World: r3wp

Stack looks useful... though as I mention in my article I sort of 
feel that it's dangerous to do it that way, and am ok with narrower 
contracts

Erm... I still can't quite figure out how to conveniently get my 
assert routine to access the system/script/header from the caller, 
without writing:

assert system/script/header 1 != 0

I'd like some shorthand FOO so I don't have to type system/script/header 
each time, so I'd settle for:

assert FOO 1 != 0

But if FOO is a function that lives in debug-functions.r then I'd 
need some parameter trick to access the calling context, the way 
bind uses a known-word for example.  Is that the case?

That won't work, even if you specify system/script/header at the 
call site, unless your assert statement is run during the execution 
of the script in question. Once a function is loaded, its source 
is gone - system/script/header refers to the currently running script. 
The only way to get your assert system working is with a preprocessor.

Ah, well, I was thinking of doing that anyway.

REBOL doesn't have declarations - it has expressions that build functions 
and assign them to variables at runtime.

You might consider extending prebol.r from the SDK - it has all of 
the preprocessing infrastructure built in. All you would have to 
do is add a #assert directive that would be replaced at preprocessing 
time with the info you want to log.

An option I considered was a refinement to assert, called /loc (for 
location) and let you pass in an identifying string for that assert. 
 Then, a tool which goes through all the asserts in the code and 
if it doesn't have a location, add the refinement and a parameter 
based on a GUID.  It will also make a database of file/lines for 
each location.  Thereafter programmers can move them around without 
much trouble.

Does R3 have guid! ??

You can make your own using binaries or strings, but nothing built 
in.

Yet - user-defined datatypes aren't done yet.

In your case, filename, line and position on the line would probably 
be sufficient.

Brian (fork); :)  Check out rebol.org - pager.r for an attempt I 
made at keeping track of source line numbers. Again, some REBOL scripts 
are built without newlines.  Best to keep index? values.  imho.  
It's what I use for the word cross referencer I'm working on  (forth 
style locate and tour)

As far as I can figure, the only things in REBOL that cares doodle 
about newline is; the semi-colon comment, that newlines are retained 
in multi-line braced strings (and kinda in blocks)  and  a newline 
breaks a quoted string.

You can just call me Fork, that's fine, no offense taken.  :)

Seems to keep things clearer... and I can feel all hacker elite to 
be one of the few on here with a handle.  :)

Only add fork that time as this is a three-way Brian chat.  :)

--[[  a Lua multi-line comment  --]]   but adding a hyphen   as in 
---[[  lets the commented block run and the --]] can remain in place 
as both bookends are now just -- single-line comments; really quite 
clever.  Rebol could use ;{  and ;}  where ;;{ opens up the 'commented' 
lines  to run (not a block comment  at that point) but leaves the 
'markers' in place until debugging or profiling or unit test or whatever 
is completed;

dumb question:

>> a: make object! [b: 'c]
>> mold a
== "make object! [^/    b: 'c^/]" ; c is lit
>> mold/all a
== "#[object! [^/    b: c^/]]" ; c is not lit


I guess the question is why c is not lit for MOLD. It means that 
when you probe an object , words in an object appear lit, when they 
are really not. (Are they?)

sorry, it should be "why is c lit for MOLD"

make evaluates the object spec block. Therefore, the lit-word is 
transformed into a word.

mold/all, on the other hand, does not evaluate the object spec block.

compare
	probe load mold a
	probe load mold/all a

They both load the same.

in your first reply, did you mean "mold evaluates the object spec 
block"?

no, I definitely meant make.

Try this:
	make object! [a: 1 + 2 print "hello"]

then I don't understand: "mold/all, on the other hand, does not evaluate 
the object spec block.". should that not produce a lit-word in the 
object?

You can see both the expression 1 + 2  and  print "hello"  are evaluated.

Sorry, I meant when you evaluate the result of mold and mold/all.

What doesn't make sense to me is that MOLD presents the lit-word. 
I don't really care that they mold and load the same and what is 
evaluated after MOLD. It's MOLD itself that bothers me. :-)

Ok, first of all, trust me that the way it is is just perfect. Now, 
I have to think how to explain it well..

A first explanation attempt :-

MOLD's basic task is to provide the specification for the value being 
molded, so that, when DOne you get the original value.

MOLD/ALL's basic task is to provide the specification for the value 
being molded, so that, when LOADed, you get the original value.

The difference is, 

 do [ make object! [...] ]    will evaluate whatever is in the spec 
 block, whereas

 do [ #[object! [...] ]          will not, no matter how many times 
 you try it.

The first one relies on 'make being evaluated as the default native 
MAKE function that we all know and love, and the second one does 
not. A simple load is all that is required to build the object.

MOLD/ALL works straight from the the object itself, whereas MOLD/ALL 
goes "back further", returning code which *produces* the specification.

The best explanation should probably come from a sound definition 
of mold and mold/all....

you wrote MOLD/ALL in both cases in the line above.

I guess you meant "MOLD goes back further"

As I read here, I understand it. I also think that it would have 
been better for MOLD/ALL to be called SERIALIZE, because it is not 
an extension of MOLD, but perhaps a different mode. But I bet that 
Carl already has thought about this.

Ingo, yes.

As I remember, there was some debate about what to call mold/all 
when it was developed. We could probably just search some archives 
to find out...

SERIALIZE implies to me that there would be a target to serialize 
to, which is SAVE.

i.e. SAVE/ALL

disagree - check the expression "serialise to disk"

serialise means "turn into a series" ie. a string. That's all.

I have been thinking about this for sometime.  I'm thinking we need 
an alternative set of mezzanines for those that need performance 
and want to build something off the /base product.  Problem with 
existing mezzanines is that they need to maintain backwards compatibilitiy 
which means we may lack some of the performance we might gain from 
a current alternative.  The goal of the new set of mezzanines is 
to be driven from the most recent distribution of the REBOL platform 
R2.  Obviously, R3 should be accomplishing this task inherently.