J3/15-179r1
To: J3
From: Malcolm Cohen
Subject: Interp on statement functions returning PDTs
Date: 2015 August 05

----------------------------------------------------------------------

NUMBER: F08/0141
TITLE: Can a statement function have a variable-length PDT result?
KEYWORDS: Statement function, parameterized derived type
DEFECT TYPE: Erratum
STATUS: J3 consideration in progress

QUESTION:

Consider

    MODULE m207c006
      TYPE string(n)
        INTEGER,LEN :: n
        CHARACTER(n) :: v
      END TYPE
    END MODULE
    PROGRAM test
      USE m207c006
      CALL s(3)
      CALL s(7)
    CONTAINS
      SUBROUTINE s(n)
        TYPE(string(n)) sf
        TYPE(string(n)) var(3)
        sf(nn) = var(nn)
        var(1) = string(n)('123456789')
        var(2) = string(n)('abcdefgh')
        PRINT *,sf(1),sf(2)
      END SUBROUTINE
    END PROGRAM

If this is conforming, it would appear that the output ought to be
  123abc
  1234567abcdefg

However, for the following reasons, perhaps this is not intended to be
conforming.

(1) A statement function of type character is required to have a
    constant length type parameter.  There is no such requirement on a
    statement function of parameterized derived type.  This seems
    inconsistent, since one can wrap a variable-length character
    in a parameterized derived type.

(2) A statement function is not permitted to invoke a nonintrinsic
    function that requires an explicit interface (and having a PDT
    result requires an explicit interface), nor is it permitted to
    contain a structure constructor.  This would seem to render such a
    statement function almost completely useless, since all it can do
    is to choose between PDT constants or variables, or use an
    intrinsic function such as TRANSFER to construct the result.

(3) Statement functions were declared to be obsolescent long before
    the addition of parameterized derived types to the standard.  It
    has been general policy not to "improve" obsolescent features.
    This would seem to indicate that statement functions were not
    intended to be capable of having parameterized derived type in the
    first place.

Q1. Are statement functions of parameterized derived type intended to
    be conforming?

Q2. If so, are statement functions of parameterized derived type with
    nonconstant length type parameters intended to be conforming?

Q3. If so, is the expected output from the example program above as
    described?

ANSWER:

A1. No, statement functions were not intended to be permitted to be of
    parameterized derived type.  The exception for statement functions
    from the explicit interface requirements was overlooked.  An edit
    is provided to prohibit these.

A2, A3.  Moot.

EDITS:

[22:11+] 1.6.2 Fortran 2003 compatibility, insert new paragraph
  "Fortran 2003 permitted a statement function to be of parameterized
   derived type; this part of ISO/IEC 1539-1 does not permit that."
{Previous corrigenda have already inserted incompatibility
 paragraphs.}

[311:34+] 12.6.4 Statement function, after C1275, insert constraint
  "C1275a A statement function shall not be of a parameterized derived
          type."

SUBMITTED BY: Malcolm Cohen

HISTORY: 15-179    m207  F08/0141  Submitted
         15-179r1  m207  Revised edits.

----------------------------------------------------------------------