World: r3wp

Terry, Rails already do that. It is called Migrations, a kind of 
versioned schema, and yes you just add a field and everything works.

Volker, compiling Rebol is not impossible, it is just very difficult 
exponentially difficult because " the order evaluation" for the a 
function call can change at any time. Depending on how the words 
in the body are defined. This is the CFG feature and problem.

>> test: func[][foo bar]
>> f1: func[][probe "foo first"]
>> f2: func[x][probe "foo first"]
>> bar: func[][probe "bar second"]
>> foo: :f1
>> test
foo first

bar second

>> foo: :f2
>> test   ;;; Surprise order of evaluation changed
!
bar second

foo first

This gets more complex as the body of the function grows, the problem 
of compilation becomes exponential

That is currently true. But makes it sense? What if  such things 
are simply forbidden for compilable code?

Now you could use fix the body of a function maybe not allowing for 
function! values to change dynamically, but this will limit the language. 
I think the best approach for getting closer to the metal is REBCode. 
But I am not sure you can hav a Metacircular Rebol.

I dont think it limitsthe language. I never used that  as a feature. 
Well maybe setting 'print to none, but i can live with "print: func[value][]"

For sourcecode that paren-saving is an advantage. But if i enforce 
that the number of arguments stays the same all thetime, i see no 
problems.

How about generators or dialects, they depend of such feature. I 
mean dialect that don't use parse.

The problem is that you can not assume that the environment hasn't 
change, because if you assumed the you have broken the semantics.

Can you give an example?

Example for which part?

Where changing argument-lists make sense.

APPLY

I can only imagine cases where that is a bug.

With a bit code around?

I can simulate lisp APPLY with a dialect. APPLY require variable 
number of args>

Same for lisp MAP.

In rebol i put those args in a block.

So some of the higher-order techniques.

Lisp does that too in a way, because it puts things always in parens.

Ah. But the point is not that you code in Rebol, the point is that 
if you implement a DIALECT that has different semantics and shares 
the value types! of rebol then you can compile such dialect.

So it would be 
  apply reduce[arg1 arg2]
istead of 
 (apply arg1 arg2)
Or do i miss something?

Yes. Lisp parens are its compilation unit.

And in lisp arg1 none of the args causes evauluation while in rebol 
the may.

I meant ".... value types! of rebol then you CAN'T compile such dialect."

And in rebol we have none when looking at sourcecode.

But actually, when a function runs, its "compilation-units" are always 
the same. Meanswhen a function is run, the lisp-parens can be inserted 
by reflection. (except of strange hacks)

Basically the eval form of lisp is fixed, it is part of syntax, and 
it is (func args ...)

And in rebol it is [func arg arg2 block-of-variable-args]

At least that would work for 95% of rebol.

Not to expensive to get a lot more speed. (interpreter would be always 
available too.)

In rebol the is not special *eval form*, evaluation depends on the 
specific  expression.

But the expression does not change. its "parens" are always on the 
same places, in each evaluation.

do [val1 val2 val3 val4] ?? What this program produce?

So that part can be compiled.

Run it once.

If the parens are different thenext time, there is usually a bug.

Not enough. It depends on the environment.

What if i enforce that, by keeping track and checking somehow? Do 
you have an example wherre that would hurt?

So you will make executing slower, because now the interpreter needs 
to keep track of the whole tree to see which values changed and which 
are a violation of contract.

Options are debug-mode, or pointing out that other things are checked 
too.

Or a compilable-function! , which would be a little bit slower until 
the jit kicks in.

compilable-function! is possible.

That is what rewrite rules and rebcode accomplish.

rebcode would be the cross-platform-target. could be still 10* faster.

Yes. I like that. But this was may point the CFG of rebol is great 
it allows for very neat tricks. But the expense is you can not compile 
easily.

IMHO its main advantage is in creating code and readability. Not 
in kind of self-modifying code.which you do when you turn calls in 
arguments by changing the  arglist-len.

BTW, compilabe- function shares some of the traits of closure. closure 
are a separte function in rebol because they are expensive. In Orca 
we made all funcs to behave like closures, but we decided to factor 
it out like in Rebol to keep the speed gains.

varargs are not the reason for not having a compilable language. 
Both Lisp and C support varargs, and both are compilable languages. 
The culprit is CFG.

From the poing of view of the compiler developer he can't make any 
assumption on how to compila a rebol expression, while in C and Lisp 
he knows that the forms are fixed.