[Bug fortran/84615] [8 Regression] Executable Segfault for some tests compiled with -m32 after r256284

[kargl at gcc dot gnu.org]

https://gcc.gnu.org/bugzilla/show_bug.cgi?id=84615

kargl at gcc dot gnu.org changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |kargl at gcc dot gnu.org

--- Comment #15 from kargl at gcc dot gnu.org ---
(In reply to Janne Blomqvist from comment #12)
> So to be clear, the problem seems to be that while the code generation for
> the function itself appears Ok, it generates the interface wrong (i.e. uses
> the declared charlen kind instead of C_SIZE_T), and thus calling it doesn't
> work properly.
> 
> I still have no idea where in the frontend code the error is.

I haven't looked, but suspect that you can force a
conversion to the correct charlen in

trans-intrinsic.c:gfc_conv_intrinsic_function_args

Well, for intrinsics, the above function has

      if (e->ts.type == BT_CHARACTER)
        {
          gfc_conv_expr (&argse, e);
          gfc_conv_string_parameter (&argse);
          argarray[curr_arg++] = argse.string_length;
          gcc_assert (curr_arg < nargs);
        }

From trans-expr.c, one has

/* Makes sure se is suitable for passing as a function string parameter.  */
/* TODO: Need to check all callers of this function.  It may be abused.  */

void
gfc_conv_string_parameter (gfc_se * se)

I suspect you can do a fold_convert here or a conversion
before the gfc_conv_expr.