Skip to main content

Coroutines for Twisted with tail call optimization support

Project description

Coroutine flow control


Generators wrapped with @txcoroutine.coroutine are otherwise identical to those wrapped with @twisted.internet.defer.inlineCallbacks, however, the object returned by it is an instance of txcoroutine.Coroutine which is a subclass of twisted.internet.defer.Deferred.

Coroutine objects provide an API otherwise identical to that of Deferred objects, however, calling pause, unpause or cancel on Coroutine objects transparently applies the same action on all nested Deferred objects that are currently waited on recursively.

Simple example

Single coroutine that calls a Deferred-returning function. The Deferred is automatically cancelled when the coroutine is stopped.

from __future__ import print_function

from twisted.internet import reactor
from twisted.internet.defer import Deferred

def get_message():
    d = Deferred(canceller=lambda _: (
        print("cancelled getting a message"),
    print("getting a message...")
    heavylifting = reactor.callLater(1.0, d.callback, 'dummy-message')
    return d

def some_process():
        while True:
            msg = yield get_message()
            print("processing message: %s" % (msg,))
    finally:  # could use `except GeneratorExit` but `finally` is more illustrative
        print("coroutine stopped, cleaning up")

def main():
    proc = some_process()
    reactor.callLater(3, proc.cancel)  # stop the coroutine 3 seconds later.



getting a message...
processing message: dummy-message
getting a message...
processing message: dummy-message
cancelled getting a message
coroutine stopped, cleaning up

Advanced example with multiple levels of coroutines and cascaded flow control

from __future__ import print_function

from twisted.internet import reactor, task
from twisted.internet.defer import Deferred

def level3_process():
    basetime = reactor.seconds()
    seconds_passed = lambda: int(round(reactor.seconds() - basetime))
        while True:
            print("iterating: %ss passed" % seconds_passed())
            yield sleep(1.0)
    finally:  # could use `except GeneratorExit` but `finally` is more illustrative
        print("level3_process stopped; cleaning up...")

def level2_process():
        yield level3_process()
        print("level2_process stopped; cleaning up...")

def root_process():
        yield level2_process()
        print("root_process stopped; cleaning up...")

def main():
    proc = root_process()
    reactor.callLater(3, proc.pause)  # pause the coroutine 3 seconds later.
    reactor.callLater(6, proc.unpause)  # then pause 3 seconds later
    reactor.callLater(9, proc.cancel)  # then finally stop it 3 seconds later

def sleep(seconds, reactor=reactor):
    """A simple helper for asynchronously sleeping a certain amount of time."""
    return task.deferLater(reactor, seconds, lambda: None)



iterating: 0s passed
iterating: 1s passed
iterating: 2s passed
iterating: 6s passed
iterating: 7s passed
iterating: 8s passed
level3_process stopped; cleaning up...
level2_process stopped; cleaning up...
root_process stopped; cleaning up...

Tail call optimisation


def fact(n, result=1):
    if n <= 1:
        noreturn(fact(n - 1, n * result))
    yield  # make sure it's a generator

n = coroutine(fact)(10000).result

Note, fact itself should not be decorated with coroutine, otherwise the recursive call would simply create another coroutine. This would still support infinite recursion but would be less efficient and consume slightly more memory per each new level introduced because, internally, all the Deferreds would be alive and chained to each other.

This is mainly meant for recursively and infinitely swapping out behaviour in long running processes. For non-coroutine/non-generator TCO, a simpler approach is also possible by delegating the function invocation directly to the trampoline. However, this would be out of the scope of this package.

Description of operation

The memory held by the caller is immediately released as it swaps itself out for another process, while the Deferred that was originally returned is still bound to the ongoing processing.

def process():
    big_obj = SomeBigObject()
    noreturn(process_state1())  # big_obj is released immediately

def process_state1():
    another_big_obj = SomeBigObject()
    noreturn(process_state2())  # another_big_obj is released immediately

def process_state2():
    yield do_something()

def some_other_coroutine():
    yield process()  # will block until state2 has returned 123

This cannot be achieved with plain @inlineCallbacks while satisfying both requirements.

Memory-efficient solution with @inlineCallbacks:

def process():
     big_obj = SomeBigObject()
     process_state1()  # big_obj is released immediately but the `Deferred` returned by process is fired immediately

Solution with @inlineCallbacks keeping Deferred consistency but not releasing memory:

def process():
     big_obj = SomeBigObject()
     yield process_state1()  # big_obj is not released until process_state1 completes


See also The blog post doesn’t mention tail-call optimisation though.

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

txcoroutine-1.0.9.tar.gz (7.0 kB view hashes)

Uploaded source

Built Distribution

txcoroutine-1.0.9.win32.exe (74.8 kB view hashes)

Uploaded any

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