[Bug fortran/30389] ACHAR() intrinsic gives erroneous errors in constant-folding.

[tkoenig at gcc dot gnu dot org]

------- Comment #3 from tkoenig at gcc dot gnu dot org  2007-01-19 21:35 -------
In order to fix this, we should know what we would prefer :-)

Constant folding maps a lot of characters (all ASCII
characters > 127, and a lot of control characters) to zero.
In the rest of the library, we treat the A* functions identically
to the non-ASCII versions. For example, we encode

      subroutine comp(a, b, t)
      implicit none
      logical t
      character(*) a, b
      t = lle(a,b)
      end

as

comp (a, b, t, _a, _b)
{
  bit_size_type D.994;
  <unnamed type> D.995;
  bit_size_type D.996;
  <unnamed type> D.997;

  D.994 = (bit_size_type) (<unnamed type>) _a * 8;
  D.995 = (<unnamed type>) _a;
  D.996 = (bit_size_type) (<unnamed type>) _b * 8;
  D.997 = (<unnamed type>) _b;
  *t = _gfortran_compare_string (_a, a, _b, b) <= 0;
}

We also encode

      program main
      integer i
      character(1) c
      i = 154
      c = achar(i)
      end

as
  i = 154;
  {
    char char.0;

    char.0 = (char) i;
    c[1]{lb: 1 sz: 1} = char.0;
  }

So, we could either

a) fix constant folding to ignore the ascii_table and xascii_table

b) fix the library and runtime conversion to honor the ascii_table.

a) is perfectly fine, as long as we document it.  I'd vote for it.

Comments?


-- 


http://gcc.gnu.org/bugzilla/show_bug.cgi?id=30389