Skip to main content

cara is a Cap'n proto Alternative RPC API.

Project description

Build Status Coverage Status Codacy Badge Documentation Status PyPI Version PyPI License

cara is a Cap’n proto Alternative RPC API. Read the docs!

Reason for creation

pycapnp is a straight C++ conversion and, while that’s great and all, it’s not pythonic. It also uses capnp’s RPC layer and friends, which is from scratch and isn’t very mature, while there are plenty of RPC layers, event loops, etc already in python and well-maintained.

Requirements

To install via setup.py (or pip), a capnproto installation must be locatable by pkg-config. Installed via a normal ‘sudo make install’ should work, other situations have not been tested.

Usage

First, generate the code from your .capnp files:

capnp compile -ocara my_structs.capnp

Then import them:

import my_structs_capnp

Example

my_structs.capnp

struct MyStruct {
    field @0 :Text;
    nested @1 :NestedStruct;
    struct NestedStruct {
        integer @0 :Int32;
    }
}

Python usage

import my_structs_capnp

my_structs_capnp.MyStruct({'field': 'some text for here'})
# -- or --
m = my_structs_capnp.MyStruct.Create(field='some different text')

# All the classes masquerade as python builtins, like dict:
msgpack.packb(m) == b'\x81\x00\xb3some different text'
# But it's slightly different... Look at Field Shrinking below to
# understand

Pseud Integration

There’s also pseud integration. Pseud supports tornado and gevent, but only tornado on Python 3, so these examples used tornado. If you use Python 2, you’re welcome to use gevent.

The first requirement imposed is that you call cara_pseud.setup_server on your server and cara_pseud.setup_client on your client. Once both are called, you can start the server and client. For the server, register an interface with the class or function you want to export. For the client, wrap the client object with the interface you want to use it as. This API allows a server to export multiple interfaces and a client to use any number of them.

Example

my_ifaces.capnp

interface SimpleEcho {
    echo (text :Text) -> (text :Text);
}

interface BackAndForth {
    interface Callback {
        callback (callback :Callback) -> (result :Text);
    }
    callMeMaybe (callback :Callback) -> ();
    otherFunc () -> ();
}

Python usage:

from cara import cara_pseud
from my_ifaces_capnp import SimpleEcho, BackAndForth

@tornado.gen.coroutine
def create_server():
  server = pseud.Server(...)
  server.bind(...)
  cara_pseud.setup_server(server)
  yield server.start()

  # A function can be used to implement an interface with a single
  # method. The name doesn't have to match either.
  @cara_pseud.register_interface(server, SimpleEcho)
  def func(text):
    return text

  # If an interface has multiple methods, a class is necessary. It also has
  # to implement all the methods, but its name can be anything, too.
  # It can subclass the interface or object, but if you choose the
  # interface, the register_interface call can infer it from the class
  # definition.
  @cara_pseud.register_interface(server)
  class Server(BackAndForth):
    def callMeMaybe(self, callback):
      # You can even use a lambda as an interface.
      callback(lambda: 'internal callback')

    def otherFunc(self):
      pass

@tornado.gen.coroutine
def create_client():
  server = pseud.Client(...)
  server.connect(...)
  cara_pseud.setup_client(server)
  yield client.start()

  echo_iface = SimpleEcho(client)
  result = yield echo_iface.echo('test')
  assert result == 'test'

  # Now let's mess with this exported interface.
  back_and_forth = BackAndForth(client)
  # This is a special combination of fortunate accidents. A method with one
  # argument that is an interface with one method can be called like a
  # decorator. Though, you need to yield it still.
  @back_and_forth.callMeMaybe
  def callback(callback=None):
    result = yield callback()
    assert result == 'internal callback'
  yield callback

io_loop.add_callback(create_server)
io_loop.add_callback(create_client)
io_loop.start()

Field Shrinking

# Notice there's no mention of 'field' in the result:
m = my_structs_capnp.MyStruct({'field': 'some text for here'})
msgpack.packb(m) == b'\x81\x00\xb3some different text'
# Yet it's there when we pack the object directly.
m = {'field': 'some different text'}
msgpack.packb(m) == b'\x81\xa5field\xb3some different text'

The difference is because a cara Struct uses the ordinals of the fields instead of their names. This will only be an issue when sending the packed bytes over to another system that isn’t using cara. If you send it back into cara, it’ll unpack the fields correctly and you can use it like the original pieces.

original = my_structs_capnp.MyStruct.Create(nested={'integer': 2})
packed = msgpack.packb(original)
unpacked = msgpack.unpackb(packed)
# --> {1: {0: 2}}
result = my_structs_capnp.MyStruct(unpacked)
# --> MyStruct({nested: NestedStruct({integer: 2})})

This allows us to serialize a struct into a much smaller bytestring, especially since 0-127 becomes a single byte in msgpack. As long as your capnp schema changes are sufficiently backwards-compatible, you can deserialize and lookup the field numbers to get the appropriate type.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

cara-0.10.3.tar.gz (53.4 kB view hashes)

Uploaded Source

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page