06:30 <mellum> Anybody knows a bit about the internal representation of OCaml?

06:31 <jemfinch_> what do you mean?

06:32 <mellum> I want to know why ints are tagged with a 1, and not with a 0 lower bit.

06:32 <mellum> Because the latter would be easier to implement IMHO.

06:32 <jemfinch_> because of machine alignment restrictions.

06:32 <jemfinch_> the tag is to differentiate ints from pointers.

06:33 <jemfinch_> pointers have to be aligned on 4 byte boundaries, which means the lower two bits will be 0 on every pointer.

06:33 <jemfinch_> and thus, ints have to be tagged with a 1 to let the garbage collector know they're ints and not pointers.

06:33 <mellum> Yes, but the choice which tag denotes what is somewhat arbitrary... one could also tag ints with 0 and pointers with 1

06:33 <mellum> then one could do integer addition directly

06:33 <jemfinch> you could, but you'd have you mask a pointer on every single dereference.

06:34 <mellum> no, you can use a constant offset of -1, which costs absolutely nothing.

06:34 <mellum> On RISC, that is.

06:34 <jemfinch> it would degrade performance by a significant amount -- pointer dereferencing happens *far* more often than integer arithmetic.

06:34 <jemfinch> well, not every machine is RISC, is it? :)

06:34 <jemfinch> in fact, most machines that O'Caml runs on aren't RISC -- they're CISC.

06:34 <mellum> But most *architectures* are RISC, and i386 is a pretty new port IIRC.

06:36 <jemfinch> integer arithmetic probably isn't a priority for the O'Caml folks, and rightly so -- not very many people at all do it.

06:36 <jemfinch> x86 might even be a recent port (I don't know) but the majority of the machines O'Caml runs on are x86.

06:37 <mellum> One could then make this choice dependend on the architecture. Shouldn't be all that difficult...

06:37 <jemfinch> I think you're really making a rather large deal about something that doesn't really matter :)

06:46 <jemfinch> I think tagging pointers instead of ints would, even with that RISC thing, be much more complicated than what they do now.

06:47 <jemfinch> you'd still have to tag every single pointer created -- and that would seriously degrade performance much more than tagging ints does.

06:48 <mellum> Well, that's one opcode per allocation. I'd expect there are a lot more integer ops than allocations.

06:49 <jemfinch> aside from that, it would make integration with C much harder, since (and I'm making assumptions about the behavior of the RISC pointer shift thingy) pointers created on the stack in C code couldn't be derefenced directly (you'd have to increment them first, and C being what it is, that would require a cast to a void* [since otherwise, an increment increases the pointer by the size of what it points to])

06:49 <jemfinch> I'd expect there are more allocations than integer ops, actually.

06:49 <jemfinch> every cons allocates 3 pointers.

06:51 <mellum> Well, dunno... an allocation probably takes a few ten cycles anyway, so it wouldn't matter all that much

06:52 <jemfinch> but that doesn't change that there are probably more allocations than integer ops, I think :)

07:33 <jemfinch> I want compile time garbage collection in O'Caml.

08:28 <jemfinch> (or sml-nj)

18:43 <jemfinch> anyone here?

18:48 <Taaus> Umm... I'm here... For a short while, anyway :)

18:48 <jemfinch> do you know about quotations?

18:48 <jemfinch> (and antiquotations)

18:49 <Taaus> Hmm... I'm not sure... Can you explain what they are? :)

18:49 <jemfinch> nope, that's why I was going to ask you :)

18:49 <Taaus> Hehe, okay..

18:49 <jemfinch> http://cm.bell-labs.com/cm/cs/what/smlnj/doc/quote.html

18:50 <Taaus> I see you were in #ruby-lang briefly :)

18:50 <jemfinch> yeah, I was just doing some research.

18:50 <jemfinch> someone said ruby had a generic api for iteration, and I didn't think it did.

18:51 <Taaus> Well, it does... Using mix-ins from module Enumerable, IIRC :)

18:51 <jemfinch> oh well.

18:51 <jemfinch> I probably dislike Ruby more than I dislike any other language.

18:51 <Taaus> Heh, well... You sure don't seem very knowledgeable about it ;)

18:52 <jemfinch> I couldn't program in it, that's true.

18:52 <Taaus> Let's stay away from the whole Ruby deal this time...

18:52 <jemfinch> :)

18:54 <jemfinch> anyway, (anti)quotations seem to be some sort of powerful preprocessor.

18:55 <Taaus> Yeah.

18:55 <samx> perhaps something similar to lisp quotes and unquotes ?

18:55 <samx> and quasiquotes

18:56 <Taaus> I admit total ignorance af far as lisp is concerned... I'm afraid of parentheses ;)

18:56 <Taaus> af->as

18:56 <jemfinch> :)

18:57 <jemfinch> I definitely think when I get back to programming (which won't be for awhile, I'm killed with other work) I'll be converting to SML.

18:58 <Taaus> Okay...

18:58 <Taaus> I have precious little time for programming myself...

18:58 <jemfinch> yeah, it stinks, doesn't it?

18:59 <Taaus> Nono, I love studying music... If I wanted to program more, I'd study CS, but there are various reasons why I don't do that. :)

19:01 <Taaus> I wonder how difficult it would be to implement a BF parser in O'Caml...

19:01 <samx> so, why are you looking at going to sml?

19:01 <jemfinch> BF?

19:01 <Taaus> Uhhh... Interpreter, not parser..

19:02 <Taaus> BF = BrainF*ck... King of the Turing tarpits :)

19:02 <jemfinch> hehe...it shouldn't be too hard :)

19:02 <jemfinch> heck, you could write it as a compiler pretty easily.

19:03 <jemfinch> samx: well, sml-nj in particular.

19:03 <jemfinch> it has first-class continuations, something I've always wanted to play around with.

19:04 <jemfinch> it also has quotations/antiquotations, which seem interesting.

19:04 <samx> checked camlp4 ? i believe it has quotations

19:04 <jemfinch> I never liked classes in ML, so I don't feel like I'm losing much in that regard.

19:05 <jemfinch> and a lot more research seems to go on in SML -- stuff that'll be interesting to play with. Concurrent ML, Ckit, compile-time garbage collection, etc.

19:06 <jemfinch> it also is a bit more focused on safety -- for instance, the functions that convert from strings all return NONE or Some <type>, instead of just returning <type> and raising an exception on failure.

19:07 <samx> the research features are pretty implementation dependant.. i don't think sml/nj e.g. has regions

19:07 <jemfinch> no, sml-nj doesn't.

19:07 <jemfinch> it's mlkit that does that.

19:07 <jemfinch> I don't know, it just seems a bit more interesting than O'Caml.

19:08 <jemfinch> I don't doubt there are disadvantages (the classes might be useful sometimes; O'Caml has a tighter Unix module, and easier integration with C, etc.)

19:08 <samx> yes, interesting implementations. but pretty much come with the same problems as ocaml

19:08 <jemfinch> but for what I want to do, I think SML will provide a better environment.

19:09 <Taaus> What is it you want to do?

19:09 <mellum> Hmm, some guy here who was at a conference last week says SML is dead, and most people convert to the CAML branch.

19:10 <jemfinch> I pretty much just write irc bots :)

19:10 <Taaus> Hehehe, that's what I thought ;)

19:10 <jemfinch> well, SML still has a stronger hold on universities.

19:10 <jemfinch> (and in commerce)

19:11 <samx> mellum: sml and caml are both pretty research languages, so if you are considering it from the users point a view, both sml and ocaml are dead

19:11 <samx> jem: stronger hold in commerce ?

19:11 <jemfinch> samx: well, there at least has been a commercial implementation of SML :)

19:12 <samx> jem, yeah.. long time ago, and it was discontinued a long time ago too :-)

19:12 <jemfinch> well, I still like SML's safety features.

19:12 <jemfinch> and sml-nj at least is still actively developed.

19:13 <jemfinch> samx: do you know much about mlkit's regions and compile-time garbage collection?

19:13 <jemfinch> specifically, why it's not implemented in other languages -- are there disadvantages to it?

19:14 <samx> so is ocaml.. but looks like most sml people are looking at abandoning sml to a new 'sml' type language.. thought as they are researchers, i'm sure that'll take 10 years before anything really happens :-)

19:16 <samx> jem, nope, not really. but after seeing that even with the regions, it's compiled code is slower that ocaml or some other sml implementations, i wonder...

19:16 <jemfinch> well, I don't think much work has gone into optimizing mlkit like the work that's gone into sml-nj or ocaml.

19:18 <samx> jem, but if you try out mlkit or look into regions, please tell me too ;-)

19:20 <jemfinch> samx: I can't, I don't run linux :)

19:20 <Taaus> jemfinch: You could always switch to Haskell and mess with monads... (Yet another subject of which I know absolutely nothing ;)

19:20 <jemfinch> Taaus: I could, but that just doesn't seem all that interesting.

19:20 <Taaus> Hehe, I can't disagree with that :)

19:21 <jemfinch> first class continuations I find much more interesting :)

19:21 <Taaus> ... Right...

19:21 <Taaus> ;)

20:07 <samx> there isn't a way to init a field in a record type with a pointer to itself, is there? That is, something like ( { my_record = r } as r)

20:11 <jemfinch> no, I don't think so.

20:11 <jemfinch> you could use Obj.magic and then set the point appropriately, though.

20:12 <jemfinch> let r = { my_record = Obj.magic () }; r.my_record <- r; r;;

20:12 <jemfinch> that might work.

20:16 <samx> what does Obj.magic actually do? I've never used it

20:17 <mellum> it's magic :)

20:17 <mellum> A cast operator.

20:17 <mellum> It does nothing except fooling the type system.

20:18 <samx> so, 'Obj.magic ()' would cast the value unit to what ever the inferred type was, in the above?

20:21 <samx> ok, played a bit with it on the top level, got it to dump a core etc.. but i think i got the idea pretty much now :-)