elemental optr. between abstract arrays (catastrophic)

[Ferdinand_T_]

Hello,

here's an other catastrophic compiler error report (version 14.0.1); maybe it helps to improve ifort!

The error is caused by invoking an elemental binary operator (as '==') on polymorphic arrays of abstract declared type.

[fortran]

module m
    implicit none
 
    ! abstract parent type with elemental equality operation
    type, abstract :: abstr_parent_type
    contains
        procedure(is_equal_interf), deferred :: is_equal
        generic :: operator(==) => is_equal
    end type
 
    abstract interface
        elemental function is_equal_interf(this, other)
            import abstr_parent_type
            class(abstr_parent_type), intent(in) :: this, other
            logical :: is_equal_interf
        end function
    end interface
 
    ! extended type overrides abstract equality operation
    type, extends(abstr_parent_type) :: child_type
        integer :: field = 0
    contains
        procedure :: is_equal => child_type_is_equal
    end type
 
contains
    elemental function child_type_is_equal(this, other)
        class(child_type), intent(in) :: this
        class(abstr_parent_type), intent(in) ::  other
        logical :: child_type_is_equal
 
        ! any comparision code
        child_type_is_equal = .true.
    end function
end module
 
program t
    use m
    implicit none
 
    ! must be array(!) of abstract(!) parent in order to crash
    class(abstr_parent_type), allocatable, dimension(:) :: arr1, arr2
 
    ! run elementwise comparisions on child_types
 
    allocate(child_type :: arr1(1), arr2(1))
 
    ! variant 1: catastrophic error
    if (all(arr1 == arr2)) then                          ! CRASH
        print *, 'arrays are equal'
    end if
 
    ! variant 2: no error (legal?)
    if (all(arr1%is_equal(arr2))) then
        print *, 'arrays are equal'                      ! works
    end if
 
end program
[/fortran]

COMPILE LINE:
ifort elementwise_type_bound.f90

OUTPUT:
catastrophic error: **Internal compiler error: internal abort** Please report this error along with the circumstances in which it occurred in a Software Problem Report.  Note: File and line given may not be explicit cause of this error.
compilation aborted for elementwise_type_bound.f90 (code 1)

WORKAROUND QUESTION:
Is variant 2 (with the elemental type bound function invocation directly on the array) a standard conforming F2003 alternative?

Thanks,
Ferdinand

[Steven_L_Intel1]

Thanks - I will send this off to the developers. There's nothing wrong with calling the function directly - indeed, the use of the operator ought to translate to a call to the function.

[Ferdinand_T_]

Steve Lionel (Intel) wrote:

There's nothing wrong with calling the function directly - indeed, the use of the operator ought to translate to a call to the function.

It's just that I never called a type bound function from an arrayvariable before, e.g. 'arr%func()' instead of 'array(index)%func()' and I am glad to hear that this is legal syntax. Thank you!

[Steven_L_Intel1]

With an elemental function, the syntax you chose passes an array which is fine.

Issue ID is DPD200249693.

[Ferdinand_T_]

While writing some code with that nice 'elemental function on array invokation' syntax I run into a probably related issue (catastrophic compiler abort on type bound elemental generic, when called from array) which I managed to reduce to the following program:

[fortran]

module m
    implicit none
 
    ! type bound elemental generic
    type :: derived_type
    contains
        ! generic interface for type bound elemental (pass or nopass)
        procedure, nopass :: do_specific
        generic :: do_generic => do_specific
    end type
 
contains
    elemental subroutine do_specific()
    end subroutine
end module
 
program p
    use m
    implicit none
 
    ! must be array(!) of derived types in order to crash
    type(derived_type), dimension(1) :: my_derived
 
    call my_derived%do_specific()         ! works
    call my_derived%do_generic()          ! CRASH
end program

[/fortran]

In the hope posting those issues here is helpful and with best regards,
Ferdinand

[Steven_L_Intel1]

Yes, your posting these is very helpful! Thanks.