Re: [rfc] Use LONGEST instead of value for index arithmetic

["Ulrich Weigand" <uweigand@de.ibm.com>]

Daniel Jacobowitz wrote:
> On Fri, Jun 26, 2009 at 05:49:07PM +0200, Ulrich Weigand wrote:
> > Hello,
> > 
> > this patch implements the change I mentioned recently: use LONGEST instead
> > of struct value to represent array indexes (and similarly the integer
> > operands to pointer arithmetic).  This has the advantage that index
> > computations can be performed in host instead of target arithmetic,
> > and not just eliminates a bunch of references to global built-in types,
> > but makes both implementation and users of the value_subscript and
> > pointer arithmetic routines quite a bit simpler.
> 
> Looks good to me.  I noticed you're removing several error calls; do
> those garbage cases still produce some error message?  I guess they
> must be untested :-(

All error messages should be preserved by my changes, but I admit this
isn't quite obvious.  For example, in the "pointer + not-integer" case,
the following error used to be emitted by value_ptradd:

-  if (!is_integral_type (value_type (arg2)))
-    error (_("Argument to arithmetic operation not a number or boolean."));

With the patch applied, due to those eval.c changes:

-      else if (ptrmath_type_p (value_type (arg1)))
-       return value_ptradd (arg1, arg2);
-      else if (ptrmath_type_p (value_type (arg2)))
-       return value_ptradd (arg2, arg1);
+      else if (ptrmath_type_p (value_type (arg1))
+              && is_integral_type (value_type (arg2)))
+       return value_ptradd (arg1, value_as_long (arg2));
+      else if (ptrmath_type_p (value_type (arg2))
+              && is_integral_type (value_type (arg1)))
+       return value_ptradd (arg2, value_as_long (arg1));
       else
        {
          binop_promote (exp->language_defn, exp->gdbarch, &arg1, &arg2);

control will fall through to the default value_binop code, and *this*
will now issue the error message:

  if ((TYPE_CODE (type1) != TYPE_CODE_FLT
       && TYPE_CODE (type1) != TYPE_CODE_DECFLOAT
       && !is_integral_type (type1))
      || (TYPE_CODE (type2) != TYPE_CODE_FLT
          && TYPE_CODE (type2) != TYPE_CODE_DECFLOAT
          && !is_integral_type (type2)))
    error (_("Argument to arithmetic operation not a number or boolean."));

(Note that when I originally introduced value_ptradd I duplicated this
message in the first place.)

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com