J3/14-126
To: J3
From: Van Snyder
Subject: UTI concerning undefinition
Date: 2014 February 12
Reference: 14-106

14-106 observed that the interp concerning undefinition that resulted in
the current text at 16.6.6p1(1) appears not to be completely correct.

[14-007:461:41+] Insert a UTI

  This still isn't quite right.

  16.6.6p1(1) doesn't cover the case of, say, a default integer
  associated with half of a double precision variable.  If the integer
  becomes defined, the double precision variable ought to become
  undefined.  The first sentence in 16.6.6p1(1) only covers the
  totally-associated case.

  The second sentence of 16.6.6p1(1) doesn't allow the case of a double
  precision pointer or associate name that is totally associated with a
  double precision subobject of another object, but only partially
  associated with the object.  Consider, for example:

    double precision, pointer :: D
    complex(kind(0.0d0)), target :: X = ( 1.0, 2.0 )
    d => x%re
    d = 42.0d0

  16.6.6p1(1) says this results in X becoming undefined.

  This is clearly not what we want.  The same problem occurs with
  derived-type objects.