World: r3wp

with liquid the difference is staggering when over 10000 objects.

well, I doubt that I would have more than a few hundred objects... 
but ...

unclear as to why yours is shared and mine isn't

but its a large object.  still using the class system, I can allocate 
1000000 nodes using about 400MB.  with instances, 10000 obects take 
much more than that.

R2 shares objects within objects. the class is not re-created at 
each make [].

Oh ....

is it objects within objects within objects?

But then the class functions are accessed through obj/class/method 
obj, insstead of obj/method.

so if I have

pharmacy: make object! [
	address: make addressobj [ ]
]


and the address object has objects in it ... they will be shared??

Yeah.

Ouch

So, how does one create nested unique objects?

Create them using the init function ??

With explicit construction in the spec block, like this:
a: make proto [
    b: make inner-proto [...]
]

Every MAKE object! takes a spec block that is an init function, in 
effect.

so , instead of 

a: make pharmacy []

I have to 

a: make pharmacy [ address: make addressobj []]

If that is the case, perhaps I need a 'create function with each 
object, and at init time, iterate thru all the objects calling their 
create function ?

As you can see, I've never played with rebol objects before ...

If you like. The standard way of doing things in REBOL is to have 
functions that create objects, rather than constructor methods.

which is the more robust method?

In terms of maintenance?

OOP in a prototype-based language with function values instead of 
methods is different. Classes are emulated if need be, but don't 
always need to be. In REBOL even delegation is explicit, unlike most 
other prototype-based object languages.


For the best maintenance use factory functions that create objects 
based on standard specs. Beyond that, different models are better 
for different tasks. Sometimes you assign function values to fields, 
sometimes you use class objects, sometimes class names that are looked 
up at runtime.

I usually creat myself a new function which calls make and the  object 
init (wether class or prototype based)  if that init needs to create 
new inner objects, then its responsible for doing so.


in your case the make-xml-object  could accept an xml string and 
then call make-xml-object recursively for each element it finds.

I guess I can set the init functions to none after they've done their 
jobs ?

not much would really be gained, but it migh trigger some GC cleanup.

i wouldn't have 100s of init functions anymore .. :)

Using your method of creating a special context is going to screw 
up my obj2xml function :(

One way to do init is as open code in the object....That only "inits" 
the original object, not anything MAKEd from it. But it may be useful 
in some cirumstances, and it does not become part of the object:
    o: make object! compose [print "init" (a: 99) a: 1 b: 2]

huh?

Code in the spec block doesn't get attached to the object.

thats another ways of doing it... instead of storing reference objects 
you store reference spec blocks.

that is how its done in GLayout to assign the same setup to many 
styles.

The spec block is like an init function that goes away after it runs, 
without needing to assign none to any fields.

so graham... instead of doing: 


; note this is an object, context is a shortcut for make object! 
,  like func is a shortcut for make function!
addressobj: context [
	number: 666
	street: "styx lane"
	city: "pandemonium"
]
pharmacy: context [
	address: make addressobj [ ]
]

you do:

; note this is a block
addressobj: [
	number: 666
	street: "styx lane"
	city: "pandemonium"
]

pharmacy: context [
	address: context addressobj
]

I'm having to create nested objects 7 levels deep ...

in the later, you can add code in the addressobj which will be executed 
everytime you create an object using it.

for example: 
addressobj: [
	number: 666
	street: "styx lane"
	city: "pandemonium"
	address: rejoin [number " " street " " city]
]

does the xml structure change a lot (lists of data, alternative or 
optional elements, etc) or is it really static?

static largely

cause with rebol its easy to hack stuff up using out of the box tricks.....

something like...

>> probe load replace {#[object! [a: #[object! [b: #[object! [c: 
"%VALUE%"]]]]]]} "%VALUE%" "tadam!"
make object! [
    a: make object! [
        b: make object! [
            c: "tadam!"
        ]
    ]
]

;-)

A hack.

The XSD I'm looking at has about 30-40 complex data types

my client has a few thousand data types to cope with ;-) rebxml allows 
you to load that up pretty quickly...but not using the xsd obviously.


its parser builds up a string and then loads it.   I found it to 
be quite fast actually.  its MUCH faster than firefox, for example.

I'm creating the XML from scratch .. and not reading it in.

Maybe I need to use the factory way ..

you only need to output xml?  then that is very easy.

MUCH easier than loading it.

what is the source of data?