Re: fold casted const variables

[Jan Hubicka <jh at suse dot cz>]

> On Sun, 16 Feb 2003, Jan Hubicka wrote:
> 
> > Hi,
> > the following testcase does not get current simplified as it should.
> > 
> > Bootstrapped/regtested on i386.  OK?
> > Honza
> > void link_error (void);
> > const double one=1.0;
> > t()
> > {
> > 	if ((int)one != 1)
> > 		link_error ();
> > }
> > Sun Feb 16 17:34:30 CET 2003  Jan Hubicka  <jh@suse.cz>
> > 
> > 	* c-typeck.c (build_c_cast):  Fold constant variables into
> > 	initial values.
> 
> You have made sure that this doesn't cause anything to be counted as an
> integer constant expression that shouldn't be and isn't currently?  (See
> c9?-const-expr-3.c and add testcases (to both files) for casting a double
> 0.0 to int.  (int)0.0 is constant (pass), (int)(0.0+0.0), (int)+0.0,
> (int)(double)0.0 aren't (probably fail), (int)zero where const double zero
> = 0.0; shouldn't be constant either.)
I've modified testcase into:

/* Test for constant expressions: broken optimization with const variables.  */
/* Origin: Joseph Myers <jsm28@cam.ac.uk> */
/* { dg-do compile } */
/* { dg-options "-std=iso9899:1990 -O2" } */
/* Note: not using -pedantic since the -std option alone should be enough
   to give the correct behavior to conforming programs.  */

static const int ZERO = 0;
static const double DZERO = 0;

int *a;
int b;
long *c;

/* Assertion that n is a constant zero: so the conditional expression
   has type 'int *' instead of 'void *'.
*/
#define ASSERT_NPC(n)	(b = *(1 ? a : (n)))
/* Assertion that n is not a constant zero: so the conditional
   expresions has type 'void *' instead of 'int *'.
*/
#define ASSERT_NOT_NPC(n)	(c = (1 ? a : (void *)(__SIZE_TYPE__)(n)))

void
foo (void)
{
  ASSERT_NPC (0);
  ASSERT_NOT_NPC (ZERO);
  ASSERT_NPC (0 + 0);
  ASSERT_NOT_NPC (ZERO + 0); /* { dg-bogus "incompatible" "bogus null pointer constant" { xfail *-*-* } } */
  ASSERT_NOT_NPC (ZERO + ZERO); /* { dg-bogus "incompatible" "bogus null pointer constant" { xfail *-*-* } } */
  ASSERT_NPC (+0);
  ASSERT_NOT_NPC (+ZERO); /* { dg-bogus "incompatible" "bogus null pointer constant" { xfail *-*-* } } */
  ASSERT_NPC (-0);
  ASSERT_NOT_NPC (-ZERO); /* { dg-bogus "incompatible" "bogus null pointer constant" { xfail *-*-* } } */
  ASSERT_NPC ((char) 0);
  ASSERT_NOT_NPC ((char) ZERO);
  ASSERT_NPC ((int) 0);
  ASSERT_NOT_NPC ((int) ZERO);
  ASSERT_NPC ((int) 0.0);
  ASSERT_NOT_NPC ((int) DZERO);
  ASSERT_NOT_NPC ((int) +0.0);
  ASSERT_NOT_NPC ((int) (0.0+0.0));
  ASSERT_NOT_NPC ((int) (double)0.0);
}

And get same set of warnings from -O2 compilation before and after my
patch, so I guess it is not making things worse.

Honza
> 
> -- 
> Joseph S. Myers
> jsm28@cam.ac.uk