[Dwarf-Discuss] PROPOSAL: DW_OP_implicit_pointer

[Jakub Jelinek]

On Thu, Aug 12, 2010 at 09:54:40AM -0400, Ron Brender wrote:
> In reviewing the proposal I feel unclear regarding what problem it
> is intended to solve.
> 
> If I understand correctly, the effect of this operator is equivalent
> to replicating the DW_AT_location or DW_AT_const_value of the
> referenced DIE followed by an DW_OP_add operation using the second
> parameter. Is this a correct understanding.
> 
> If so, it appears that the primary intended benefit is a space
> saving by avoiding replication of potentially large _location or
> _const_value attributes.
> 
> If I have got it all wrong, please clarify.

No, this isn't something that can be expressed currently in DWARF4.

Consider:

struct S
{
  int *x, y;
};

int u[6];

static inline void
add (struct S *a, struct S *b, int c)
{
  *a->x += *b->x;
  a->y += b->y;
  u[c + 0]++;
  a = (struct S *) 0;
  u[c + 1]++;
  a = b;
  u[c + 2]++;
}

int
foo (int i)
{
  int j = i;
  struct S p[2] = { {&i, i * 2}, {&j, j * 2} };
  add (&p[0], &p[1], 0);
  p[0].x = &j;
  p[1].x = &i;
  add (&p[0], &p[1], 3);
  return i + j;
}

int
bar (int i)
{
  int *j = &i;
  int **k = &j;
  int ***l = &k;
  i++;
  return i;
}

An optimizing compiler will likely optimize away all the pointers
and the variables that had address taken (like i, j, k) might
be also optimized away (have just implicit location descriptions)
or not be addressable (live in registers).  Variable p might also
be scalarized and the different pieces kept in different registers.

When the object, to whose address the optimized away pointer points
to, is not addressable, with DWARF4 all the compiler can do is not
emit DW_AT_location (nor DW_AT_const_value) at all.  With this extension
e.g. in bar, while you can't print j's value, you can print *j.
Similarly in the inlined add calls you can print *a->x, a->y, *b->x, b->y
or in foo print say *p[0].x.

	Jakub