World: r4wp

- not exactly

(that would be quite incomfortable)

Interested in any alternative fixes, but I see no discomfort for 
the above.

Well, there is some discomfort. REB_FUNCTION values are not garbage 
collected.

However, fortunately, the ARGS series is unique to function (since 
it is created specifically for the function), so it can be used instead 
of the body.

That's indeed rather annoying. Seems the GC leaves much to be desired.

(even if two functions have the same SPEC and BODY, they always have 
distinct ARGS)

Ok. That's better (even though it is a somewhat ugly hack).

Well, but the fact that REB_FUNCTIONs don't need GC is not ugly IMO.

One negation too much.

(Meaning: sorry, I'm not sure I understand what you are trying to 
say.)

You mean you dislike it?

Integers, decimals, REB_FUNCTIONS, whatnot... don't need GC.

Only series and GOBs need GC

Quite fortunate that you know already know enough of the GC to not 
mistakenly walk down the wrong road.

I didn't yet check how REB_FUNCTIONs are created, but why should 
they _not_ be GC'd?

Are they pooled or managed using a custom allocator?

Exactly for the same reason why integers do not need GC. They are 
not allocated, so they don't need deallocation

Ah, so they can only be contained within a series.

Ok.

(Or a gob, maybe.)

Yes, the reason why GOBs needed GC was that they did not fit within 
128 bits.

(GOB is 512 bits)

So there's a "gob reference" value type?

Yes

Ah, I see. A "gob value" is just a pointer to the real gob structure.

REBGOB (the part needing GC) is 512 bits, while Reb_Gob (fits within 
128 bits and points to a REBGOB)

REBGBO! :)

Thanks for clarifying.

Yes, sorry, it is just struct Reb_Gob called REBGBO.

BTW, REBGBO looks quite ugly to me

(I mean just the name)

Maybe REBGBI (in analogy to REBSRI) would be better?

Is the "index" field of REBGBO presently used?

yes

you can write:

next gob

So, actually, the "full GOB" is 544 bits, not just 512

(when summing REBGOB and REBGBO while subtracting the pointer.

I am quite curious whether it would be possible to fit a Rebol value 
to less than 256 bits when using 64-bit memory pointers

I originally guessed 160 bits might suffice, but I would not bet 
on it now.

if not wanting to make some "big adjustments", it looks like absolutely 
necessary to go to at least 224 bits.

I'd be quite interested in that as well.

And I'd generally try to stay 64-bit aligned.

...which yields exactly 256 bits :-(

192 or 256, yes.

some "values" contain 3 pointers, which gives 192 + type information 
+ alignment = 256

Hi, all, a "stupid" question: R3 is still called "alpha" (and there 
*are* issues I want solved before moving it to beta). One of the 
issues is the "gotcha" represented by the DECIMAL! name. I know that 
it is used consistently in Rebol, but it is still a "gotcha" for 
any possible newcomers actually stating something like: "here mathematics 
is not welcome", which is not true so much as I (mathematician by 
the education) would say.


Also, having a "truly decimal" datatype called MONEY! in R3, I would 
prefer a rename:

MONEY! rename to DECIMAL!

DECIMAL! rename to REAL! or FLOAT! (or something else that could 
be popular)


So, how many of you prefer to keep the DECIMAL! name for the 64-bit 
IEEE 754 binary floating point format used in Rebol and

how many of you prefer to rename the DECIMAL! datatype to something 
else?

I would not mind this change.

I'm strongly in favour of this change (and would prefer float! over 
real!).

Not a stupid question, a hard one. 

1) Keep money! as it is.
2) Use new decimal type for decimal!  +1
3) Use float! (not real!) as the name for IEEE754  +1