[plt-scheme] Behind the scenes, is everything running using continuation passing style?

From: Matthias Felleisen (matthias at ccs.neu.edu)
Date: Wed Nov 14 17:33:21 EST 2007

On Nov 14, 2007, at 2:43 PM, Ryan Culpepper wrote:

> On Nov 14, 2007, at 9:38 AM, Matthias Felleisen wrote:
>> [...]
>> ;; ---
>> When you do implement call/cc with cps, you give up the "native"  
>> stack. Doable but a high cost for integrating with the rest of the  
>> world. As SK points out and I emphasize :-), the best synthesis is  
>> to map cps continuation manipulations into small "native" stack  
>> operations. Dybvig and Hieb's strategy of lazily copying the stack  
>> is by far the best. Clinger has an excellent survey paper on the  
>> topic, comparing different strategies, their [dis]advantages and  
>> benchmarks.
>> But of course, you really have to believe that call/cc and friends  
>> are essential to go through all this work.
> Going through that (substantial) work does have another benefit. If  
> you implement contexts in a way that supports call/cc, you often  
> also avoid artificial limits on the size of your contexts ("for  
> free"). That is, no "Stack overflow" or "Recursion limit exceeded"  
> errors. I would rank elimination of artificial stack limits up  
> there with proper tail recursion in terms of supporting clean  
> program structure.

I do not believe that you need the full complexity of Dybvig-Hieb for  
that though I acknowledge that you get close. -- Matthias

Posted on the users mailing list.