[Bug fortran/46678] [4.6 Regression] Wrong code with strings

[burnus at gcc dot gnu.org]

http://gcc.gnu.org/bugzilla/show_bug.cgi?id=46678

--- Comment #5 from Tobias Burnus <burnus at gcc dot gnu.org> 2010-11-27 
09:54:36 UTC ---
(In reply to comment #4)
> While it makes a bit more sense, I admittedly still fail to understand why
>   int .string
>   {
>      .string = 4
>   }
>   // use .string
> fails

'cause I cannot read. One has:

sub (integer(kind=4)[0:D.1579] * restrict a, integer(kind=4) & restrict n)
{
  character(kind=1) string[1:.string];
  integer(kind=4) .string;
  {
    .string = MAX_EXPR <...>
    character(kind=1) string[1:.string];
    // working
  }
  // failing
}

which makes sense but indicates that we somehow mishandle variable-length
strings; in particular, it indicates that it won't work that way with
allocatable string lengths.