[virtmach] I am new here ...

[Sergei Lodyagin]

OK. Here I've tried explain my design decisions.

1. For what it is?

VM it is slightly overloaded term, or, more correctly, there are 
different applications in
which VMs are used. I suppose that VM is a tool which allow describe 
program in
some common operations (or definitions) to allow its automatic 
transmission and running in
any environment where this VM is running.


2. Why neither bytecode nor AST.

Well, bytecode is a term I've never seen a complete definition. When 
people say "bytecode"
the first association is a stack-based virtual machine. This machine is 
a very simple for
implementation and it can be managed by very small set of operations ( < 
256 - all of them can
be coded as 1 byte). It is a very popular methods becouse, as already 
was mentioned,
it is very simple.

But one will have problems when will try translate such program for a 
real machine - processor.
The problems is a complexity of obtaining information needed for 
high-quality optimization from
this intermediate representation. (May be it is a good topic for 
discussion).

The author of Juice make the same conclusion.
He suggests use an intermediate compiler representation - Abstract 
Syntax Tree (others propose
use for the same purposes quards etc.). It seams to me that this 
approach is good for build
high optimized code, but it is very tied with a source language (for 
Slim Binaries it is Oberon),
it makes an execution part very expensive (becouse it is a good part of 
a whole compiler)
and it doesn't allow interpret the code.


3. The real problem is not VM, the real problem is an environment.

There is a fashion include in VM specification some uniform operation 
system superstructure
to allow VM interreact with resources, other components of software 
(written in another languages) etc.
It seams to me that good design should make a border between these two 
different things.
My approach is: the VM code shouldn't be a user of resources but one of 
the resources for using.

4. VM should have possibility to present a generic code.

Generic means not instantiated (it is Ada terminology, C++ programmers 
use the term "template").
It seams to me that VM can give a power to this conception (sometimes 
they name it
"static inheritance"). Only VM based models allow use generic code by 
the most natural
way - instantiate it with different parameters "on fly".

Well, I don't like very long letters (especially the letters which are 
full of my poor language :(  ).
May be I miss some important topics. If you show interest in it I will 
write a bit more.

Sergei Lodyagin.


> On Sat, 22 Dec 2001, Sergei Lodyagin wrote:
>
>> If you like you can see on my project - http://quirks-uvm.sf.net .  
>> It is my experiment. It is VM which is neither bytecode (like Java and
>> its company) nor AST (like Juice).
>> Also I constantly search for new links, different comparisons and
>> other stuff about VMs. It will be good if sometimes this information
>> will appear in this list.
>
>
> Well, to get things moving a little bit, why don't you tell us a bit
> about your VM design and why you went away from bytecode, ASTs etc?
> (I know you have stuff at your website, doing it here may spark some
> discussions..)
>
> niall
>
>