18 KiB

Raw Blame History

[WIP] SimpleSerialize (SSZ) Spec

This is the work in progress document to describe simpleserialize, the current selected serialization method for Ethereum 2.0 using the Beacon Chain.

This document specifies the general information for serializing and deserializing objects and data types.

About

SimpleSerialize was first proposed by Vitalik Buterin as the serialization protocol for use in the Ethereum 2.0 Beacon Chain.

The core feature of ssz is the simplicity of the serialization with low overhead.

Terminology

Term	Definition
`big`	Big Endian
`byte_order`	Specifies endianness: Big Endian or Little Endian.
`len`	Length/Number of Bytes.
`to_bytes`	Convert to bytes. Should take parameters `size` and `byte_order`.
`from_bytes`	Convert from bytes to object. Should take `bytes` and `byte_order`.
`value`	The value to serialize.
`rawbytes`	Raw serialized bytes.

Constants

Constant	Value	Definition
`LENGTH_BYTES`	4	Number of bytes used for the length added before a variable-length serialized object.

Overview

Serialize/Encode

uint: 8/16/24/32/64/256

Convert directly to bytes the size of the int. (e.g. uint16 = 2 bytes)

All integers are serialized as big endian.

Check to perform	Code
Size is a byte integer	`int_size % 8 == 0`

assert(int_size % 8 == 0)
buffer_size = int_size / 8
return value.to_bytes(buffer_size, 'big')

bool

Convert directly to a single 0x00 or 0x01 byte.

Check to perform	Code
Value is boolean	`value in (True, False)`

assert(value in (True, False))
return b'\x01' if value is True else b'\x00'

Address

The address should already come as a hash/byte format. Ensure that length is 20.

Check to perform	Code
Length is correct (20)	`len(value) == 20`

assert( len(value) == 20 )
return value

Hash

Hash Type	Usage
`hashN`	Hash of arbitrary byte length `N`.

Checks to perform	Code
Length in bytes is correct for `hashN`	`len(value) == N`

hashN

assert(len(value) == N)

return value

Bytes

For general bytes type:

Get the length/number of bytes; Encode into a 4-byte integer.
Append the value to the length and return: [ length_bytes ] + [ value_bytes ]

Check to perform	Code
Length of bytes can fit into 4 bytes	`len(value) < 2**32`

assert(len(value) < 2**32)
byte_length = (len(value)).to_bytes(LENGTH_BYTES, 'big')
return byte_length + value

List/Vectors

Lists are a collection of elements of the same homogeneous type.

Check to perform	Code
Length of serialized list fits into 4 bytes	`len(serialized) < 2**32`

Get the number of raw bytes to serialize: it is len(list) * sizeof(element).
- Encode that as a 4-byte big endian uint32.
Append the elements in a packed manner.

Note on efficiency: consider using a container that does not need to iterate over all elements to get its length. For example Python lists, C++ vectors or Rust Vec.

Example in Python


serialized_list_string = b''

for item in value:
   serialized_list_string += serialize(item)

assert(len(serialized_list_string) < 2**32)

serialized_len = (len(serialized_list_string).to_bytes(LENGTH_BYTES, 'big'))

return serialized_len + serialized_list_string

Container

A container represents a heterogenous, associative collection of key-value pairs. Each pair is referred to as a field. To get the value for a given field, you supply the key which is a symbol unique to the container referred to as the field's name. The container data type is analogous to the struct type found in many languages like C or Go.

To serialize a container, obtain the set of its field's names and sort them lexicographically. For each field name in this sorted list, obtain the corresponding value and serialize it. Tightly pack the complete set of serialized values in the same order as the sorted field names into a buffer. Calculate the size of this buffer of serialized bytes and encode as a 4-byte big endian uint32. Prepend the encoded length to the buffer. The result of this concatenation is the final serialized value of the container.

Check to perform	Code
Length of serialized fields fits into 4 bytes	`len(serialized) < 2**32`

To serialize:

Get the names of the container's fields and sort them.
For each name in the sorted list, obtain the corresponding value from the container and serialize it. Place this serialized value into a buffer. The serialized values should be tightly packed.
Get the number of raw bytes in the serialized buffer. Encode that number as a 4-byte big endian uint32.
Prepend the length to the serialized buffer.

Example in Python

def get_field_names(typ):
    return typ.fields.keys()

def get_value_for_field_name(value, field_name):
    return getattr(value, field_name)

def get_type_for_field_name(typ, field_name):
    return typ.fields[field_name]

serialized_buffer = b''

typ = type(value)
for field_name in sorted(get_field_names(typ)):
    field_value = get_value_for_field_name(value, field_name)
    field_type = get_type_for_field_name(typ, field_name)
    serialized_buffer += serialize(field_value, field_type)

assert(len(serialized_buffer) < 2**32)

serialized_len = (len(serialized_buffer).to_bytes(LENGTH_BYTES, 'big'))

return serialized_len + serialized_buffer

Deserialize/Decode

The decoding requires knowledge of the type of the item to be decoded. When performing decoding on an entire serialized string, it also requires knowledge of the order in which the objects have been serialized.

Note: Each return will provide deserialized_object, new_index keeping track of the new index.

At each step, the following checks should be made:

Check to perform	Check
Ensure sufficient length	`length(rawbytes) >= current_index + deserialize_length`

uint: 8/16/24/32/64/256

Convert directly from bytes into integer utilising the number of bytes the same size as the integer length. (e.g. uint16 == 2 bytes)

All integers are interpreted as big endian.

assert(len(rawbytes) >= current_index + int_size)
byte_length = int_size / 8
new_index = current_index + int_size
return int.from_bytes(rawbytes[current_index:current_index+int_size], 'big'), new_index

Bool

Return True if 0x01, False if 0x00.

assert rawbytes in (b'\x00', b'\x01')
return True if rawbytes == b'\x01' else False

Address

Return the 20 bytes.

assert(len(rawbytes) >= current_index + 20)
new_index = current_index + 20
return rawbytes[current_index:current_index+20], new_index

Hash

hashN

Return the N bytes.

assert(len(rawbytes) >= current_index + N)
new_index = current_index + N
return rawbytes[current_index:current_index+N], new_index

Bytes

Get the length of the bytes, return the bytes.

Check to perform	code
rawbytes has enough left for length	`len(rawbytes) > current_index + LENGTH_BYTES`
bytes to return not greater than serialized bytes	`len(rawbytes) > bytes_end`

assert(len(rawbytes) > current_index + LENGTH_BYTES)
bytes_length = int.from_bytes(rawbytes[current_index:current_index + LENGTH_BYTES], 'big')

bytes_start = current_index + LENGTH_BYTES
bytes_end = bytes_start + bytes_length
new_index = bytes_end

assert(len(rawbytes) >= bytes_end)

return rawbytes[bytes_start:bytes_end], new_index

List/Vectors

Deserialize each element in the list.

Get the length of the serialized list.
Loop through deserializing each item in the list until you reach the entire length of the list.

Check to perform	code
rawbytes has enough left for length	`len(rawbytes) > current_index + LENGTH_BYTES`
list is not greater than serialized bytes	`len(rawbytes) > current_index + LENGTH_BYTES + total_length`

assert(len(rawbytes) > current_index + LENGTH_BYTES)
total_length = int.from_bytes(rawbytes[current_index:current_index + LENGTH_BYTES], 'big')
new_index = current_index + LENGTH_BYTES + total_length
assert(len(rawbytes) >= new_index)
item_index = current_index + LENGTH_BYTES
deserialized_list = []

while item_index < new_index:
   object, item_index = deserialize(rawbytes, item_index, item_type)
   deserialized_list.append(object)

return deserialized_list, new_index

Container

Refer to the section on container encoding for some definitions.

To deserialize a container, loop over each field in the container and use the type of that field to know what kind of deserialization to perform. Consume successive elements of the data stream for each successful deserialization.

Instantiate a container with the full set of deserialized data, matching each member with the corresponding field.

Check to perform	code
rawbytes has enough left for length	`len(rawbytes) > current_index + LENGTH_BYTES`
list is not greater than serialized bytes	`len(rawbytes) > current_index + LENGTH_BYTES + total_length`

To deserialize:

Get the names of the container's fields and sort them.
For each name in the sorted list, attempt to deserialize a value for that type. Collect these values as they will be used to construct an instance of the container.
Construct a container instance after successfully consuming the entire subset of the stream for the serialized container.

Example in Python

def get_field_names(typ):
    return typ.fields.keys()

def get_value_for_field_name(value, field_name):
    return getattr(value, field_name)

def get_type_for_field_name(typ, field_name):
    return typ.fields[field_name]

class Container:
    # this is the container; here we will define an empty class for demonstration
    pass

# get a reference to the type in some way...
container = Container()
typ = type(container)

assert(len(rawbytes) > current_index + LENGTH_BYTES)
total_length = int.from_bytes(rawbytes[current_index:current_index + LENGTH_BYTES], 'big')
new_index = current_index + LENGTH_BYTES + total_length
assert(len(rawbytes) >= new_index)
item_index = current_index + LENGTH_BYTES

values = {}
for field_name in sorted(get_field_names(typ)):
    field_name_type = get_type_for_field_name(typ, field_name)
    values[field_name], item_index = deserialize(data, item_index, field_name_type)
assert item_index == start + LENGTH_BYTES + length
return typ(**values), item_index

Tree_hash

The below tree_hash algorithm is defined recursively in the case of lists and containers, and it outputs a value equal to or less than 32 bytes in size. For the final output only (ie. not intermediate outputs), if the output is less than 32 bytes, right-zero-pad it to 32 bytes. The goal is collision resistance within each type, not between types.

We define hash(x) as BLAKE2b-512(x)[0:32].

uint: 8/16/24/32/64/256, bool, address, hash32

Return the serialization of the value.

bytes, hash96

Return the hash of the serialization of the value.

List/Vectors

First, we define some helpers and then the Merkle tree function. The constant CHUNK_SIZE is set to 128.

# Returns the smallest power of 2 equal to or higher than x
def next_power_of_2(x):
    return x if x == 1 else next_power_of_2((x+1) // 2) * 2

# Extends data length to a power of 2 by minimally right-zero-padding
def extend_to_power_of_2(data):
    return data + b'\x00' * (next_power_of_2(len(data)) - len(data))

# Concatenate a list of homogeneous objects into data and pad it
def list_to_glob(lst):
    if len(lst) == 0:
        return b''
    if len(lst[0]) != next_power_of_2(len(lst[0])):
        lst = [extend_to_power_of_2(x) for x in lst]
    data = b''.join(lst)
    # Pad to chunksize
    data += b'\x00' * (CHUNKSIZE - (len(data) % CHUNKSIZE or CHUNKSIZE))
    return data

# Merkle tree hash of a list of items
def merkle_hash(lst):
    # Turn list into padded data
    data = list_to_glob(lst)
    # Store length of list (to compensate for non-bijectiveness of padding)
    datalen = len(lst).to_bytes(32, 'big')
    # Convert to chunks
    chunkz = [data[i:i+CHUNKSIZE] for i in range(0, len(data), CHUNKSIZE)]
    # Tree-hash
    while len(chunkz) > 1:
        if len(chunkz) % 2 == 1:
            chunkz.append(b'\x00' * CHUNKSIZE)
        chunkz = [hash(chunkz[i] + chunkz[i+1]) for i in range(0, len(chunkz), 2)]
    # Return hash of root and length data
    return hash((chunkz[0] if len(chunks) > 0 else b'\x00' * 32) + datalen)

To tree_hash a list, we simply do:

return merkle_hash([tree_hash(item) for item in value])

Where the inner tree_hash is a recursive application of the tree-hashing function (returning less than 32 bytes for short single values).

Container

Recursively tree hash the values in the container in order sorted by key, and return the hash of the concatenation of the results.

return hash(b''.join([tree_hash(getattr(x, field)) for field in sorted(value.fields)))

Implementations

Language	Implementation	Description
Python	https://github.com/ethereum/beacon_chain/blob/master/ssz/ssz.py	Beacon chain reference implementation written in Python.
Rust	https://github.com/sigp/lighthouse/tree/master/beacon_chain/utils/ssz	Lighthouse (Rust Ethereum 2.0 Node) maintained SSZ.
Nim	https://github.com/status-im/nim-beacon-chain/blob/master/beacon_chain/ssz.nim	Nim Implementation maintained SSZ.
Rust	https://github.com/paritytech/shasper/tree/master/util/ssz	Shasper implementation of SSZ maintained by ParityTech.
Javascript	https://github.com/ChainSafeSystems/ssz-js/blob/master/src/index.js	Javascript Implementation maintained SSZ

18 KiB Raw Blame History

[WIP] SimpleSerialize (SSZ) Spec

ToC

About

Terminology

Constants

Overview

Serialize/Encode

uint: 8/16/24/32/64/256

bool

Address

Hash

hashN

Bytes

List/Vectors

Container

Deserialize/Decode

uint: 8/16/24/32/64/256

Bool

Address

Hash

hashN

Bytes

List/Vectors

Container

Tree_hash

uint: 8/16/24/32/64/256, bool, address, hash32

bytes, hash96

List/Vectors

Container

Implementations

18 KiB

Raw Blame History