J3/06-223

To:      J3
From:    Bill Long
Subject: C13 - UTI 61 : BITS_KIND
Date:    28-jul-2006

---------------------

Unresolved Technical Issue 61 (page 361) recommends removing from the
result value description of BITS_KIND the sentences about double
complex and non-default integer arguments.

Reply: Removing the sentence about double complex is fine. It is also
the case that the final sentence of the result value description
provides a duplicate specification for the case of double
precision. Edits are provided to fix both of these issues.

The argument for non-default integers is less convincing. The specific
example of legacy Cray hardware is faulty.  It is true that, by
default, integer multiplication and division operations were done with
46 bits of precision using the floating point units. However, BIT_SIZE
for these objects was still returned as 64 because for the purposes of
the bit manipulation routines, such as IAND, all 64 bits of the words
were used. All 64 bits were used for integer add and subtract as well,
but that is not the relevant issue. BIT_SIZE for an integer is related
only to the behavior when the object is being used in the context
where it is treated as sequence of bits in the bit manipulation
routines. This has a detectable effect for operations such as a
circular shift. The goal for BITS_KIND with an integer argument is to
return a bits kind value that will specify a bits object with the same
bit manipulation characteristics as the integer argument.  Since the
integer type is special in that the bits manipuation routines are
defined for integer arguments, singling out integer arguments to
BITS_KIND is reasonable.

The argument that BIT_SIZE(default_integer) is not necessarily the
same as NUMERIC_STORAGE size seems theoretical, as I know of no actual
implementation where this is the case.  However, this issue can be
circumvented by changing description of the result value for a default
integer argument. Edits are provided for this.


Edits:
------

[361:8] Replace "default integer, default real," with "default real".

[361:10-11] Remove the sentence "If X is of type complex with the same
kind type parameter as that of double precision, the result has the
value 4 x NUMERIC_STORAGE_SIZE."

[361:12] Replace "of type non-default integer" with "of type integer".

[361:12-13] Replace "If X is of a non-default logical or non-default
non-integer numeric type" with "Otherwise".

[361:14] Replace "of that type and kind" with "of the type and kind of
X".