To:         J3 Members                                       J3/17-155
From:       Van Snyder
Subject:    C856 overlooks three things, and prohibits an impossibility
References: 17-007r1
Date: 2017 May 18

1. Discussion
=============

C856 prohibits a variable with the PROTECTED attribute that is accessed
by use association from being the <data-target> in a pointer assignment
statement, but it doesn't prohibit it from being the <data-target> in a
<component-data-source> for a pointer component in a structure
constructor, or the <initial-data-target> in a variable declaration or
type definition.

C856 prohibits a nonpointer object with the PROTECTED attribute from
being the <proc-target> in a pointer assignment statement.  The only
objects that can be the <proc-target> in a pointer assignment statement
are procedures and procedure pointers, but only procedure pointers can
have the PROTECTED attribute.  There is no nonpointer object with the
PROTECTED attribute that can appear as the <proc-target> in a pointer
assignment statement.  There is no point to constrain against it.

According to C854, the only nonpointer object that can have the
PROTECTED attribute is a variable.  C856 should more precisely say
"variable" instead of "object."

2. Edits
========

[111:6-8 C856 8.5.15 PROTECTED attribute] Replace C856:

"C856 A nonpointer variable that has the PROTECTED attribute and is
      accessed by use association shall not appear in a variable
      definition context (19.6.7), as the <data-target> in a
      <pointer-assignment-stmt> or in a <component-data-source> in a
      <structure-constructor>, or as the <initial-data-target> in a
      <component-initialization> in a <component-decl> in a derived type
      definition or in an <initialization> in a type declaration
      statement."

or

"C856 A nonpointer variable that has the PROTECTED attribute and is
      accessed by use association shall not appear
      o  in a variable definition context (19.6.7),
      o  as the <data-target> in a
         -  <pointer-assignment-stmt> or
         -  <component-data-source> in a <structure-constructor>, or
      o  as the <initial-data-target> in
         -  a <component-initialization> in a <component-decl> in a
            derived type definition or
         -  an <initialization> in a type declaration statement."

3. Alternative Edits
====================

It might seem simpler to say that a nonpointer object that has the
PROTECTED attribute does not have the TARGET attribute in scoping units
where it is accessed by use association.  That might have the bizarre
indirect effect of prohibiting to use a pointer that has such a variable
as its target, which could be the case if the association is established
in the module where the PROTECTED attribute is specified or where the
variable does not have the PROTECTED attribute, from being the
<data-target> in a pointer assignment statement etc., because the target
of the pointer doesn't have the TARGET attribute.  If it is nonetheless
acceptable to say so:

[111:2 8.5.15p1 PROTECTED attribute] Append a sentence: "A nonpointer
variable that has the PROTECTED attribute does not have the TARGET
attribute in scoping units where it is accessed by use association."

[111:6 8.5.15 PROTECTED attribute] Replace "object" with "variable".

[111:7-8 C856 8.5.15 PROTECTED attribute] Delete "or as the
<data-target> or <proc-target> in a <pointer-assignment-stmt>".

4. Question for the Editor
==========================

Does this need to be mentioned in the Introduction?

[xviii Introduction under "Data usage and computation] Insert a sentence
"A nonpointer variable with the PROTECTED attribute is prohibited to
appear in three contexts that were formerly allowed."

5. If this technical change is out of order at this time
========================================================

[111:6 C856 8.5.15 PROTECTED attribute] Replace "object" with
"variable".

[111:7 C856 8.5.15 PROTECTED attribute] Delete "or <proc-target>".