To:        J3                                      11-173r1
From:      Bill Long/Steve Lionel
Subject:   CFI_type_other split
Date:      2011 June 28
Reference: WG5/N1854

Discussion:
-----------

The collection of types in the current type code CFI_type_other bin
includes some of substantially different nature. Also, the sign of
CFI_type_other is confusing, as some types covered could suggest a
positive value, while others clearly indicated negative.

Splitting the collection would provide greater functionality. In
particular, one of the presumed uses of the new interoperability
facility is to provide interfaces to C routines that, in effect, copy
blocks of data.  Variables of types that represent only raw "data" are
reasonable candidates for such copying, whereas types with components
that have ALLOCATABLE or POINTER attributes are not. Thus, it would
be advantageous to separate out the CFI_type_other to be only the
non-copyable types. Furthermore, a type specifier for an interoperable
derived type is a useful special case of copyable types.  By moving
Fortran intrinsic types for which there is no C counterpart (such as
128-bit real) out of the CFI_type_other collection, future expansion
of the type list should be less disruptive.

A new processor-dependent group of nonnegative specifiers for Fortran
intrinsic types that are not represented by other type codes and
Fortran derived types that do not have type parameters, type-bound
procedures, final procedures, or components that are allocatable,
pointer, or correspond to CFI_type_other is proposed.  Also,
CFI_type_struct is added to represent interoperable structs.

The value of CFI_type_other is negative.


Edits to N1854:
--------------

[11:16] In the second sentence of 5.2.4 Macros para 7, after "shall
be" insert " negative and"

[11:17] After "specifiers." add a new sentence: "CFI_type_struct
specifies a C struct that is interoperable with a Fortran derived
type; its value shall be positive and distinct from all other type
specifiers."

[11:18+] Add a new paragraph following para 7 of 5.2.4 Macros:

"Additional nonnegative processor-dependent type specifier values may
be defined for Fortran intrinsic types that are not represented by
other type specifiers and noninteroperable Fortran derived types that
do not have type parameters, type-bound procedures, final procedures,
nor components that have the ALLOCATABLE or POINTER attributes, or
correspond to CFI_type_other."


[12:Table 5.2] Before then entry for CFI_type_other, add a new line:

    "CFI_type_struct         interoperable struct"

[12:Table 5.2] Replace the C Type entry for CFI_type_other with:

   "Not otherwise specified"

[21:4+] Add a new bullet item before the last bullet:

  "if the dynamic type of the effective argument is an intrinsic type
  with no corresponding type listed in Table 5.2, or a
  noninteroperable derived type that does not have type parameters,
  type-bound procedures, final procedures nor components that have the
  ALLOCATABLE or POINTER attributes, or correspond to CFI_type_other,
  one of the processor-dependent nonnegative type specifier values;"