Re: ObjC and speed for Smalltalk interpretation

[Travis Griggs]

Stefan Urbanek wrote:
> Hi all,
>
> I have a question:
>
> Currently the Smalltalk interpreter for StepTalk is using bytecodes. 
> Basically they are pop, push, send, return and few others. I was 
> thinking about another solution, and that is to compile into some kind 
> of command-object hierarchy (reflecting source code) isntead of 
> bytecodes and execute it command by command by sending it a message. I 
> think, that some optimalisations can be done using that kind of 
> execution, however, i am not expert in that and therefore I would like  
> to ask you.
>
> For example: command MessageSendCommand with method: 
> executeInContext:environment:receiver:
>
> Perhaps some message sends (fetching a bytecode, object manipulation, 
> ...) can be saved with that... Or not? I think that CommandObjects can 
> adapt to the execution process to create kind of speed up (caching per 
> command in loops for example).
>
> What is better way to go?
>
> Ideas?

Stefan,

Have you asked in any of the Smalltalk forums. As an (ardent Smalltalk) 
lurking in this list, I find your project really interesting. I wonder 
what some of the hard core VM engineers wouldn't have some ideas for 
you. Names that come to mind, Eliot Miranda (VisualWorks VM guru), 
Vasilli Bykov (implemented ST80 BlueBook as an Smalltalk inside a 
Smalltalk and got faster than the original performance), John Brant 
(architect of one of the "Smalltalk in .net" environments), David 
Simmons (another of the "Smalltalk in .net" flavors, and a compiler guru 
in general), and many others. Post some of your questions at 
comp.lang.smalltalk maybe?

--
Travis Griggs
Objoligist
Key Technology
Conformity is not an excuse for Cranial Convalescence