[Stackless] Marketing? Re: Blog posts related to Stackless and benchmarking

[Andrew Francis]

>Date: Wed, 4 Mar 2009 11:04:44 -0500
>From: Richard Tew <richard.m.tew at gmail.com>
>Subject: Re: [Stackless] Blog posts related to Stackless and
    benchmarking
>To: Phoenix Sol <phoenix at burninglabs.com>
>Cc: "stackless at stackless.com" <stackless at stackless.com>
>Message-ID:
>    <952d92df0903040804l51caa344xa6ccebd88df2fb8 at mail.gmail.com>
>Content-Type: text/plain; charset=ISO-8859-1

>While Stackless may have faster scheduling and allow for code
>naturally written without boilerplate to unknowingly block within any
>called function, there's a lot to be said for being able to run
>against standard Python.

Hi Richard:

Some thoughts. Although I have not be methodical, I have run tests between pure Twisted programmes, programmes that use both Stackless and Twisted, and programmes where I trade off tasklets/channels for deferreds. So far, the performance difference is low (between %5 and 10%). My conclusion is that implementing the scheduling in C, makes a substantial difference. 

A selling point of Stackless is that it is a superset of Python. You and the folks at CCP Games do a great job of keeping Stackless in sync with CPython. And the PyPy guys are chartering a course for the future. The only reason I keep a CPython and a Stackless implementation handy is sometimes I want a debugger to step through complicated code. I believe if a little more documentation existed concerning the nitty-gritty of tasklet internals, it would make it easier for third-parties to integrate Stackless into existing frameworks. Thus giving system administrators and eventually distro makers fewer reasons not to stock Stackless Python from the get-go.

Cheers,
Andrew