Re: [Factor-talk] context switch benchmarking

[Slava P. <sl...@fa...>]

On 17-Apr-07, at 10:29 AM, Csaba Henk wrote:

> Ah, thanks, superb. I did look into the coroutines module but  
> apparently
> I was somewhat absent-minded as I looked over the co* words.
>
> The benchmark result is not in the top category but I think it's still
> not bad given that continuations are implemented in the language  
> itself
> (and, as you say, they run interpreted).
>
>  python:                      0.02 s

Python's generators are not full continuations, in fact I believe  
they're rewritten to an iterator as a compile-time transform, so the  
resulting bytecode does not look like the original code at all. This  
explains why Python is so much faster than the others on this benchmark.

>  gambit scheme (interpreted): 0.25 s
>  python with greenlets based
>    generators:                0.45 s
>  factor:                      0.73 s
>  ruby:                        1.35 s

Once Factor's continuations are compiled, performance will improve here.

Slava