World: r4wp

So, we get both worlds. If we manage to call R3 code from the Red 
section and vice versa, that would be great. We could use the compiled 
speed for inner loops and let the interpreter do all the non-speed 
relevant things.

Robert, I guess you mean Red/System and not Red? Well, the license 
are compatibles, so if someone wants to include Red/System into R3, 
he doesn't need to ask anyone for permission.

Or the other way around, no? With Kaj's system, you can write R3 
extensions in Red/System, no?

Yep, RedCode for R3 is already here

It could be integrated better, but that would require Red to run 
on R3

Kaj - but it is not like Cyphre creted JIT for R2, which looked like 
REBOL and was just faster. Red/System code surely can't run that 
way - it is for offline stuff, where you first have to compile the 
app, no runtime stuff ...

Cyphre's code also needed to be JITed first. I don't see a fundamental 
difference

Pekr: the difference between AOT and JIT compilation is much thiner 
than you think. Just load Red/System compiler code to your R2 app, 
pass it any source code at runtime, use the link?: no option and 
you get compiled code and related data in form of binary! values...and 
voilà! :-) The rest is same as for Cyphre's JIT, you need a way to 
call native code in memory, something that is hardly possible in 
R2, but maybe Cyphre found a hole to achieve it anyway.

something that is hardly possible in R2
 - not a problem

Ladislav: I was thinking about an internal only solution, I know 
it's possible to do it by using a tiny C code in a shared lib.   
        If you've figured out a way to do it without any external 
dependency, I would be glad to learn how you did it.

I know it's possible to do it by using a tiny C code in a shared 
lib.
 - does that look like a problem?

It breaks the "fit all in one binary" REBOL philosophy, requires 
to compile it for each platform and maintain OS-specific code...Not 
a problem per se, but IMHO a sub-optimal solution, that's why I was 
interested in a possibly purely "internal" solution, in case I would 
have missed it.

The link?: no option doesn't seem to be available yet. Is it in the 
unreleased AVR backend?

It's available but not documented as it is only used by the compiler 
internally for now. You can add it to any of the target definition 
block in %config.r for testing (or create a %custom-targets.r file 
instead). It will put the compiler in an "incremental" mode (it can 
compile incrementally as many source file as you want). Once compilation 
has finished, no file will be generated and compiler state will not 
be reset. You can then inspect the result of the compilation from 
console using:

>> probe system-dialect/compiler/job
>> probe emitter/symbols
>> probe emitter/code-buf
>> probe emitter/data-buf
>> probe system-dialect/compiler/imports


You can basically get most of these data in logs when compiling using 
-v 9 option.

Cool, thanks

R3 extensions break that philosophy already, so go for it :)

Does not linking mean that no binary file is output, or does it also 
mean that addresses in the code are not resolved?

Only the latter.

You can link to something that only exists in RAM.

That can't be right

People overestimate what linking means. It really is as simple as 
a pointer, the same thing that it means in a linked list.

Kaj: that's right. If you want addresses to be resolved, we need 
to add a in-memory linker. It will basically just resolve all the 
addresses. But that would not work as is with the imports that expect 
to be statically linked. So, a custom dynamic loader would be required 
(same requirements as for implementing LOAD/LIBRARY and MAKE ROUTINE!).

Not a big deal to implement though, just time-consuming.

I know what linking is; I wrote a relocating loader in Atari 8-bit 
assembler

I don't mean external symbols that are for the operating system to 
resolve, I mean the internal linking that the Red/System linker, 
presumably %red-system/linker.r, does inside a Red/System compile, 
possibly consisting of multiple sources

To put it concretely, the AVR backend needs to generate a memory 
image, without external symbols because there is no operating system, 
but with all internal addresses resolved. Can that memory image be 
constructed by appending data-buf and code-buf?

Nope, you need to do some processing first. All global addresses 
in code are zeroed by default. The linker fills the blank by calculating 
the right code or data addresses.

It's basically what does linker/resolve-symbol-refs

OK. Any idea when you'll publish the AVR backend? It would be useful 
for reference

You mean the linker backend?

Yes, but anything that's fit for inclusion :-)

It's already published since several months: red-system/formats/Intel-HEX.r

Oh, great :-)

As you will notice, it uses a minimal C-based kernel and some imports 
are linked to it using syscalls.

An Arduino? I'll look into it

Yes, that's a kernel for a Arduino Uno board.

OK, great

I would like to get rid of the compiled C part, but never found the 
time to recode it in Red/System. It would also needs some addition 
to Red/System, like interruption handling (already planned) and other 
non-planned features, like a way to initialize RAM/SRAM from Flash 
memory (basically, it needs to copy the firmware data section from 
ROM to RAM), or initialize properly the timer clock (which should 
be doable with the hardware I/O support I've planned already).

In fact, once done, it would be the first nano-kernel written in 
Red/System. ;-)

Thanks, that's useful info

Here are the syscalls that should map to that kernel version (I've 
changed it often):

#syscall [
	init-runtime: 02F0h []
	
	init-serial: 05C8h [
		object	[int16!]
		baud	[integer!]
	]
	prin: 084Eh [
		object	[int16!]
		msg		[int16!]		; string pointer!
	]
	prin-int16: 09FEh [
		object	[int16!]
		value	[int16!]
		type    [int16!]		; unsure about the meaning of this parameter
	]
	prin-int32: 0992h [
		object	[int16!]
		value	[integer!]
		type    [int16!]		; unsure about the meaning of this parameter
	]
	set-pin-mode: 046Ah	[
		pin		[byte!]
		mode	[byte!]
	]
	digital-write: 04B6h [
		pin		[byte!]
		value	[byte!]
	]
	wait: 023Eh [
		delay	[integer!]
	]
]

The kernel does not use pure C, but Processing, a C/C++ hybrid.

Do you have int16! implemented in the AVR backend?

Well, I do have something, but it's messy, buggy and incomplete. 
I can send you the AVR8 backend if you want to play with it, I don't 
want to publish it until it gets a stable and correct support for 
basic datatypes.

I'll keep it in mind, but don't need it now

In Red/System, can I assign a value to a variable of different type? 
say    a: 1 a: "1"

Nope, Red/System is statically typed, you can never change the type 
of a variable. You can just do type casting to convert the variable's 
value to a compatible type.

Jerry, remember that variables are, semantically, just labels to 
variable-size containers in Red/System. The memory model is similar 
to C's one, so totally different from Red or REBOL.

I've activated the doc-strings in all the Red/System bindings that 
I had prepared when I wrote them

If someone writes the doc tool that Doc proposed, you could generate 
separate documentation for the bindings