Title: Notes on a Fortran Conversion, part 2 Date: 06/10/2020 Category: Programming

I found this a fascinating example of the following programmer’s mantra:

“It can’t possibly not work!” “I found it why it doesn’t work” “How did that ever work?”

subroutine set_fast(flag)

    real(kind=8) :: xmeas, xmv
    common /pv/ xmeas(42), xmv(12)

    real(kind=8), intent(in) :: time
    character(len=20), intent(in) :: flag
    logical :: init = .false.
    integer :: i
    character(len=10) :: tmp, mode = ""

    select case(flag)
        case ("--xmv") ! program
            if (.not. init) then
                call get_command_argument(2, tmp)
                read(tmp, *) i
                call get_command_argument(3, tmp)
                read(tmp, *) mode
            end if
            select case (mode)
                case ("MAX")
                    xmv(i) = 100.
!   omitted for brevity
            end select
            
    end select
    init = .true.
end subroutine

I wanted to add another case, as you do:

        case ("--xmeas") ! program
            if (.not. init) then
                call get_command_argument(2, tmp)
                read(tmp, *) i
                call get_command_argument(3, tmp)
                read(tmp, *) mode
            end if
            select case (mode)
                case ("MAX")
                    xmeas(i) = 200.
                case ("MIN")
                    xmeas(i) = -10.

            end select ! xmeas variations

And… segfault. Oh dear. Worse, only on a release build - debug handles this just fine. Did you see it? If not - spoilers ahead - it’s the index value, which is only read the first time. Shamefully, I only resolved this with print statements. Somehow, without the second case, i retained the correct value. With the second case, even if never invoked, i ended up as LARGE_INT on subsequent runs. Adding integer:: i = 0 tells the compiler to retain the value for future invocations.