Robin Vowels writes:
 > > Date:          Fri, 18 Oct 2002 17:09:46 -0400
 > > From:          Tom Clune <[log in to unmask]>
 > > To:            [log in to unmask]
 >
 > > According to the Fortran standard, how should host-accessible variables
 > > be treated in a recursive routine?  I would have thought that only a single
 > > instance would be created, but due to a bug, I encountered a case where
 > > at least one compiler appears to have created a separate instance at each level.
 >

 > Are you sure?
 > The codes that you give contradict your description.
 > The actual (long) code has "i" defined in the recursive subroutine "SubInit".
 > The wrapping subroutine "ut_tree_iter" not define "i", and your comment
 > says that you accidentally omitted "i" from this routine.

Apologies - my quick stripped down version was not quite complete, and
in my hasty recreation of the problem in the "real" code, I commented
out the wrong declaration of "i".

I include here a full program which demonstrates the problem by
getting different answers on different compilers.  Note that I am now
also making reference to an integer "total" that is also only declared
in main, as a simple mechanism to demonstrate the differing behavior.

!---------------------------------------
Program Main
  Implicit None

  ! These are only declared in main, but used in
  ! the recursive routine sub_init() as well.
  Integer :: i
  Integer :: total

  Integer, Parameter :: N_CHILDREN = 4
  Integer, Parameter :: MAX_DEPTH  = 3

  total = 0
  Call sub_init(1, 0)
  Print*, total

Contains

  Recursive Subroutine sub_init(id, depth)
    Implicit None
    Integer, Intent(In) :: id
    Integer, Intent(In) :: depth

    Integer :: ch_id
    ! Integer "i" is not declared here, but in main instead

    Do i = 1, N_CHILDREN
       If (depth < MAX_DEPTH) Then
          ch_id = N_CHILDREN*id + i
          Call sub_init(ch_id, depth + 1)
          total = total + ch_id
       End If
    End Do

  End Subroutine sub_init

End Program Main
!---------------------------------------

Under absoft:
% ./a.out
   7938

Under Compaq Fortran V5.5-2602
% ./a.out
       12993

Under MIPSpro 7.3.1.1m
% ./a.out
 111

Under Cray CF90 3.5.0.4
# default optimization
% ./a.out
 26
# -O0 optimization gives same as absoft
 7938

From these results I infer that the code is, in fact, not standard
conforming, and I should have declared the integer "i" in routine
sub_init().  In fact, once I have done so, I get the same answer
(7938) under all of the above compilers.  (Well, the default
options for Cray Cf90 still give 26, but I suspect that a compiler
bug is involved here.)  Apparentl, I was just "unfortunate" enough
to be developing under the one compiler that gave the intended
answer even with the illegal code.

Cheers,

- Tom

--

--
Thomas Clune, Ph.D.                   <[log in to unmask]>
Advanced Software Technology Group    Work: 301-286-4635
Science Computing Branch 931          Fax:  301-286-1634
NASA Goddard Space Flight Center
Greenbelt, MD 20771