From martin@anduin.ocf.llnl.gov Wed Jul 3 22:56:14 1996 Received: from anduin.ocf.llnl.gov (anduin.ocf.llnl.gov [134.9.49.1]) by dkuug.dk (8.6.12/8.6.12) with SMTP id WAA08419 for ; Wed, 3 Jul 1996 22:56:12 +0200 Received: by anduin.ocf.llnl.gov (4.1/SMI-4.0) id AA24674; Wed, 3 Jul 96 13:42:00 PDT Date: Wed, 3 Jul 96 13:42:00 PDT From: martin@anduin.ocf.llnl.gov (Jeanne T Martin) Message-Id: <9607032042.AA24674@anduin.ocf.llnl.gov> To: sc22wg5@dkuug.dk Subject: Allocatable Components - User Input Here is a message I received from some "real" Fortran 90 users at LLNL. ______________________________________________________________________________ > From: "Gayle Sugiyama" > Date: Wed, 3 Jul 1996 09:06:42 -0700 > To: jtm@llnl.gov > Subject: F90 standards : pointers inside derived type > Cc: jleone@anvil.llnl.gov > > I know that officially LLNL has no representative on the F90 standards > board/committee, but I am hoping that you will be able to pass on > this input from the ARAC F90 users (we took a course from you last > year). > We have been working with derived types containing allocatable > pointers. In some cases the ONLY reason we are using pointers is that > allocatable arrays are not permitted in a derived type. This leads > to what we consider inconsistent behavior of the = versus the => > assignments. As stated in the Fortran Handbook p276, rule 14 > "If a structure has a component with the POINTER attribute and the > structure is assigned a value using an instrinsic derived-type assignment, > POINTER ASSIGNMENT is used for each component with the POINTER > attibute." From our perspective this means that when you do a > derived_type1 = derived_type2 > assignment the pointer array inside the derived type is not > copied but acts like a => assignment instead. This has been very > confusing to us since if we wanted that we would have used > derived_type1 => derived_type2 > instead. To get a copy of derived_type2, we must go in by hand and > assign each of the elements of the derived type by hand > derived_type1%pointer_array = derived_type2%pointer_array > derived_type1%real_var = derived_type2%real_var > and so on. > In fact overall we think that it would have been nicer if > allocatable arrays and pointers would more clearly distinct - > i.e. arrays could be allocatable/deallocatable and pointers > associated/nullified. There are probably good reasons why this > wasn't done and to insure backwards compatability we presume > this won't be changed, so our suggestion is to add allocatable > arrays as an allowed element in derived types for which > derived type assignments would work like = and not =>. > > --- Gayle Sugiyama and John Leone > > A simple test code example of the above written by JL : > > module met_definition > > implicit none > > type met_data > real, pointer, dimension(:,:,:) :: val > real :: time > character(len = 20) :: met_str > end type met_data > > end module met_definition > > > program test_ta > > use met_definition > > implicit none > > type (met_data), allocatable, dimension(:, :) :: u > > integer :: imax = 3, jmax = 1, kmax = 1, num_data = 2, & > i, j, k, nd > > allocate(u(1, num_data)) > > do nd = 1, num_data > allocate(u(1, nd)%val(imax, jmax, kmax)) > u(1, nd)%val = nd > u(1, nd)%time = nd * 100. > write(u(1, nd)%met_str, '(a, i2.2)') 'u data set ', nd > end do > > write (*, *) 'PART I' > write (*, *) 'u data types after allocation and initial assignment' > do nd = 1, num_data > write (*, '(/, a, i2.2)') 'data set ', nd > write (*, *) u(1, nd)%met_str, u(1, nd)%time > write (*, *) u(1, nd)%val > enddo > > u(1, 1) = u(1, 2) > > write (*, '(///,a)') 'PART II' > write (*, *) 'u data types after setting u(1, 1) = u(1, 2)' > do nd = 1, num_data > write (*, '(/, a, i2.2)') 'data set ', nd > write (*, *) u(1, nd)%met_str, u(1, nd)%time > write (*, *) u(1, nd)%val > enddo > > u(1, 2)%val = 3 > u(1, 2)%time = 3 * 100. > write(u(1, 2)%met_str, '(a, i2.2)') 'u data set ', 3 > > write (*, '(///,a)') 'PART III' > write (*, *) 'u data types after reassigning u(1, 2) but not u(1, 1)' > do nd = 1, num_data > write (*, '(/, a, i2.2)') 'data set ', nd > write (*, *) u(1, nd)%met_str, u(1, nd)%time > write (*, *) u(1, nd)%val > enddo > > end program > > -- > ------------------------------------------------------------------------------ > Gayle Sugiyama > sugiyama@llnl.gov > L-103 > Lawrence Livermore National Lab > Livermore, CA 94550 > (510)422-7266 > ------------------------------------------------------------------------------ > Jeanne Martin email: jtm@llnl.gov phone: (510) 422-3753 postal code: L-561 fax: (510) 423-8704 Lawrence Livermore National Laboratory Livermore, CA 94550