To: J3                                                          11-246
From: Van Snyder
Subject: PDTR 29113 Lifetimes and Pointers
Date: 2011 October 11
Reference: WG5/N1869

Discussion
----------

In PDTR 29113, paragraph 2 of subclause 5.3.7, one sees that if a dummy
argument does not have the TARGET, ASYNCHRONOUS, or VOLATILE attributes,
all C pointers to any part of the object its C descriptor describes
become undefined on return from the call.

The reason for ASYNCHRONOUS to be in the list is that Rolf Rabenseifner
objected to the possibility of performance degradation caused by the
TARGET attribute.  If the criterion were based on the TARGET attribute
alone, paragraphs 12.5.2.4p8-12 of 1539-1 would have covered all the
cases.

There appears to be no reason for VOLATILE to be in the list.

In PDTR 29113, paragraph 3 of subclause 5.3.7, one sees the sentence "A
Fortran pointer variable that is associated with the object described by
a C descriptor shall not be accessed beyond the end of the lifetime of
the C descriptor and the object it describes."

There are at least three problems with this sentence.

1.  It applies to all Fortran pointers that ever become associated with
an object described by a C descriptor, no matter when or how.

2.  The prohibition against accessing the Fortran pointer after the
lifetime of the C descriptor is pointless, since the pointer is
associated with the object the C descriptor describes, not with the C
descriptor.  Even if the C function has taken a copy to get a contiguous
object, it's still the object described by the C descriptor that is
under discussion, not a copy of it.  If the caller has taken a copy to
get a contiguous object, paragraphs 12.5.2.4p8-12 govern the pointer's
association status.

3.  In 1539-1, the word "accessed" is used with respect to coarrays, but
more generally applies to accessibility by host or use association.  It
is not used in relation to pointers.  Taken literally, it could be
interpreted to mean that the pointer could not be the <pointer-object>
in a pointer assignment statement.

The first and third of these are definitely mistakes in the construction
of subclause 5.3.7.  The second is covered by 12.5.2.4p8-12 and
16.5.2.4-5 of 1539-1.

There is no discussion of the effect of the lifetime of an ordinary C
object that is the target of a Fortran pointer.

Edits
-----

In the second paragraph of subclause 5.3.7, replace "TARGET,
ASYNCHRONOUS, or VOLATILE" by "TARGET or ASYNCHRONOUS.

In the third paragraph of subclause 5.3.7, remove the final sentence "A
Fortran pointer ... it describes."

After the fourth paragraph of subclause 5.3.7, insert a paragraph

"If a Fortran pointer becomes associated with a C object, the
association status of the Fortran pointer becomes undefined when the
lifetime of the C object ends."

Before subclause 6.10, add a subclause "6.9a Edits to clause 16"  Within
that subclause add the edit to 1539-1

"After 16.5.2.4p1(3) insert a list item
  '(3a) the target of the pointer is a C object and the lifetime of the
        C object ends,'
"