Hello world,
the attached patch fixes a 7/8/9 regression where dependency checking
was for class arrays and a scalar value was mishandled when the
dependency happened in an elemental function.
There was an ICE for the test case which is handled by
fixing up the class refs in gfc_walk_variable_expr.
Once this was gone, a wrong-code issue appeared which was fixed
by the part in gfc_scalar_elemental_arg_saved_as_reference
(is that the longest function name in gfortran?).
Regression-tested. OK for all affected branches?
Regards
Thomas
2019-03-06 Thomas Koenig
PR fortran/66089
* trans-array.c (gfc_scalar_elemental_arg_saved_as_reference):
Return false if a scalar tempoary is needed.
(gfc_walk_variable_expr): Fix up class refs.
2019-03-06 Thomas Koenig
PR fortran/66089
* gfortran.dg/dependency_53.f90: New test.
Index: trans-array.c
===
--- trans-array.c (Revision 269260)
+++ trans-array.c (Arbeitskopie)
@@ -2699,6 +2699,9 @@ gfc_scalar_elemental_arg_saved_as_reference (gfc_s
if (ss_info->type != GFC_SS_REFERENCE)
return false;
+ if (ss_info->data.scalar.needs_temporary)
+return false;
+
/* If the actual argument can be absent (in other words, it can
be a NULL reference), don't try to evaluate it; pass instead
the reference directly. */
@@ -10515,6 +10518,8 @@ gfc_walk_variable_expr (gfc_ss * ss, gfc_expr * ex
{
gfc_ref *ref;
+ gfc_fix_class_refs (expr);
+
for (ref = expr->ref; ref; ref = ref->next)
if (ref->type == REF_ARRAY && ref->u.ar.type != AR_ELEMENT)
break;
! { dg-do run }
! PR fortran/66089 - used to ICE and, after that ICE was fixed,
! gave wrong results.
type :: t
integer :: c
end type t
class(t), dimension(:), allocatable :: b,c
allocate (b(5), source=t(7))
allocate(c(5), source=t(13))
c = plus(c(1), b)
if (any(c%c /= 20)) call abort
c = t(13)
c = plus(b, c(1))
if (any(c%c /= 20)) call abort
contains
elemental function plus(lhs, rhs)
class(t), intent(in) :: lhs, rhs
type(t) :: plus
plus%c = lhs%c + rhs%c
end function plus
end
