J3/06-292

Date:    24 August 2006
To:      J3
From:    Richard Maine
Subject: Misleading text in Note 15.17

Note 15.17 in F2003 is misleading, as described below. I propose
that this be fixed in F2008.

The note says that polymorphic, allocatable, or pointer arrays are
never interoperable. This much is accurate.

But then it says that this is because such arrays are not explicit
shape or assumed size. While that is true for allocatable and
pointer, I can't find any justification for it for polymorphic
arrays. A polymorphic variable must be an allocatable, a pointer or a
dummy. I see no restriction against a polymorphic dummy being
explicit shape. (I'm not sure whether one can be assumed-size or not,
but I didn't look very hard at that because one counter-example
seemed sufficient).

I don't think that a polymorphic array can be interoperable, but the
reason has nothing to do with explicit shape or assumed size. I think
the reason is the same one that applies to scalar polymorphic
variables - an extensible type cannot be interoperable (see the
definition of an extensible type, which excludes BIND(C) types).

I propose the following edits:

In Note 15.17 in subclause 15.2.5 on pg 399 of F2003,
replace the Note with

  "An allocatable or pointer array is never interoperable. Such
   an array does not meet the requirement of being explicit shape
   or assumed size."

Add a new Note at the end of subclause 15.2.4, at [399:5+]

  "A polymorphic variable is never interoperable because
   extensible types are not interoperable types."

I made the polymorphic part a separate note in the subclause
on scalar variables because it has nothing in particular to
do with arrayness.

P.S. As a separate but related matter, I'm slightly concerned
that the standard might allow a dummy class(*) argument to be
considered interoperable if its dynamic type is an interoperable
intrinsic one. That is the one way I can think of that the
dynamic type of a polymorphic variable can be an interoperable
type. I tend to think of interoperability as being a static
matter, but our usual convention is that "type" means "dynamic
type" if not specified otherwise. Perhaps we should specify
otherwise in the definitions of interoperability of variables. I
am not sure whether this is a real problem, but J3 might want to
consider the question.