[Stackless] Advice on Writing a Stackless Reactor for Twisted

[Christopher Armstrong]

On 1/16/07, David Wyand <TheHeadGnome at gnometech.com> wrote:
> From: "Christopher Armstrong" <radix at twistedmatrix.com>
> > It doesn't look like this does anything that a simple
> > twisted.internet.task.LoopingCall(ReactorTick).start(0) wouldn't do.
> > If you used that much simpler technique, you also wouldn't restrict
> > yourself to the select reactor.
>
> I'm certainly new at this so any guidance is greatly appreciated, although I
> don't want to hijack Andrew's original query.
>
> In reading the API docs for t.i.task.LoopingCall() it sounds like this
> relies on the Twisted application loop being the driver rather than Twisted
> itself being treated as a tasklet and 'ticked' elsewhere.  Would your code
> fragment above not require a stackless.tasklet() call somewhere to make
> Twisted go along with the uthread-like application loop I've implemented in
> my maintest.py?  Or have I just completely misunderstood?
>
> When I approached this problem I saw that two different paths could be
> taken.  The first would be to allow Twisted to be the driver and wrap all
> tasklets as deferreds (the approach I understand you've taken).  The second
> is to treat Twisted as just another tasklet along with all my other network,
> GUI, and game tasklets, which is what I've posted.  However I've not yet
> gone down either path far enough other than some basic AI simulations to
> discover any potential pitfalls.

What you're relying on is blocking the mainloop between calls to
'simulate'; my proposed change simply changes the blocking to be
slightly deeper down inside the reactor. The blocking will still
happen. Try it out.

-- 
Christopher Armstrong
International Man of Twistery
http://radix.twistedmatrix.com/
http://twistedmatrix.com/
http://canonical.com/

_______________________________________________
Stackless mailing list
Stackless at stackless.com
http://www.stackless.com/mailman/listinfo/stackless