Re: RFC: fix cast to void

[Pedro Alves <palves at redhat dot com>]

On 11/28/2012 04:41 PM, Tom Tromey wrote:
> This patch fixes a bug I noticed while debugging another problem.
> 
> In some cases a cast to void will fail like so:
> 
>     (gdb) p (void) v_int_pointer 
>     Attempt to dereference a generic pointer.
> 
> The problem is that the check for cast-to-void comes after the
> lval_memory case.
> 
> Fixed as appended.  New test included.
> Built & regtested on x86-64 Fedora 16.
> 

We should move it even before:

  else if (TYPE_LENGTH (type) == TYPE_LENGTH (type2))
    {
      if (code1 == TYPE_CODE_PTR && code2 == TYPE_CODE_PTR)
	return value_cast_pointers (type, arg2, 0);

      arg2 = value_copy (arg2);
      deprecated_set_value_type (arg2, type);
      set_value_enclosing_type (arg2, type);
      set_value_pointed_to_offset (arg2, 0);	/* pai: chk_val */
      return arg2;
    }

because TYPE_LENGTH of void is 1, so with:

 char foobar = 1;

this:

 (gdb) p (void) foobar

enters that TYPE_LENGTH==TYPE_LENGTH branch, and returns with an arg2 that
has the wrong VALUE_LVAL.  Here's a test for that:

(gdb) p & (void) foobar
$6 = (void *) 0xc215d9 <foobar>

This should be an error.

-- 
Pedro Alves