To: Deborah Donovan
From: David R Tribble on Thu, Mar 5, 1998 9:24 PM
Subject: Comments on ISO/IEC CD 9899 (C9X)
Message-Id: <2.2.32.19980306013532.00f1a340@central.beasys.com>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Date: Thu, 05 Mar 1998 19:35:32 -0600
To: ansi-c9x <ddonovan@itic.nw.dc.us>
From: David R Tribble <david.tribble@central.beasys.com>
Subject: Comments on ISO/IEC CD 9899 (C9X)

Public Comment Number(s) PC-____ to PC-____

ISO/IEC CD 9899 (SC22N2620) Public Comment
===========================================

Date:               1998-03-03
Author:             David R. Tribble
Author Affiliation: Self
Postal Address:     6004 Cave River Dr.
                    Plano, TX 75093-6951
                    USA
E-mail Address:     dtribble@technologist.com
                    david.tribble@central.beasys.com
                    dtribble@flash.net
Telephone Number:   +1 972 738 6125, 16:00-00:00 UTC
                    +1 972 964 1729, 01:00-05:00 UTC
Fax Number:         +1 972 738 6111
Number of individual comments: 1

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

Comment 1. 

Category: Request for clarification

Committee Draft subsection: 6.5.2.1, 6.1.2.8.2, K.2, K.3.9, K.5.8.

Title: Bitfield widths

Detailed description:

 Section 6.5.2.1 #8 states that a bitfield can be declared as type
 'int', 'signed int', or 'unsigned int'.  It further states that "a
 bitfield is interpreted as a signed or unsigned integer type consisting
 of the specified number of bits".

 However, there is no mention of whether or not the number of bits in a
 bitfield is limited to the number of bits in an '[unsigned] int'.
 6.5.2.1#8 implies that a bitfield can be declared with the number of
 bits in /any/ integer type.  This implies that bitfields can be
 declared to have any number of bits, up to the number of bits in a
 'long' or even a 'long long' integer type.

 Presumably, then, the following declaration is conforming and portable:

    struct Widdle
    {
        unsigned int    a: 29;      // Exceeds bits in UINT_MAX
        unsigned int    b: 33;      // Exceeds bits in ULONG_MAX
    };

 The width of member 'a' exceeds the minimum ISO width of 'unsigned int'
 of 16 bits, so it must take on the type of 'unsigned long int'.
 Similarly, the width of member 'b' exceeds the minimum ISO width of
 'unsigned long int' of 32 bits, so it must take on the type of
 'unsigned long long int'.

 This seems like a reasonable interpretation.  But is it correct?

 Sections K.2, K.3.9, and K.5.8 state that a compiler may allow types
 other than 'signed int' and 'unsigned int' for bitfields as an
 extension.  Assuming that the interpretation above is correct, such
 extensions are superfluous; all bitfield types would be equivalent, and
 would only be distinguished by their signedness.

Suggestion:

 The maximum allowable size for a bitfield should be mentioned in the
 standard, perhaps as a footnote to section 6.5.2.1#8:

    6.5.2.1
      ...
    [#8]
      ...
      A bitfield is interpreted as a signed or unsigned integer type
      consisting of the specified number of bits *.

      * This implies that a bitfield may contain any number of bits,
      up to the number of bits in the widest integer type (umaxint_t).

 If instead it is deemed a limitation that bitfields cannot have a width
 that exceeds the number of bits in an '[unsigned] int', then this
 should be mentioned, perhaps as a footnote.  Further, it should be
 noted in appendix K that exceeding this limit is implementation-defined
 behavior.

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