X3J3/96-175
                                                         Page 1 of 1

Date:     November 13, 1996
To:       X3J3
From:     R. Baker Kearfott
Subject:  Choices for Interval Mixed Mode

The discussion group (producers of X3J3/ 96-156) decided that
constants should be handled differently from variables in mixed
mode expressions, for the following reasons.

1. It is natural in certain contexts to put non-interval
   constants into expressions.  For example, one may write

   F = X**2 + 3 X + 0.1

2. Many constants, such as 3, 0.25, are representable exactly
   on most processors, while others are not.

3. It is generally known what the user expects when a constant
   such as 0.1 is written, whereas, if a non-interval variable
   such as R appears in an expression, it is unknown how it has
   arisen, or how accurately it represents a user-intended
   "true" value.

These considerations, along with subsequent discussion at X3J3,
have led to the following modification of 96-156.

===================================================================

Whenever a constant or variable is converted to interval, it is
converted to an interval [A,B] that contains the exact value
represented by the constant or variable.  When the conversion
occurs shall be consistent with general rules for mixed mode
expressions.

(Note:  I have removed the requirement in 96-156 that constants be
        converted as if they were first converted to interval, with
        no intervening conversion to other floating point values.)

===================================================================

In any case, language to users stating that, in general,
mixed mode arithmetic is to be avoided, is advisable.  (Users
can avoid mixed mode arithmetic with interval constructors or
with the function IVAL.)  The reason is that mixed mode arithmetic
may not always give users the expected enclosures.