_state_for (void *pc_target, stru
*** 600,606 ****
    /* Then the insns in the FDE up to our target PC.  */
    end = next_fde (f);
    pc = f->pc_begin;
!   while (insn < end && pc < pc_target)
      insn = execute_cfa_insn (insn, &state, &info, &pc);
  
    memcpy (state_in, &state.s, sizeof (state.s));
--- 600,606 ----
    /* Then the insns in the FDE up to our target PC.  */
    end = next_fde (f);
    pc = f->pc_begin;
!   while (insn < end && pc <= pc_target)
      insn = execute_cfa_insn (insn, &state, &info, &pc);
  
    memcpy (state_in, &state.s, sizeof (state.s));

-- 
Andreas Schwab                                      "And now for something
schwab@issan.informatik.uni-dortmund.de              completely different"