There are two types of fundamental items one can encode in RLP:

  1. Strings of bytes
  2. Lists of other items

In this package, byte strings are represented either as Python strings or as bytearrays. Lists can be any sequence, e.g. lists or tuples. To encode these kinds of objects, use rlp.encode():

>>> from rlp import encode
>>> encode('ethereum')
>>> encode('')
>>> encode('Lorem ipsum dolor sit amet, consetetur sadipscing elitr.')
b'\xb88Lorem ipsum dolor sit amet, consetetur sadipscing elitr.'
>>> encode([])
>>> encode(['this', ['is', ('a', ('nested', 'list', []))]])

Decoding is just as simple:

>>> from rlp import decode
>>> decode(b'\x88ethereum')
>>> decode(b'\x80')
>>> decode(b'\xc0')
>>> decode(b'\xd9\x84this\xd3\x82is\xcfa\xcd\x86nested\x84list\xc0')
[b'this', [b'is', [b'a', [b'nested', b'list', []]]]]

Now, what if we want to encode a different object, say, an integer? Let’s try:

>>> encode(1503)
>>> decode(b'\x82\x05\xdf')

Oops, what happened? Encoding worked fine, but rlp.decode() refused to give an integer back. The reason is that RLP is typeless. It doesn’t know if the encoded data represents a number, a string, or a more complicated object. It only distinguishes between byte strings and lists. Therefore, pyrlp guesses how to serialize the object into a byte string (here, in big endian notation). When encoded however, the type information is lost and rlp.decode() returned the result in its most generic form, as a string. Thus, what we need to do is deserialize the result afterwards.

Sedes objects

Serialization and its couterpart, deserialization, is done by, what we call, sedes objects (borrowing from the word “codec”). For integers, the sedes rlp.sedes.big_endian_int is in charge. To decode our integer, we can pass this sedes to rlp.decode():

>>> from rlp.sedes import big_endian_int
>>> decode(b'\x82\x05\xdf', big_endian_int)

For unicode strings, there’s the sedes rlp.sedes.binary, which uses UTF-8 to convert to and from byte strings:

>>> from rlp.sedes import binary
>>> encode(u'Ðapp')
>>> decode(b'\x85\xc3\x90app', binary)
>>> print(decode(b'\x85\xc3\x90app', binary).decode('utf-8'))

Lists are a bit more difficult as they can contain arbitrarily complex combinations of types. Therefore, we need to create a sedes object specific for each list type. As base class for this we can use rlp.sedes.List:

>>> from rlp.sedes import List
>>> encode([5, 'fdsa', 0])
>>> sedes = List([big_endian_int, binary, big_endian_int])
>>> decode(b'\xc7\x05\x84fdsa\x80', sedes)
(5, b'fdsa', 0)

Unsurprisingly, it is also possible to nest rlp.List objects:

>>> inner = List([binary, binary])
>>> outer = List([inner, inner, inner])
>>> decode(encode(['asdf', 'fdsa']), inner)
(b'asdf', b'fdsa')
>>> decode(encode([['a1', 'a2'], ['b1', 'b2'], ['c1', 'c2']]), outer)
((b'a1', b'a2'), (b'b1', b'b2'), (b'c1', b'c2'))

What Sedes Objects Actually Are

We saw how to use sedes objects, but what exactly are they? They are characterized by providing the following three member functions:

  • serializable(obj)
  • serialize(obj)
  • deserialize(serial)

The latter two are used to convert between a Python object and its representation as byte strings or sequences. The former one may be called by rlp.encode() to infer which sedes object to use for a given object (see Sedes Inference).

For basic types, the sedes object is usually a module (e.g. rlp.sedes.big_endian_int and rlp.sedes.binary). Instances of rlp.sedes.List provide the sedes interface too, as well as the class rlp.Serializable which is discussed in the following section.

Encoding Custom Objects

Often, we want to encode our own objects in RLP. Examples from the Ethereum world are transactions, blocks or anything send over the Wire. With pyrlp, this is as easy as subclassing rlp.Serializable:

>>> import rlp
>>> class Transaction(rlp.Serializable):
...    fields = (
...        ('sender', binary),
...        ('receiver', binary),
...        ('amount', big_endian_int)
...    )

The class attribute fields is a sequence of 2-tuples defining the field names and the corresponding sedes. For each name an instance attribute is created, that can conveniently be initialized with __init__():

>>> tx1 = Transaction(b'me', b'you', 255)
>>> tx2 = Transaction(amount=255, sender=b'you', receiver=b'me')
>>> tx1.amount

At serialization, the field names are dropped and the object is converted to a list, where the provided sedes objects are used to serialize the object attributes:

>>> Transaction.serialize(tx1)
[b'me', b'you', b'\xff']
>>> tx1 == Transaction.deserialize([b'me', b'you', b'\xff'])

As we can see, each subclass of rlp.Serializable implements the sedes responsible for its instances. Therefore, we can use rlp.encode() and rlp.decode() as expected:

>>> encode(tx1)
>>> decode(b'\xc9\x82me\x83you\x81\xff', Transaction) == tx1

Sedes Inference

As we have seen, rlp.encode() (or, rather, rlp.infer_sedes()) tries to guess a sedes capable of serializing the object before encoding. In this process, it follows the following steps:

  1. Check if the object’s class is a sedes object (like every subclass of rlp.Serializable). If so, its class is the sedes.
  2. Check if one of the entries in rlp.sedes.sedes_list can serialize the object (via serializable(obj)). If so, this is the sedes.
  3. Check if the object is a sequence. If so, build a rlp.sedes.List by recursively infering a sedes for each of its elements.
  4. If none of these steps was successful, sedes inference has failed.

If you have build your own basic sedes (e.g. for dicts or floats), you might want to hook in at step 2 and add it to rlp.sedes.sedes_list, whereby it will be automatically be used by rlp.encode().

Further Reading

This was basically everything there is to about this package. The technical specification of RLP can be found either in the Ethereum wiki or in Appendix B of Gavin Woods Yellow Paper. For more detailed information about this package, have a look at the API Reference or the source code.