Access violation in assignment(=)

[Wolf_W_]

Hello,

i got an error with the XE17.1 compiler. The following code produces two different crashes in line 35 (coming from 68), dependend on the type of "vec" in "T_VECTOR". The code worked with the XE16.X compilers.

module M_T_VECTOR
  implicit none

  type :: T_VECTOR
!forrtl: severe (408): fort: (7): Attempt to use pointer ASSIGN_SOURCE when it is not associated with a target
    real :: vec(1:3) = [0.0,0.0,0.0]

!forrtl: severe (157): Program Exception - access violation
!    doubleprecision :: vec(1:3) = [0.0,0.0,0.0]
  contains

    procedure :: assignTo
    generic :: assignment(=) => assignTo
    procedure ::add
    generic :: operator(+) => add

  end type T_VECTOR

  type, extends(T_VECTOR) :: T_VECTOR_EXT

    type(T_VECTOR),  allocatable :: disp(:)
    class(T_VECTOR_EXT), pointer :: pt_a    => null()
    class(T_VECTOR_EXT), pointer :: pt_b(:) => null()

  end type T_VECTOR_EXT

!--------------------------------------------------------
contains

  elemental subroutine assignTo(assign_target, assign_source)
    class(T_VECTOR), intent(inout) :: assign_target
    class(T_VECTOR), intent(in   ) :: assign_source

! Exception thrown here
    assign_target%vec = assign_source%vec

  end subroutine

  elemental function add(lhs, rhs) result(ergebnis)
    class(T_VECTOR), intent(in   ) :: lhs, rhs
    type(T_VECTOR)                 :: ergebnis

    ergebnis%vec = lhs%vec + rhs%vec
  end function

end module M_T_VECTOR

!========================================================

program main
  use M_T_VECTOR
  implicit none

  type(T_VECTOR_EXT), target :: a,b(2)

  a%pt_b => b
  a%pt_a => a

  allocate(a%disp(2))
  allocate(b(1)%disp(2))
  allocate(b(2)%disp(2))

  !This works
  a%pt_a%disp(1) = a%pt_a%disp(1) + a%pt_b(1)%disp(1)
  a%pt_a%disp(2) = a%pt_a%disp(2) + a%pt_b(1)%disp(2)

  !This crashes
  a%pt_a%disp(:) = a%pt_a%disp(:) + a%pt_b(1)%disp(:)

end program

Greetings

Wolf

[jimdempseyatthecove]

Try adding ", target" to lines 6 and 21

Jim Dempsey

[Wolf_W_]

Thanks for your advice, but that gives me the compile-time error:

error #6516: This attribute specification is not valid for a component definition statement.   [TARGET]

Wolf

[Kevin_D_Intel]

Thank you for the concise test case. I reproduced the access violation for both data types noted in the test case and escalated this to Development (internal tracking id noted below).

I was unable to reproduce the noted "forrtl: severe (408): fort: (7)" failure on Linux and I do not have 17.0 update 1 available for Windows at the moment. This appears to be a regression with the latest 17.0 update 1. I confirmed as you noted the test case compiles/runs successful with previous 16.0 compilers, and I also found it runs successfully when using the initial 17.0 release (Version 17.0.0.109 Build 20160721) on both Linux and Windows. I will let you know if I learn about any source code work around for this defect.

(Internal tracking id: DPD200416395)

[Wolf_W_]

Thank you,

it can be worked around with a loop.

Wolf

[Kevin_D_Intel]

Ok, thank you for sharing that.

[FortranFan]

Kevin D. (Intel) wrote:

.. I will let you know if I learn about any source code work around for this defect. ..

@Kevin D. (Intel),

You may want to run the following with your development and see if it helps any in terms of diagnosing the defect with compiler 17.  I'm pressed for time to go into my observations of Intel Fortran and interpretations of the standard, but I've run into issues previously with coders in our team defining generic assignments with a polymorphic right-hand side and often it leads to ambiguities too.  See the effect of changing the declaration of the second parameter in the assignTo procedure below.

module M_T_VECTOR
   
   implicit none

   type :: T_VECTOR
   
      real :: vec(1:3) = [0.0,0.0,0.0]

   contains

      procedure :: assignTo
      generic :: assignment(=) => assignTo
      procedure ::add
      generic :: operator(+) => add

   end type T_VECTOR

   type, extends(T_VECTOR) :: T_VECTOR_EXT

      type(T_VECTOR),  allocatable :: disp(:)
      class(T_VECTOR_EXT), pointer :: pt_a    => null()
      class(T_VECTOR_EXT), pointer :: pt_b(:) => null()

   end type T_VECTOR_EXT

   !--------------------------------------------------------
contains

   elemental subroutine assignTo(assign_target, assign_source)

      class(T_VECTOR), intent(inout) :: assign_target
      type(T_VECTOR),  intent(in   ) :: assign_source  !<-- see change here

      ! Exception thrown here
      assign_target%vec = assign_source%vec

   end subroutine

   elemental function add(lhs, rhs) result(ergebnis)

      class(T_VECTOR), intent(in   ) :: lhs, rhs
      type(T_VECTOR)                 :: ergebnis

      ergebnis%vec = lhs%vec + rhs%vec

   end function

end module M_T_VECTOR

program main

   use M_T_VECTOR, only : T_VECTOR_EXT

   implicit none

   type(T_VECTOR_EXT), target :: a,b(2)

   a%pt_b => b
   a%pt_a => a

   allocate(a%disp(2))
   allocate(b(1)%disp(2))
   allocate(b(2)%disp(2))

   b(1)%disp(1)%vec = 1.0
   b(1)%disp(2)%vec = 2.0

   a%pt_a%disp(:) = a%pt_a%disp(:) + a%pt_b(1)%disp(:)

   print *, "a%disp(1)%vec = ", a%disp(1)%vec, " (expected values are all 1.0)"
   print *, "a%disp(2)%vec = ", a%disp(2)%vec, " (expected values are all 2.0)"

   stop

end program

Upon execution with compiler 17, update 1:

 a%disp(1)%vec =  1.000000 1.000000 1.000000  (expected values are all 1.0)
 a%disp(2)%vec =  2.000000 2.000000 2.000000  (expected values are all 2.0)

[Kevin_D_Intel]

Thank you FortranFan. I shared your update with them. Yours produces unexpected results w/17.0 initial release and expected results w/17.0 update 1.