Hi consultants:

The following small test code stops on lightning with the following
runtime error (after compiling with pgf90):

ln0126en% a.out
deallocating gethist%lev..
0: DEALLOCATE: memory at bfffed18 not allocated
ln0126en% 


!----------------------------------------------------------------------
      module hist
      implicit none
      type history
        integer :: nlev
        real,pointer :: lev(:)
      end type history
      end module hist
!----------------------------------------------------------------------
      module get_hist
      use hist,only: history
      implicit none
      contains
!
      type(history) function gethist(nlev)
      integer,intent(in) :: nlev
      integer :: ier

      gethist%nlev = nlev
      if (associated(gethist%lev)) then
        write(6,"('deallocating gethist%lev..')")
        deallocate(gethist%lev)
      endif
      write(6,"('allocating gethist%lev: gethist%nlev=',i3)") 
     |  gethist%nlev
      allocate(gethist%lev(gethist%nlev),stat=ier)
      write(6,"('after allocate gethist%lev: gethist%nlev=',i3,
     |  ' ier=',i3)") gethist%nlev,ier

      end function gethist
      end module get_hist
!----------------------------------------------------------------------
      program testalloc
      use hist,only: history
      use get_hist,only: gethist
      implicit none
      type(history) :: h
      integer :: nlev
!
      nlev = 100
      h = gethist(nlev)
      end program testalloc
!----------------------------------------------------------------------


So the associated() intrinsic function is returning true (even tho
the pointer has not yet been associated), but then the deallocate 
statement fails saying the memory wasnt allocated after all.

Now if I replace function gethist with a subroutine, and pass
h as an argument, it works.  But this is a distillation of a
much larger code (a serial post-processor), and it would take
quite a bit of work to change it, and I don't see why it should
fail as is anyway. It runs fine on xlf90 bs and bv, and f90 on
tempest. It also fails on our local HAO linux boxes, using a
slightly different version of pgf90. 

Do you see any other way around this beside replacing the
function with a subroutine?  And do you think this warrants
a bug report to the Portland Group?  Thanks,

--Ben