[EXTERNAL] Re: [Patch, F03, RFC] FINAL support in 4.8

[Rouson, Damian]

Janus,

Everything you stated below about the behavior of the IBM, NAG, and Intel
compilers is correct.  The version of Intel's compiler I just tested is
12.1.5.339.  I should be able to get access to the Intel's 13.0 release
within a few days.

Damian

On 10/29/12 1:38 AM, "Janus Weil" <janus@gcc.gnu.org> wrote:

>Hi Damian,
>
>> My apologies.  The code I sent earlier was wrong in two places (the
>> default initialization and the "merge" call).  The corrected version
>>below
>> indicates that NAG (and IBM) finalize the new_child() function result
>>but
>> not "infant".  This makes more sense relative to the standard. Item 1 of
>> section 6.7.1.3 of the Fortran 2008 standard suggests that allocating a
>> variable that already has the "allocated" status causes an error
>> condition. If the variable does not already have the "allocated" status
>> (as is the case in this example), then finalizing it makes no sense to
>>me.
>
>yes, concerning the ALLOCATE statement, it makes sense that the result
>of "new_child" should be finalized, and not the "infant" variable.
>Unfortunately our patch currently gets this wrong and finalizes the
>"infant" variable instead (when copying the data from "new_child"
>through the _copy procedure).
>
>In addition, I was wondering whether a finalization of 'infant' at the
>end of the main program should occur (which is what gfortran currently
>does, as a consequence of infant's automatic deallocation, which may
>or may not be what the standard demands).
>
>In summary, for the following program (together with the modules from
>Damian's last test case):
>
>program main
>  use child_module
>  implicit none
>  class(child), allocatable :: infant
>  print *,"AAA"
>  allocate(infant,source=new_child())
>  print *,"BBB"
>end
>
>gfortran currently prints:
>
> AAA
> subroutine finalize called on infant
> BBB
> subroutine finalize called on new_child() result
>
>[The last line does not literally mean that a new_child() result is
>finalized. Instead it comes from the finalization of 'infant' at the
>end of the main program (after it got its value from new_child).]
>
>ifort 12.1.2 segfaults after printing "AAA". If I understand Damian's
>description correctly, IBM and NAG should print
>
> AAA
> subroutine finalize called on new_child() result
> BBB
>
>Cheers,
>Janus
>