To:	J3                                   J3-01/324
From: Larry Meadows
Subject: BOZ constants
Date: August 21, 2001

BOZ constants were promoted to full-fledged integer constants
in the F2K draft. This causes many existing codes, that use
a common extension to treat BOZ constants as typeless, to fail.
For example, on an IEEE machine, consider the following code:

	REAL R
	R = Z'3F800000'
	PRINT *,R
	END

The above code is illegal F95. On many existing compilers it
will print 1.0 . In F2K it will print the integral value in
decimal of 3F800000 .

It has been suggested that the transfer intrinsic can be used
to mimic the defacto-standard behavior of existing compilers.
However, the transfer intrinsic cannot be used in DATA statements,
forcing the user to resort to equivalence or other subversion.
Also, the use of BOZ constants as typeless in existing code
is often in legacy code, and the user will not wish to modify
such code.

This paper proposes that BOZ constants be disallowed in most
contexts, so that the defacto-standard behavior can continue
to exist as a vendor extension.
Edits:
[32:41+] Insert constraint:
Constraint:
A /boz-literal-constant/ shall not appear except in the following contexts:

1)  in a DATA statement where it corresponds to an integer variable,

2)  as an /initialization-expr/ in a /type-declaration-stmt/ where the
data entity being initialized is an integer data entity.

3)  as an /initialization-expr/ in a PARAMETER statement where the
named constant is an integer named constant,

4)  as the /expr/ in an intrinsic assignment statement where the
/variable/ is an integer variable,

5)  as an /expr/ in a structure constructor that corresponds to an
integer component of the derived type,  and

6)  as an /expr/ that is an /ac-value/ in an array constructor that contains at
least one /ac-value/ that is an integer /expr/ and is not a
/boz-literal-constant/.