CBOR is "Concise Binary Object Representation", defined by RFC 8949. Like JSON, but binary, more compact, and supporting a much broader range of data types.
cborg focuses on strictness and deterministic data representations. CBORs flexibility leads to problems where determinism matters, such as in content-addressed data where your data encoding should converge on same-bytes for same-data. cborg helps aleviate these challenges.
cborg is also fast, and is suitable for the browser (is Uint8Array
native) and Node.js.
cborg supports CBOR tags, but does not ship with them enabled by default. If you want tags, you need to plug them in to the encoder and decoder.
import { encode, decode } from 'cborg'
const decoded = decode(Buffer.from('a16474686973a26269736543424f522163796179f5', 'hex'))
console.log('decoded:', decoded)
console.log('encoded:', encode(decoded))
decoded: { this: { is: 'CBOR!', yay: true } }
encoded: Uint8Array(21) [
161, 100, 116, 104, 105, 115,
162, 98, 105, 115, 101, 67,
66, 79, 82, 33, 99, 121,
97, 121, 245
]
When installed globally via npm
(with npm install cborg --global
), the cborg
command will be available that provides some handy CBOR CLI utilities. Run with cborg help
for additional details.
Convert a JSON object into CBOR bytes in hexadecimal format.
$ cborg json2hex '["a", "b", 1, "๐"]'
84616161620164f09f9880
Convert a hexadecimal string to a JSON format.
$ cborg hex2json 84616161620164f09f9880
["a","b",1,"๐"]
$ cborg hex2json --pretty 84616161620164f09f9880
[
"a",
"b",
1,
"๐"
]
Convert a hexadecimal string to a CBOR diagnostic output format which explains the byte contents.
$ cborg hex2diag 84616161620164f09f9880
84 # array(4)
61 # string(1)
61 # "a"
61 # string(1)
62 # "b"
01 # uint(1)
64 f09f # string(2)
f09f9880 # "๐"
import { encode } from 'cborg'
const { encode } = require('cborg')
Encode a JavaScript object and return a Uint8Array
with the CBOR byte representation.
- Objects containing circular references will be rejected.
- JavaScript objects that don't have standard CBOR type representations (without tags) may be rejected or encoded in surprising ways. If you need to encode a
Date
or aRegExp
or another exotic type, you should either form them into intermediate forms before encoding or enable a tag encoder (see Type encoders).- Natively supported types are:
null
,undefined
,number
,bigint
,string
,boolean
,Array
,Object
,Map
,Buffer
,ArrayBuffer
,DataView
,Uint8Array
and all otherTypedArray
s (the underlying byte array of TypedArrays is encoded, so they will all round-trip as aUint8Array
since the type information is lost).
- Natively supported types are:
Number
s will be encoded as integers if they don't have a fractional part (1
and1.0
are both considered integers, they are identical in JavaScript). Otherwise they will be encoded as floats.- Integers will be encoded to their smallest possible representations: compacted (into the type byte), 8-bit, 16-bit, 32-bit or 64-bit.
- Integers larger than
Number.MAX_SAFE_INTEGER
or less thanNumber.MIN_SAFE_INTEGER
will be encoded as floats. There is no way to safely determine whether a number has a fractional part outside of this range. BigInt
s are supported by default within the 64-bit unsigned range but will be also be encoded to their smallest possible representation (so will not round-trip as aBigInt
if they are smaller thanNumber.MAX_SAFE_INTEGER
). LargerBigInt
s require a tag (officially tags 2 and 3).- Floats will be encoded in their smallest possible representations: 16-bit, 32-bit or 64-bit. Unless the
float64
option is supplied. - Object properties are sorted according to the original RFC 7049 canonical representation recommended method: length-first and then bytewise. Note that this recommendation has changed in RFC 8949 to be plain bytewise (this is not currently supported but pull requests are welcome to add it as an option).
- The only CBOR major 7 "simple values" supported are
true
,false
,undefined
andnull
. "Simple values" outside of this range are intentionally not supported (pull requests welcome to enable them with an option). - Objects, arrays, strings and bytes are encoded as fixed-length, encoding as indefinite length is intentionally not supported.
float64
(boolean, defaultfalse
): do not attempt to store floats as their smallest possible form, store all floats as 64-bittypeEncoders
(object): a mapping of type name to function that can encode that type into cborg tokens. This may also be used to reject or transform types as objects are dissected for encoding. See the Type encoders section below for more information.
import { decode } from 'cborg'
const { decode } = require('cborg')
Decode valid CBOR bytes from a Uint8Array
(or Buffer
) and return a JavaScript object.
- Integers (major 0 and 1) that are outside of the safe integer range will be converted to a
BigInt
. - The only CBOR major 7 "simple values" supported are
true
,false
,undefined
andnull
. "Simple values" outside of this range are intentionally not supported (pull requests welcome to enable them with an option). - Indefinite length strings and byte arrays are intentionally not supported (pull requests welcome to enable them with an option). Although indefinite length arrays and maps are supported by default.
allowIndefinite
(boolean, defaulttrue
): when the indefinite length additional information (31
) is encountered for any type (arrays, maps, strings, bytes) or a "break" is encountered, an error will be thrown.allowUndefined
(boolean, defaulttrue
): when major 7, minor 23 (undefined
) is encountered, an error will be thrown. To disallowundefined
on encode, a custom type encoder for'undefined'
will need to be supplied.allowInfinity
(boolean, defaulttrue
): when an IEEE 754Infinity
or-Infinity
value is encountered when decoding a major 7, an error will be thrown. To disallowInfinity
and-Infinity
on encode, a custom type encoder for'number'
will need to be supplied.allowNaN
(boolean, defaulttrue
): when an IEEE 754NaN
value is encountered when decoding a major 7, an error will be thrown. To disallowNaN
on encode, a custom type encoder for'number'
will need to be supplied.allowBigInt
(boolean, defaulttrue
): when an integer outside of the safe integer range is encountered, an error will be thrown. To disallowBigInt
s on encode, a custom type encoder for'bigint'
will need to be supplied.strict
(boolean, defaultfalse
): when decoding integers, including for lengths (arrays, maps, strings, bytes), values will be checked to see whether they were encoded in their smallest possible form. If not, an error will be thrown.- Currently, this form of deterministic strictness cannot be enforced for float representations, or map key ordering (pull requests very welcome).
useMaps
(boolean, defaultfalse
): when decoding major 5 (map) entries, use aMap
rather than a plainObject
. This will nest for any encountered map. During encode, aMap
will be interpreted as anObject
and will round-trip as such unlessuseMaps
is supplied, in which case, allMap
s andObject
s will round-trip asMap
s. There is no way to retain the distinction during round-trip without using a custom tag.tags
(array): a mapping of tag number to tag decoder function. By default no tags are supported. See Tag decoders.
The typeEncoders
property to the options
argument to encode()
allows you to add additional functionality to cborg, or override existing functionality.
When converting JavaScript objects, types are differentiated using the method and naming used by @sindresorhus/is (a custom implementation is used internally for performance reasons) and an internal set of type encoders are used to convert objects to their appropriate CBOR form. Supported types are: null
, undefined
, number
, bigint
, string
, boolean
, Array
, Object
, Map
, Buffer
, ArrayBuffer
, DataView
, Uint8Array
and all other TypedArray
s (their underlying byte array is encoded, so they will all round-trip as a Uint8Array
since the type information is lost). Any object that doesn't match a type in this list will cause an error to be thrown during decode. e.g. encode(new Date())
will throw an error because there is no internal Date
type encoder.
The typeEncoders
option is an object whose property names match to @sindresorhus/is type names. When this option is provided and a property exists for any given object's type, the function provided as the value to that property is called with the object as an argument.
If a type encoder function returns null
, the default encoder, if any, is used instead.
If a type encoder function returns an array, cborg will expect it to contain zero or more Token
objects that will be encoded to binary form.
Token
s map directly to CBOR entities. Each one has a Type
and a value
. A type encoder is responsible for turning a JavaScript object into a set of tags.
This example is available from the cborg taglib as bigIntEncoder
(import { bigIntEncoder } as taglib from 'cborg/taglib'
) and implements CBOR tags 2 and 3 (bigint and negative bigint). This function would be registered using an options parameter { typeEncoders: { bigint: bigIntEncoder } }
. All objects that have a type bigint
will pass through this function.
import { Token, Type } from './cborg.js'
function bigIntEncoder (obj) {
// check whether this BigInt could fit within a standard CBOR 64-bit int or less
if (obj >= -1n * (2n ** 64n) && obj <= (2n ** 64n) - 1n) {
return null // handle this as a standard int or negint
}
// it's larger than a 64-bit int, encode as tag 2 (positive) or 3 (negative)
return [
new Token(Type.tag, obj >= 0n ? 2 : 3),
new Token(Type.bytes, fromBigInt(obj >= 0n ? obj : obj * -1n - 1n))
]
}
function fromBigInt (i) { /* returns a Uint8Array, omitted from example */ }
This example encoder demonstrates the ability to pass-through to the default encoder, or convert to a series of custom tags. In this case we can put any arbitrarily large BigInt
into a byte array using the standard CBOR tag 2 and 3 types.
Valid Token
types for the second argument to Token()
are:
Type.uint
Type.negint
Type.bytes
Type.string
Type.array
Type.map
Type.tag
Type.float
Type.false
Type.true
Type.null
Type.undefined
Type.break
Using type encoders we can:
- Override the default encoder entirely (always return an array of
Token
s) - Override the default encoder for a subset of values (use
null
as a pass-through) - Omit an object type entirely from the encode (return an empty array)
- Convert an object to something else entirely (such as a tag, or make all
number
s into floats) - Throw if something should that is supported should be unsupported (e.g.
undefined
)
By default cborg does not support decoding of any tags. Where a tag is encountered during decode, an error will be thrown. If tag support is needed, they will need to be supplied as options to the decode()
function. The tags
property should contain an array where the indexes correspond to the tag numbers that are encountered during decode, and the values are functions that are able to turn the following token(s) into a JavaScript object. Each tag token in CBOR is followed by a data item, often a byte array of arbitrary length, but can be a more complex series of tokens that form a nested data item. This token is supplied to the tag decoder function.
This example is available from the cborg taglib as bigIntDecoder
and bigNegIntDecoder
(import { bigIntDecoder, bigNegIntDecoder } as taglib from 'cborg/taglib'
) and implements CBOR tags 2 and 3 (bigint and negative bigint). This function would be registered using an options parameter:
const tags = []
tags[2] = bigIntDecoder
tags[3] = bigNegIntDecoder
decode(bytes, { tags })
Implementation:
function bigIntDecoder (bytes) {
let bi = 0n
for (let ii = 0; ii < bytes.length; ii++) {
bi = (bi << 8n) + BigInt(bytes[ii])
}
return bi
}
function bigNegIntDecoder (bytes) {
return -1n - bigIntDecoder(bytes)
}
cborg is designed with deterministic encoding forms as a primary feature. It is suitable for use with content addressed systems or other systems where convergence of binary forms is important. The ideal is to have strictly one way of mapping a set of data into a binary form. Unfortunately CBOR has many opportunities for flexibility, including:
- Varying number sizes and no strict requirement for their encoding - e.g. a
1
may be encoded as0x01
,0x1801
,0x190001
,1a00000001
or1b0000000000000001
. - Varying int sizes used as lengths for lengthed objects (maps, arrays, strings, bytes) - e.g. a single entry array could specify its length using any of the above forms for
1
. Tags can also vary in size and still represent the same number. - IEEE 754 allows for
NaN
,Infinity
and-Infinity
to be represented in many different ways, meaning it is possible to represent the same data using many different byte forms. - Indefinite length items where the length is omitted from the additional item of the entity token and a "break" is inserted to indicate the end of of the object. This provides two ways to encode the same object.
- Tags that can allow alternative representations of objects - e.g. using the bigint or negative bigint tags to represent standard size integers.
- Map ordering is flexible by default, so a single map can be represented in many different forms by shuffling the keys.
- Many CBOR decoders ignore trailing bytes that are not part of an initial object. This can be helpful to support streaming-CBOR, but opens avenues for byte padding.
By default, cborg will always encode objects to the same bytes by applying some strictness rules:
- Using smallest-possible representations for ints, negative ints, floats and lengthed object lengths.
- Always sorting maps using the original recommended RFC 7049 map key ordering rules.
- Omitting support for tags (therefore omitting support for exotic object types).
- Applying deterministic rules to
number
differentiation - if a fractional part is missing and it's within the safe integer boundary, it's encoded as an integer, otherwise it's encoded as a float.
By default, cborg allows for some flexibility on decode of objects, which will present some challenges if users wish to impose strictness requirements at both serialization and deserialization. Options that can be provided to decode()
to impose some strictness requirements are:
strict: true
to impose strict sizing rules for int, negative ints and lengths of lengthed objectsallowNaN: false
andallowInfinity
to prevent decoding of any value that would resolve toNaN
,Infinity
or-Infinity
, using CBOR tokens or IEEE 754 representationโas long as your application can do without these symbols.allowIndefinite: false
to disallow indefinite lengthed objects and the "break" tag- Not providing any tag decoders, or ensuring that tag decoders are strict about their forms (e.g. a bigint decoder could reject bigints that could have fit into a standard major 0 64-bit integer).
- Overriding type decoders where they may introduce undesired flexibility.
Currently, there are two areas that cborg cannot impose strictness requirements (pull requests welcome!):
- Smallest-possible floats, or always-float64 cannot be enforced on decode.
- Map ordering cannot be enforced on decode.
There are a number of forms where an object will not round-trip precisely, if this matters for an application, care should be taken, or certain types should be disallowed entirely during encode.
- All
TypedArray
s will decode asUint8Array
s, unless a custom tag is used. - Both
Map
andObject
will be encoded as a CBORmap
, as will any other object that inherits fromObject
that can't be differentiated by the @sindresorhus/is algorithm. They will all decode asObject
by default, orMap
ifuseMaps
is set totrue
. e.g.{ foo: new Map() }
will round-trip to{ foo: {} }
by default.
Copyright 2020 Rod Vagg
Licensed under the Apache License, Version 2.0 (the "License"); you may not use this file except in compliance with the License. You may obtain a copy of the License at https://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.