https://gcc.gnu.org/bugzilla/show_bug.cgi?id=128833
Bug ID: 128833
Summary: Segfault in intrinsic assignment to a scalar pointer
when a component has a type-bound defined assignment
Product: gcc
Version: 16.2.1
Status: UNCONFIRMED
Severity: normal
Priority: P3
Component: fortran
Assignee: unassigned at gcc dot gnu.org
Reporter: jvdelisle at gcc dot gnu.org
Target Milestone: ---
This one shared to me on MatterMost. We have a patch.
This one segfaults:
module m
implicit none
type :: c
integer :: i = 0
contains
final :: c_final
procedure :: c_assign
generic :: assignment(=) => c_assign
end type c
type :: t
type(c) :: comp
integer, pointer :: p => null()
end type t
contains
subroutine c_final (self)
type(c), intent(inout) :: self
end subroutine c_final
subroutine c_assign (to, from)
class(c), intent(out) :: to
class(c), intent(in) :: from
to%i = from%i
end subroutine c_assign
end module m
program p
use m
implicit none
type(t), pointer :: q
allocate (q)
q = t (c (1))
if (q%comp%i /= 1) stop 1
end program p
$ gfortran -g pr.f90 && ./a.out
Program received signal SIGSEGV: Segmentation fault - invalid memory
reference.
... at pr.f90:30 (the assignment q = t (c (1)))
Each of these makes it pass: an ALLOCATABLE instead of POINTER q; no
pointer component in t; no FINAL procedure for c. Without the FINAL
procedure it still fails if the first argument of c_assign is
INTENT(INOUT) instead of INTENT(OUT).
Cause: generate_component_assignments in resolve.cc copies the lhs into
a temporary when the defined assignment's first argument is INTENT(INOUT),
or is INTENT(OUT) and finalizable. get_temp_from_expr gives that
temporary the attributes of the lhs, so for a pointer q it is a pointer
(-fdump-tree-original shows "struct t * _F.DA3;") that is never
allocated, and the copy writes through a null pointer.