Can I guarantee `allocatable` variables to start out not allocated?

Help
1

In C#, if you want to indicate that an integer may or may not be there, you can mark it nullable, like this:

int? x = null;

This is useful when you don’t have a value that’s “invalid”, like 0, that lets you know the value isn’t there. In Fortran, marking a scalar allocatable achieves a similar result

integer, allocatable :: x

I know you can’t rely on local variables to have any value when declared. Is the same true for allocatable variables? My actual use-case involves an array of custom types:

type :: my_type
  real, allocatable :: ent
  real, allocatable :: lvg
end type

subroutine s(n)
  integer, intent(in) :: n
  type(my_type), allocatable :: arr(:)

  allocate(arr(n))
  ! ...
end subroutine

Can I count on all elements of arr to have unallocated ent and lvg variables?

2

Yes, the language states that all allocatable variables and components have an initial status of unallocated.

2 Likes
3

You may find gfortran’s behavior surprising in this regard. You can bypass the surprise (allocated status upon second call) by explicitly deallocating the local allocatable objects before exiting the procedure.

4

Which behavior are you referring to?

The following

implicit none

type :: my_type
  real, allocatable :: ent
  real, allocatable :: lvg
end type

call s(5)
call s(6)

contains
    subroutine s(n)
      integer, intent(in) :: n
      integer :: i
      type(my_type), allocatable :: arr(:)

      allocate (arr(n))
      print*,'size(arr)=', size(arr)
      print*,(allocated(arr(i)%ent), allocated(arr(i)%lvg), i = 1, n)
      ! ...
    end subroutine
end

Seems to work just fine:

$ gfortran gfortran-alloc.f90 && ./a.out
 size(arr)=           5
 F F F F F F F F F F
 size(arr)=           6
 F F F F F F F F F F F F
5

This is a relevant discussion:

Maybe the new gfortran has improved.

1 Like
6

I’ve come across the following question on StackOverflow, which may be related to the unexpected behavior of allocatable by Gfortran (linked in the above post, which is still reproducible with Gfortran-14 on Godbolt). Indeed, this may be a bug of the option -fmax-stack-var-size itself, though not sure about details.

EDIT: Here is the man page for this option.

This option currently only affects local arrays declared with constant bounds, and may not apply to all character variables.

Also a previous thread on this option (not related to the above issue):

1 Like