J3/15-239
To:         J3 Members
From:       Van Snyder
Subject:    Problem with TARGET, dummy arguments, undefinition
Date: 2015 October 09

1. Introduction
===============

Suppose one has

  type :: T
    real, allocatable :: C1(:)
    real, pointer :: P2(:,:)
  end type T

  subroutine S ( A, I, J )
    type(t), intent(out), target :: A
    allocate ( a%c1(i*j)
    a%p2(1:i,1:j) => a%c1
  end subroutine S

Since we haven't provided the TARGET attribute for components, this
requires the effective argument to have the TARGET attribute, else the
association status of the P2 component becomes undefined when S returns.

This is absurd.

I hope we don't need to go to WG5 to get permission to repair this
oversight.

Maybe it needs an interp.

2. Edits
========

[300:13 12.5.2.4p11 Ordinary dummy variables] After "completes" insert
", unless the pointer and target are components of the same derived-type
dummy argument".