Gordian dCBOR: A Deterministic CBOR Application Profile

Introduction CBOR has many advantages over other data serialization formats. One of its strengths is specifications and guidelines for serializing data deterministically, such that multiple agents serializing the same data automatically achieve consensus on the exact byte-level form of that serialized data. This is particularly useful when data must be compared for semantic equivalence by comparing the hash of its contents. Nonetheless, determinism is an opt-in feature of CBOR, and most existing CBOR codecs put the primary burden of correct deterministic serialization and validation of deterministic encoding during deserialization on the engineer. This document specifies a set of requirements for the application profile "dCBOR" that MUST be implemented at the codec level. These requirements include but go beyond §4.2.

Conventions and Definitions The key words "MUST", "MUST NOT", "REQUIRED", "SHALL", "SHALL NOT", "SHOULD", "SHOULD NOT", "RECOMMENDED", "NOT RECOMMENDED", "MAY", and "OPTIONAL" in this document are to be interpreted as described in BCP 14 when, and only when, they appear in all capitals, as shown here.

Application Profile The dCBOR Application Profile specifies the use of Deterministic Encoding as defined in Section 4.2 of together with some application-level rules specified in this section. The application-level rules specified here do not "fork" CBOR. A dCBOR implementation produces well-formed, deterministically encoded CBOR according to , and existing generic CBOR decoders will therefore be able to decode it, including those that check for deterministic encoding. Similarly, generic CBOR encoders will be able to produce valid dCBOR if handed dCBOR conforming data model level information from an application. Note that the separation between standard CBOR processing and the processing required by the dCBOR application profile is a conceptual one: Both dCBOR processing and standard CBOR processing may be combined into a unified dCBOR/CBOR codec. The requirements in this document apply to encoding or decoding of dCBOR data, regardless of whether the codec is a unified dCBOR/CBOR codec operating in dCBOR-compliant modes, or a single-purpose dCBOR codec. Both of these are generically referred to as "dCBOR codecs" in this document. This application profile is intended to be used in conjunction with an application, which typically will use a subset of CBOR, which in turn influences which subset of the application profile is used. As a result, this application profile places no direct requirement on what subset of CBOR is implemented. For instance, there is no requirement that dCBOR implementations support floating point numbers (or any other kind of number, such as arbitrary precision integers or 64-bit negative integers) when they are used with applications that do not use them. However, this document does place requirements on dCBOR implementations that support negative 64-bit integers and 64-bit or smaller floating point numbers.

Base Requirements dCBOR encoders MUST only emit CBOR conforming to the requirements "Core Deterministic Encoding Requirements" of §4.2.1. To summarize, dCBOR codecs:

MUST encode variable-length integers using the shortest form possible.
MUST encode floating-point values using the shortest form that preserves the value.
MUST NOT encode indefinite-length arrays or maps.
MUST sort map keys in bytewise lexicographic order of their deterministic encodings.

In addition, dCBOR decoders: 5. MUST validate and return errors for any encoded CBOR that is not conformant to any part of this specification.

Duplicate Map Keys Standard CBOR defines maps with duplicate keys as invalid, but leaves how to handle such cases to the implementor (§2.2, §3.1, §5.4, §5.6). dCBOR codecs:

MUST NOT emit CBOR that contains duplicate map keys.
MUST reject encoded maps with duplicate keys.

Numeric Reduction dCBOR codecs that support floating point numbers (CBOR major type 7):

MUST reduce floating point values with no fractional part to the shortest integer encoding that can accurately represent it.
MUST reduce floating point values with a non-zero fractional part to the shortest floating point encoding that can accurately represent it.
MUST support floating point binary16 as the most-preferred encoding for floating point values, followed by binary32, then binary64.

This practice still produces well-formed CBOR according to the standard, and all existing generic decoders will be able to read it. It does exclude a map such as the following from being validated as dCBOR, even though it would be allowed in standard CBOR because:

10.0 is an invalid numeric value in dCBOR, and
using the unsigned integer value 10 more than once as a map key is not allowed.

Reduction of Negative Zero defines a negative zero value -0.0. dCBOR codecs that support floating point:

MUST reduce all negative zero values to the integer value 0.
MUST reject any encoded negative zero values.

Therefore with dCBOR, 0.0, -0.0, and 0 all encode to the same canonical single-byte value 0x00.

Reduction of NaNs and Infinities defines the NaN (Not a Number) value . This is usually divided into two types: quiet NaNs and signalling NaNs, and the sign bit is used to distinguish between these two types. The specification also includes a range of "payload" bits. These bit fields have no definite purpose and could be used to break determinism or exfiltrate data. dCBOR encoders that support floating point:

MUST reduce all NaN values to the binary16 quiet NaN value having the canonical bit pattern 0x7e00.
MUST reject any other encoded NaN values.
MUST reduce all +INF values to the binary16 +INF having the canonical bit pattern 0x7c00 and likewise with -INF to 0xfc00.
MUST reject any encoded INF or -INF values other than these.

65-bit Negative Integers The largest negative integer that can be represented in 64-bit two's complement (STANDARD_NEGATIVE_INT_MAX) is -2^63 (0x8000000000000000). However, standard CBOR major type 1 can encode negative integers as low as CBOR_NEGATIVE_INT_MAX, which is -2^64 (two's complement: 0x10000000000000000, CBOR: 0x3BFFFFFFFFFFFFFFFF). Negative integers in the range [CBOR_NEGATIVE_INT_MAX ... STANDARD_NEGATIVE_INT_MAX - 1] require 65 bits of precision, and are thus not representable in typical machine-sized integers. Because of this incompatibility between standard CBOR and typical machine-size representations, dCBOR disallows encoding negative integer values in the range [CBOR_NEGATIVE_INT_MAX ... STANDARD_NEGATIVE_INT_MAX - 1]. dCBOR codecs:

MUST NOT encode these values as CBOR major type 1.
MUST reject these encoded major type 1 CBOR values.