T
o:         J3                                       J3/14-191r1
From:       Van Snyder and Nick Maclaren
Subject:    Delete <do-block> and Range of the DO construct
Date: 2014 June 25
References: 14-007r1, 14-166

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

In 14-166, Malcolm recommended to delete the <do-block> syntax term and
the "Range of the DO construct" subclause.  We can't delete the "Range
of the DO construct" subclause without replacing references to "the
range of the loop" with "the <block> of the loop" in several places, and
prohibiting branching to the <end-do-stmt> from outside the construct.

Sections 3 and 4 describe this change.

Section 5 includes a proposal to forbid branching to all end construct
statements from outside the construct and make it a constraint.  This
could be regarded as fixing a historical defect.  It is additional to
the previous sections.


2. Specification
================

Delete the <do-block> syntactic term and the "Range of the DO construct"
subclause.

Optionally, add a constraint to forbid branching to all end-construct
statements from outside their construct.  Currently, it is not forbidden
for all of them, and is not a constraint in any case.


3. Edits to 14-007r1
====================

[xviii] Introduction

Add to the bullet "Data usage and computation" the sentence "Branching
to the end statement of an executable construct containing a block from
outside the construct now requires a diagnostic."

[174:14 R813] Replace "<do-block>" with "<block>".

[175:5 R823] Delete R823 (the <do-block> syntax rule).
subclause by
    "It is permitted to branch to the <end-do> of a DO construct only
     from within that DO construct."
[175:15 8.1.6.3p1] Replace "<do-block>" with "<block> of the
<do-construct>".

[176:3 8.1.6.5.1p2] Replace "<do-block>" with "<block>".

In Annex D, so omit if deleted
    [593 R813] Replace "<do-block>" with "<block>".


4. Replacing "range"
====================

[46:25-26] Replace "range" with "<block>", twice.

[174:9 8.1.6.1p3] Replace "range" with "<block>".

[175:14-17 81.6.3p0+] Replace "Range of the DO construct" subclause by
    "It is permitted to branch to the <end-do> of a DO construct only
     from within that DO construct."

[176:34 8.1.6.5.3p1(2)] Replace "range" with "<block>".

[177:5-6 8.1.6.5.3p3] Replace "range" with "<block>", twice.

[177:8 8.1.6.5.4p1] Replace "<range>" with "<block>".

[177:8-9 8.1.6.5.4p1] Replace "<range> of the loop" with "<block> of the
construct".

[177:11-12 C822] Replace "range" with "<block>", twice.

[177:29 8.1.6.5.5p1] Replace "range" with "<block>".

[177:31-33 8.1.6.5.5p1] Replace "range" with "<block>", thrice.

[178:23+5 Note 8.11] Replace "range" with "<block>".

[179:0+12 Note 8.14] Replace "range" with "<block>".

[179:0+25 Note 8.15] Replace "range" with "<block>".

[186:5 C846] Replace "range" with "<block>".

[222:5 9.6.4.5.1p9] PROBLEM: I don't know what to do here

In Annex D, so omit if deleted
    [594 C822] Replace "range" with "<block>", twice.
    [596 C846] Replace "range" with "<block>".


5. Constraining against branching to <end-...-stmt> from
   outside its construct
========================================================

Subclause 8.1.6.3 "Range of the DO construct" prohibits branching to an
<end-do> statement from outside the construct, but it is not a
constraint.

Prohibitions against branching to an <end-associate-stmt>,
<end-if-stmt>, <end-select-stmt>, or <end-select-type-stmt> are not
constraints, and are not in the subclause on branching.  There is no
prohibition against branching to an <end-block-stmt> or
<end-critical-stmt> from outside its construct (nor is there explicit
permission for it, as there once was for END IF).

This could be regarded as a historical defect.  If it is wished to
change this, the following changes need to be made:

The above edit
    [175:14-17 81.6.3] Replace "Range of the DO construct" ...
should be replaced by the edit
    [175:14-17 81.6.3] Delete "Range of the DO construct" subclause

[25:10+ 1.6.3p1+] Within the list of incompatibilities with Fortran
2008, add a list item:

    "Fortran 2008 did not prohibit branching to an <end-block-stmt> or
     <end-critical-stmt> from outside its construct.  This standard
     prohibits branching to those, and a diagnostic is required for
     those and other similar cases."

[187:12+ 8.2.1p1+] Insert a constraint:

"C847a If a branch target statement is an <end-associate-stmt>,
       <end-block-stmt>, <end-critical-stmt>, <end-do-stmt>,
       <end-if-stmt>, <end-select-stmt>, or <end-select-type-stmt>, the
       branching statement that references the label of that branch
       target statement shall be within the construct completed by that
       branch target statement."

[170:35 8.1.3.2p3] Delete the paragraph.

[180:29-30 8.1.7.2p2] Delete the paragraph.

[182:33 8.1.8.2p3] Delete the paragraph.

[185:33 8.1.9.2p9] Delete the paragraph.