A minimal fast protobuf implementation with cython. Benchmark shows that it's much faster than google official expremental cpp-python implementation.

I've been using it in production since 2013, only tested with python2.7, feedback on other python release is welcome.

Benchmark

$ ./setup.py build_ext --inplace
$ cd benchmark
$ ./bench.sh
encode[google official pure python]:
10 loops, best of 3: 68.8 msec per loop
encode[google official cpp python]:
100 loops, best of 3: 19.4 msec per loop
encode[py-protobuf][cprotobuf]:
100 loops, best of 3: 3.58 msec per loop
decode[google official pure python]:
10 loops, best of 3: 47.5 msec per loop
decode[google official cpp python]:
100 loops, best of 3: 4.55 msec per loop
decode[py-protobuf][cprotobuf]:
100 loops, best of 3: 3.98 msec per loop

Tutorial

Use plugin

You write a person.proto file like this:

package foo;

message Person {
  required int32 id = 1;
  required string name = 2;
  optional string email = 3;
}

And a people.proto file like this:

package foo;
import "person.proto";

message People {
  repeated Person people = 1;
}

Then you compile it with provided plugin:

$ protoc --cprotobuf_out=. person.proto people.proto

If you have trouble to run a protobuf plugin like on windows, you can directly run protoc-gen-cprotobuf like this:

$ protoc -ofoo.pb person.proto people.proto
$ protoc-gen-cprotobuf foo.pb -d .

Then you get a python module foo_pb.py , cprotobuf generate a python module for each package rather than each protocol file.

The generated code is quite readable:

# coding: utf-8
from cprotobuf import ProtoEntity, Field
# file: person.proto
class Person(ProtoEntity):
    id              = Field('int32',        1)
    name            = Field('string',       2)
    email           = Field('string',       3, required=False)

# file: people.proto
class People(ProtoEntity):
    people          = Field(Person, 1, repeated=True)

Actually, if you only use python, you can write this python module, avoid code generation.

The API

Now, you have this lovely python module, how to parse and serialize messages?

When design this package, We try to minimise the effort of migration, so we keep the names of api akin to protocol buffer's.

Note

Since this is no need to reuse a message instance and call Clear on it in python, It don't provide Clear api, so ParseFromString is more like MergeFromString in official implementation, because it don't call Clear at first.

encode/decode

>>> from foo_pb import Person, People
>>> msg = People()
>>> msg.people.add(
...    id = 1,
...    name = 'jim',
...    email = '[email protected]',
... )
>>> s = msg.SerializeToString()
>>> msg2 = People()
>>> msg2.ParseFromString(s)
>>> len(msg2)
1
>>> msg2.people[0].name
'jim'

reflection

>>> from foo_pb import Person, People
>>> dir(Person._fields[0])
['__class__', '__delattr__', '__doc__', '__format__', '__get__', '__getattribute__', '__hash__', '__init__', '__new__', '__pyx_vtable__', '__reduce__', '__reduce_ex__', '__repr__', '__setattr__', '__sizeof__', '__str__', '__subclasshook__', 'index', 'name', 'packed', 'repeated', 'required', 'wire_type']
>>> Person._fields[0].name
'email'
>>> Person._fieldsmap
{1: <cprotobuf.Field object at 0xb74a538c>, 2: <cprotobuf.Field object at 0xb74a541c>, 3: <cprotobuf.Field object at 0xb74a5c8c>}
>>> Person._fieldsmap_by_name
{'email': <cprotobuf.Field object at 0xb74a5c8c>, 'name': <cprotobuf.Field object at 0xb74a541c>, 'id': <cprotobuf.Field object at 0xb74a538c>}

repeated container

We use RepeatedContainer to represent repeated field, RepeatedContainer is inherited from list, so you can manipulate it like a list, or with apis like google's implementation.

>>> from foo_pb import Person, People
>>> msg = People()
>>> msg.people.add(
...    id = 1,
...    name = 'jim',
...    email = '[email protected]',
... )
>>> p = msg.people.add()
>>> p.id = 2
>>> p.name = 'jake'
>>> p.email = '[email protected]'
>>> p2 = Person(id=3, name='lucy', email='[email protected]')
>>> msg.people.append(p2)
>>> msg.people.append({
...     'id' : 4,
...     'name' : 'lily',
...     'email' : '[email protected]',
... })

encode raw data fast

If you already have your messages represented as list and dict, you can encode it without constructing intermidiate objects, getting ride of a lot of overhead:

>>> from cprotobuf import encode_data
>>> from foo_pb import Person, People
>>> s = encode_data(People, [
...     { 'id': 1, 'name': 'tom', 'email': '[email protected]' }
... ])
>>> msg = People()
>>> msg.ParseFromString(s)
>>> msg.people[0].name
'tom'

Utility APIs

>>> from cprotobuf import encode_primitive, decode_primitive
>>> encode_primitive('uint64', 10)
bytearray(b'\x01')
>>> decode_primitive(b'\n', 'uint64')
(10, 1)

Run Tests

$ nosetests