Re: [Cbor] Robert Wilton's No Objection on draft-ietf-cbor-7049bis-14: (with COMMENT)

Carsten Bormann <cabo@tzi.org> Mon, 21 September 2020 20:15 UTC

Return-Path: <cabo@tzi.org>
X-Original-To: cbor@ietfa.amsl.com
Delivered-To: cbor@ietfa.amsl.com
Received: from localhost (localhost [127.0.0.1]) by ietfa.amsl.com (Postfix) with ESMTP id 2A0D63A07E5; Mon, 21 Sep 2020 13:15:48 -0700 (PDT)
X-Virus-Scanned: amavisd-new at amsl.com
X-Spam-Flag: NO
X-Spam-Score: -1.919
X-Spam-Level:
X-Spam-Status: No, score=-1.919 tagged_above=-999 required=5 tests=[BAYES_00=-1.9, RCVD_IN_MSPIKE_H4=-0.01, RCVD_IN_MSPIKE_WL=-0.01, SPF_HELO_NONE=0.001, SPF_PASS=-0.001, URIBL_BLOCKED=0.001] autolearn=ham autolearn_force=no
Received: from mail.ietf.org ([4.31.198.44]) by localhost (ietfa.amsl.com [127.0.0.1]) (amavisd-new, port 10024) with ESMTP id aksf0cTW2Nxd; Mon, 21 Sep 2020 13:15:45 -0700 (PDT)
Received: from gabriel-vm-2.zfn.uni-bremen.de (gabriel-vm-2.zfn.uni-bremen.de [134.102.50.17]) (using TLSv1.2 with cipher AECDH-AES256-SHA (256/256 bits)) (No client certificate requested) by ietfa.amsl.com (Postfix) with ESMTPS id 9900A3A07CE; Mon, 21 Sep 2020 13:15:44 -0700 (PDT)
Received: from [192.168.217.124] (p548dcc60.dip0.t-ipconnect.de [84.141.204.96]) (using TLSv1.2 with cipher ECDHE-RSA-AES256-GCM-SHA384 (256/256 bits)) (No client certificate requested) by gabriel-vm-2.zfn.uni-bremen.de (Postfix) with ESMTPSA id 4BwG0G3QCcz101q; Mon, 21 Sep 2020 22:15:42 +0200 (CEST)
Content-Type: text/plain; charset=utf-8
Mime-Version: 1.0 (Mac OS X Mail 13.4 \(3608.120.23.2.1\))
From: Carsten Bormann <cabo@tzi.org>
In-Reply-To: <159965254327.29648.15632221792885283453@ietfa.amsl.com>
Date: Mon, 21 Sep 2020 22:15:42 +0200
Cc: The IESG <iesg@ietf.org>, draft-ietf-cbor-7049bis@ietf.org, CBOR Working Group <cbor-chairs@ietf.org>, cbor@ietf.org, Francesca Palombini <francesca.palombini@ericsson.com>
X-Mao-Original-Outgoing-Id: 622412141.911689-1267ac048c8cda487f210a424e552034
Content-Transfer-Encoding: quoted-printable
Message-Id: <3647F203-B61C-4096-953D-07A87130D342@tzi.org>
References: <159965254327.29648.15632221792885283453@ietfa.amsl.com>
To: Robert Wilton <rwilton@cisco.com>
X-Mailer: Apple Mail (2.3608.120.23.2.1)
Archived-At: <https://mailarchive.ietf.org/arch/msg/cbor/pQwe6egPI8OBXuwtN-bY79Vph8g>
Subject: Re: [Cbor] Robert Wilton's No Objection on draft-ietf-cbor-7049bis-14: (with COMMENT)
X-BeenThere: cbor@ietf.org
X-Mailman-Version: 2.1.29
Precedence: list
List-Id: "Concise Binary Object Representation \(CBOR\)" <cbor.ietf.org>
List-Unsubscribe: <https://www.ietf.org/mailman/options/cbor>, <mailto:cbor-request@ietf.org?subject=unsubscribe>
List-Archive: <https://mailarchive.ietf.org/arch/browse/cbor/>
List-Post: <mailto:cbor@ietf.org>
List-Help: <mailto:cbor-request@ietf.org?subject=help>
List-Subscribe: <https://www.ietf.org/mailman/listinfo/cbor>, <mailto:cbor-request@ietf.org?subject=subscribe>
X-List-Received-Date: Mon, 21 Sep 2020 20:15:48 -0000

Hi Rob,

thank you for your kind words about CBOR and for mentioning the signed
integer range issue.

> ----------------------------------------------------------------------
> COMMENT:
> ----------------------------------------------------------------------
>
> Hi,
>
> Thank you for your work on this document, and bringing this to full standard.
> Since I'm a big fan of CBOR and try to evangelize it whenever possible I'm
> please to see this happening.
>
> However, I have one minor annoyance with CBOR, which is the range of negative
> numbers that are encoded in major type 1.  My gripe is that the encoding allows
> for negative integers that are not easily representable in a simple form in
> most programming languages without using something equivalent to BigInteger.

Right.  The underlying observation was that unsigned integers are way more likely to be used in constrained environments than signed integers.  We didn't want to waste a significant part of the code space, so negative integers were designed to be non-overlapping with unsigned ones, and that gives us a bit more than the usual bit-stealing sign representation used for signed integers.  That may be less familiar than just mapping C's data types, but it certainly makes sense for the application.  So you can have a range from -256..255 that fits into a single byte (plus the initial byte).

> E.g., all values below -2^63 won't fit into a int64 type, and the value 2^64
> won't even fit into an uint64 that was used to represent a negative number
> (obviously unless it followed the CBOR encoding semantics of being offset by 1)

We don't go into good platform representations of 65-bit signed integers.

> For a generic decoder I presume that this isn't an issue since it can fallback
> to something like BigInteger.  But for other decoders handling normal sized
> integer datatypes I would presume that they would effectively presumably regard
> anything smaller than -2^63 as not well-formed for their specific problem
> domain.

Certainly well-formed, but not "expected" according to the set of
application data models they can support.

When we do an update of the CDDL spec (RFC 8610), we might spend a little more time on explaining why you rarely want to use the full 65-bit signed types but instead be more specific about what you need.  But CBOR itself simply happily can represent what you give it, and a bit more than you'd expect at first.

> I'm not suggesting that this should be changed (hence comment not a discuss),
> but there are a couple of places in the document that it might be helpful to
> warn implementors about this, that I have mentioned below.
>
> Other minor comments:
>
>    3.  Specification of the CBOR Encoding
>
>       Major type 0:  an integer in the range 0..2**64-1 inclusive.  The
>          value of the encoded item is the argument itself.  For example,
>          the integer 10 is denoted as the one byte 0b000_01010 (major type
>          0, additional information 10).  The integer 500 would be
>          0b000_11001 (major type 0, additional information 25) followed by
>          the two bytes 0x01f4, which is 500 in decimal.
>
>       Major type 1:  a negative integer in the range -2**64..-1 inclusive.
>          The value of the item is -1 minus the argument.  For example, the
>          integer -500 would be 0b001_11001 (major type 1, additional
>          information 25) followed by the two bytes 0x01f3, which is 499 in
>          decimal.
>
> Would writing "0 to 2**64-1" be more clear than 0..2**64-1?

(Sorry, CDDL thinking here, https://tools.ietf.org/html/rfc8610#section-2.2.2.1.)
We use this notation some 35 times in the document, so adding some text in the terminology section makes sense (now in 1.2).
(We won't use the opportunity to delete all 4 instances of "inclusive", though.)

> Or otherwise
> perhaps mention that in the terminology section that "x..y" is used to
> represent an inclusive range set of all values from x to y, including x and y.
> Also, noting that here where ".." has been used it explicit states that it is
> inclusive, but that doesn't appear to be the case everywhere.
>
> I suggest changing "Major type 0:  an integer ..." back to "Major type 0:  an
> unsigned integer", as in RFC7049, because the type is referred to as "Unsigned
> integer".  It also makes it more consistent with the definition of Major type 1.

Indeed; this lack of symmetry already came up in other discussions.

>    3.2.1.  The "break" Stop Code
>
>       The "break" stop code is encoded with major type 7 and additional
>       information value 31 (0b111_11111).  It is not itself a data item: it
>       is just a syntactic feature to close an indefinite-length item.
>
>       If the "break" stop code appears anywhere where a data item is
>       expected, other than directly inside an indefinite-length string,
>       array, or map -- for example directly inside a definite-length array
>       or map -- the enclosing item is not well-formed.
>
> I was wondering whether it would be helpful to clarify that by
> indefinite-length string it means text or byte string?  Although this becomes
> clear in section 3.2.3 anyway ...  My thinking is that section 3.2 lists 4
> types that can have indefinite length, and then in this section both types are
> string are treated together.

When discussing CBOR data models, we tend to group major types 2 and 3
as "strings", and 4 and 5 as "containers".  We don't use the latter in
7049bis; we do use unqualified "string" though.

We used the end of 3.2 to add another expansion of this term.

>    3.2.3.  Indefinite-Length Byte Strings and Text Strings
>
> Would it be helpful to clarify that the chunks must be the same type.  E.g. you
> cannot have a byte string that contains text string chunks and vice-versa?

We do say:

   If any item between the indefinite-length string indicator
   (0b010_11111 or 0b011_11111) and the "break" stop code is not a
   definite-length string item of the same major type, the string is
   not well-formed.

https://cbor-wg.github.io/CBORbis/draft-ietf-cbor-7049bis.html#rfc.section.3.2.3

>
>    3.4.5.2.  Expected Later Encoding for CBOR-to-JSON Converters
>
> "Tags number 21 to 23 ..." => "Tag numbers 21 to 23 ..."

Fixed.

>    4.2.1.  Core Deterministic Encoding Requirements
>
>          Floating-point values also MUST use the shortest form that
>          preserves the value, e.g. 1.5 is encoded as 0xf93e00 and 1000000.5
>          as 0xfa49742408.  (One implementation of this is to have all
>          floats start as a 64-bit float, then do a test conversion to a
>          32-bit float; if the result is the same numeric value, use the
>
> I find this paragraph slightly opaque, and I would suggest spelling out that
> 1.5 has been encoded as a 16 bit IEEE float, whereas 1.00000005 has been
> encoded as a 32 bit IEEE float.  The same comment applies to 4.2.2 as well.

Added the terms binary16/32/64 as used in other parts of the document.

> I also noticed that in most places the document refers to "floating-point" but
> in a few places "floating point" is used instead.

Fixed a few occurrences, thanks.
(We did this previously, and then Brownian motion crept in.)

>    5.5.  Numbers
>
> As per my top comment, I think that it would be useful to raise in this section
> that CBOR can encode negative values that cannot normally be represented in a
> compact form.

Added in the middle of the first paragraph:

  Another example is that, since CBOR keeps the sign bit for its integer
  representation in the major type, it has one bit more for signed
  numbers of a certain length (e.g., -2\*\*64..2\*\*64-1 for 1+8-byte
  integers) than the typical platform signed integer representation of
  the same length (-2\*\*63..2\*\*63-1 for 8-byte int64_t).


>    6.1.  Converting from CBOR to JSON
>
>       Most of the types in CBOR have direct analogs in JSON.  However, some
>       do not, and someone implementing a CBOR-to-JSON converter has to
>       consider what to do in those cases.  The following non-normative
>       advice deals with these by converting them to a single substitute
>       value, such as a JSON null.
>
>       *  An integer (major type 0 or 1) becomes a JSON number.
>
> It is worth referencing back to section 5.5 on Javascript numbers and
> explicitly warn that not all CBOR integers can be precisely represented as JSON
> numbers, and there may be a loss of precision?

We added a paragraph at the end of 6.1 that points to RFC 7493.

>    Appendix C.  Pseudocode
>
>       Major types 0 and 1 are designed in such a way that they can be
>       encoded in C from a signed integer without actually doing an if-then-
>       else for positive/negative (Figure 2).  This uses the fact that
>       (-1-n), the transformation for major type 1, is the same as ~n
>       (bitwise complement) in C unsigned arithmetic; ~n can then be
>       expressed as (-1)^n for the negative case, while 0^n leaves n
>       unchanged for non-negative.  The sign of a number can be converted to
>       -1 for negative and 0 for non-negative (0 or positive) by arithmetic-
>       shifting the number by one bit less than the bit length of the number
>       (for example, by 63 for 64-bit numbers).
>
> This was another place where I thought that it might be useful to warn the
> reader about decoding negative integers and the risk of overflow taking a major
> 1 value into an int64 native type.

This paragraph is about describing the pseudocode that follows, which
is an encoder.

We don't have pseudocode that shows how a decoder might range-check
encoded data items for fit into the receiving data structures; this is
so platform specific that I don't think we should provide any
(and certainly not at this stage of the document).

All these changes are in ae2bab7 and 94f98be, now PR #216 (merged):
https://github.com/cbor-wg/CBORbis/pull/216

Grüße, Carsten