IRC log started Tue Dec 7 00:00:00 1999 [msg(TUNES)] permlog 1999.1207 Tril: you awake? 12:20am -:- SignOff td: #TUNES (td has no reason) -:- SignOff Tril: #TUNES (BitchX-75+Deb1an -- just do it.) -:- smoke [smoke@16dyn117.delft.casema.net] has joined #tunes hi hey smoke 03:10am -:- SignOff smkl: #TUNES (Ping timeout for smkl[glubimox.yok.utu.fi]) -:- smkl_ [sami@glubimox.yok.utu.fi] has joined #tunes -:- smkl_ is now known as smkl -:- smokie [tw026024@zaalf17.twi.tudelft.nl] has joined #tunes -:- SignOff smokie: #TUNES (BitchX-75+Deb1an -- just do it.) -:- SignOff Fare: #TUNES (Ping timeout for Fare[quatramaran.ens.fr]) -:- SignOff smkl: #TUNES (Ping timeout for smkl[glubimox.yok.utu.fi]) -:- smkl [sami@glubimox.yok.utu.fi] has joined #tunes -:- Fare [fare@quatramaran.ens.fr] has joined #Tunes -:- mibin [mibin@62.11.105.103] has joined #tunes -:- fridge [fridge@wolly28.zip.com.au] has joined #tunes hey what is "orthogonal persistence of objects"? hey fridge what is the context of that phrase? fridge: that a persistant objects should fit in as any transient object in the system For instance, we want to replace filesystems with orthogonal persistence of objects s/objects/object/ you as a programmer should not have to worry about whether the object is over the network, in a database or is in memory 10:10am okay now that makes sense =) and naturally transient objects should be possible to make persitent in a simple way (typo day) did that make sense (despite my poor spelling)? yes thank you -:- fridge [fridge@wolly28.zip.com.au] has left #tunes [] 10:20am -:- tmf [tmf@193.217.63.152] has joined #tunes hi :) hullo what's up? hmm, not much I haven't been around for awhile. Is there any progress? abi, any progress? mibin: i haven't a clue heh :| me neither mibin, what's your intension? i have still not contributed in any way to tunes sigh* are you working on a related project? hmm, not really philosophical programming makes me depressed :) so, you're looking for mp3's :) hmm, not here, hehe no, i dont actually download many mp3s mibin: are you in need of a fun project? :) just a joke. once in a while someone's bewildered into #tunes, lokking for "tunes" sigh* Fufie, whats the fun project ? haha 10:50am tmf: then he asks for midis.. not mp3 :p main() { fork(); main() } mibin, you're confuzing me now, please stop ... bzzzzz tmf: I need more front-ends to a language-tool.. so if you like a non-mainstream language and want to have a call-tree tool, class-browser, documentation etc.. parse it and dump info to xml and you have a nice tool for later use how's the non-mainstream language? because the mainstream languages c, c++ and java are basically already covered ok, what's the back-end? have a look at what it is: http://sds.yi.org/ I will (btw, fufie, I see you're at UiB :) yes :) been there! do i know you in RL? :) RL? real life don't think so. hmm.. ok because you're at ii too, right? I was. Currently I'm working at a computer company here in bergen ok small world :) yepp have you been with tunes for a long time? not really.. are you working on an os thing? not really.. I just work on related higher level things.. distributed hypermedia systems and language-dabbling but I have interest in a decent OS underneath don't we all 11:00am do you dabble with OS-writing? -:- SignOff mibin: #TUNES (Ping timeout for mibin[62.11.105.103]) I'm ocasionally working on a programming language, but currently my day consists of php3 and sql :| is the language getting anywhere? I like to think that it is somewhere. But reaching the final goal is tricky. is the language related to any other language(-families)? lisp, c, pascal, ml? It's an oo language, but it has similarities with list and markup languages It's supposed to be very clean and smooth and everything ( list = lisp :) have you got some examples of code in the langueage? I like new languages :) yes, lets see ... ( there is a meta level and a oo level. an meta level example first: [ # [ 0 < 0 > ] [ 1 < 1 > ] ] [ ++ [ 0 < 1 > # ] [ 1 < 0 > ] ] this defines a '++' operator! 11:10am you can write ( ++ 1 1 0 1 ) to get 0 0 1 1 ( binary with most significant bit last ) :) hmm.. I won't go into detail on that, but I'll write a OO sentence: ( object ( new: (::) !) .) creating a new object named 'name' ( name ( state add: (::) !) .) adding a new attribute to 'name' named 'first-name' The syntax is somewhat wierd ( all languages are ) because the oo level is just the meta level with some new words defined. it seems syntactically to be somewhat lisp-like, why not use lisp-syntax? Because lisp syntax isn't really adequate in itself ( controverial :) (let ((object (new 'name))) (add-state object 'first-name)))) what's the ''' for ? makeing it a symbol (it's typo day for me today) phone.. brb 11:20am but my point is that by using a known syntax you can use existing tools sure but... if you look at the way 'state add' is converted to 'add-state', the conversion looses something important - state is a relation with a method 'add'. There is no independent thing called add-state. add-state was a function-like construct. where most other languages use keywords, lispers usually use function-syntax yepp, but the restriction I'm working under is that _everything_must_be_an_object_. So keywords in the common sence does not exist. that's good but functions are objects, right? 'methods' directly inherit 'object'. I use a prototyping model - no classes prototyping is an ok way to do it yes, I like it because it eliminates the distinction between objects and classes but back to add on state.. from what you say, the lisp-conversion is: (add state object 'first-name) where the method is first and then dispatch is done on all arguments. and I think it iwll work nice in a graphical environment yes, if you use functions, you tend too write functions like fun( mode, parameter ) where you need the function to vary according to context 11:30am ( mode is something that alters the behaviour ) lisp syntax is nice for functions, surely, but objects are not functions. objects can be functions yes, and functions should be objects, but the reverse isn't _neccasarily_ so! are there any cases where you would not need methods/functions/operators/relations? yes! REFERENCES! so, the oo level consists of 'object', 'method' and 'reference' ( in my case anyways ) what does the reference part do? It is a way of refering to other objects. It's basically a path. for example (::) would refere to a specific attribute inside an object. could that not be done by either a function: (attr-value obj 'some-attribute) ? :) (sorry if I sound rude, but I try to understand your language.. and then I become very inquisitive) I like these questions! or (attr obj 'some-attr) if you want to specify the location, have a reference instead of the value of the location/place/attr like: (set (attr obj 'some-attr) some-value) yes, you could do it with that function, if the attribute knew which objects it was contained in. 11:40am attr-value or attr would do a lookup in the object and either return the value or the reference/location but if you use structure/objects, you don't have to do lookup in the whole scope, just the local attributes in the object. that was what I meant basically wheter or not that is an emulation of references with functions or an abstract way to implement references with funtions is question what I' -:- Kaufmann [Kaufmann@200.224.105.142] has joined #tunes m trying to say is that references must be underneath somewhere D00DZ! W07Z UP! K-R4D D00D! *ahem* hi tmf: yes, I agree with you. There must be some way to refer to the location of an object, ie a reference.. but by choosing function-syntax for it you avoid lots of parsing hi kaufie :) hey Fufie tmf: keep the language as simple as possible :) what's the discussion? we're just discussing new ways of how to express language concepts ah... groovy :) and I have been Scheming too much lately and want everything really simple ie as few obstacles and special-cases as possible What do you mean? good question :) 11:50am what I mean is something like: See, it's true... nobody in this field ever has any idea what the hell they're talking about :) brb foo = new Object; is basically an operation/function.. (set foo (new Object)) mmmm hmmm there is no need for special parsin for '=' and 'new' except for the infix syntax (I have also been hacking too much on a front-end for c++ and really really hate parsing ugly languages when it can be done in a simple way) yes.. except for the infix syntax.. and the ending semicolon ;) but is the infix syntax and the ending semicolon necessary when it creates obstacles? Of course not... that's why we have SEXP ahh.. you are enlightened ;) the best syntax is no syntax at all a true DWIM? :) well, of course :) what I mean, though, is that SEXP is good exactly because it frees the programmer from the syntactic details and frees him to care about the program's semantics, which is of course what is actually important unless you're a language implementor :) especially if you are a language implementor.. :) heh of course, parsing is the easy part with sexp you basically have the AST right away 12:00pm then you can actually start doing the real work fufie: infix syntax: (like I now should patch this c++ code and hate it) a=b+c in oo: a.=( b.+(c) ) LOL just loose the dots! yes.. but in prefix you get: (set a (+ b c)) no dots: and then b and c are dispatched upon and the correct + is found that is oo or in OO perl: set $a plus $b $c; a = ( b + ( c ) ) there is a limit to how simple syntax can be why are we discussing this anyway? religion, my dear Kaufie :p ahh no.. just kidding.. dammit, just read in the programs as lists and get it over with :) In the beginning there was the word .- lists? but I just thought that thomas' syntax was too xomplex.. but what is obviously the interesting part is his meta-language.. s/xomplex/complex/ this is typo day the meta-language is way simple! (I also think that Thomas has read some Beta literature :) I think not it seems so far to go in the same direction There are dusins of oo like languages but please tell me more about the meta-language.. ok, I consists of 3 things: words, contexts and meanings words have meaning dependent on context Thats the most clearifying thing I could say about it 12:10pm that was enlightening.. :D think, about it - what more is there to (programming)laguages? a lot like what? like right now I work on a meta-language at type-level where dispatch in methods is also done on sets just doing dispatch on individual objects is too limiting for my language tool and so I needed to add to the language I am writing in the simplest case is the OR.. imagine that you want to make a method ADD where + is already defined and works on integers and reals and you want to use that code but also want to have complex numbers brb Fufie, pattern-matching? then you might want to define methods over OR-sets pattern-matching is a Good Thing (defmethod ADD ((obj (or integer real)) (other-obj (or integer real))) (+ obj other-obj))) and that is the same as writing four methods if you would have to write it on each combination and then you have much less to maintain.. and then you can concentrate on e.g (defmethod ADD ((obj complex) (other-obj integer)) ...) Fufie, how do you overload the meaning of the or form in this case? I mean, or is usually the boolean inclusive-OR combinator, not the union-set constructor defmethod is a macro so I parse OR and give it special meaning here.. it is not the usual OR ah what I say is that the first method should be used if the first argument is integer or real and the second argument is integer or real. four combinations 12:20pm yes, I understood that basically, pattern-matching the problem is to define a total-order on how the dispatch should be done in a way it is pattern-matching so... for OR it is simple because it expands in this case to the four defmethods and they all call a helper-function brb back overloading is nice why is this not included in "context, words and meanings" ? 12:30pm everything is included in "context, words and meanings" including most natural languages and verbal languages most? I haven't heard them all.. I am pulling a Hume here :) I don't think you need to How is the type language you're designing? it's basically a gedanken-project still with an implementation of the OR case which saves me a few hundred lines of code ( hahaha - I got it, "pulling Hume". I thought you meant humor/joke :) (I meant David Hume, the philosopher ;) right the OR case, as you put it, is an interesting point of view. I deal with the problem differently. but Fare provided me with some links, and I hope to add more of this functionality to my basic toolchest Isn't it better simply to modify the language behaviour so that the generic function does method-dispatch by pattern-matching? E.g., say we have a function FOO that has four methods associated with it: M1 for when it's called with two arguments, both of type real or integer, M2 for one argument of type string or symbol, M3 for any number of float arguments, and M4 for anything else. (cont') (cont'd) So there are three patterns (in regexp-like syntax here): /(Real | Integer) (Real | Integer)/ -> M1; /(String | Symbol)/ -> M2; /(Float+)/ -> M3; /.*/ -> M4. So the generic function would keep a table of these relations, and when it was called, it'd try to match each of these patterns to the arguments, in order of specificity (sp?); the method associated with the most specific one that matched would be dispatched. yes that is what I am aiming for the problem is finding the best match The way I do this, is that all objects must be well defined. there's plenty of literature on pattern-matching (especially regexp-related) around tmf: what is well-defined? kauf: but does it take into accound subtypes? account 12:40pm Of course, the concept not restricted to types... ideally, you can specify a method's arguments to any level of detail. For example, the factorial definition: (define (fact 1) 1) (define (fact (:: n Integer) (* n (fact (- n 1))))) well defined simply means that something is valid in all possible situations. (wherein :: is the "is of type" operator) kuf: I got that and I agree with you on that, and it is mostly provided for in the system I use.. kaufmann, I like the 'fact' definition e.g you have subtype dispatch and EQL dispatch: (defmethod FACT ((obj (eql 1))) ...) (defmethod FACT ((obj fixnum)) ...) Fufie, of course but the first step is to add logical operators as well But general, standard pattern-matching already provides for logical operators unfortunately the language I use doesn't support that.. yet but you're right I want a pattern matcher that allows subtyping as well 12:50pm christmas is comming up quickly. mabe santas got a pattern matcher for you:) "wow, thanks mummy.. how did you know that I wanted a pattern-matching system on types.. and it also supports subtypes" :) great! Well, if, say, EvenInteger is a subtype of Integer, the pattern matcher will check the class' inheritance list, in effect asking the instance of EvenInteger 'are you an Integer'? Of course, this could be extended in a less strict OO environment, where individual objects can have properties besides those of their class LOL Fufie but if eveninteger inherits integer, then eveninteger can always replace integer anyway! and integer can only replace eveninteger if it is even. Yes... every EvenInteger is an Integer, but not every Integer is an EvenInteger what I said. Where does pattern matching come in? it is basically a pattern I am going to match with the arguments and unlike static languages I can do this at runtime In a sufficiently advanced environment, you could simply tell the system rules such as "all instances of Integer that are divisible by 2 are also instances of EvenInteger; all that are not are instances of OddInteger". This goes beyond the common, restrictive view of OO that prevails in current systems. as in formal and actual parameters? ( I can do runtime too :) kauf: yes, that's the idea and why the TYPEP function and DEFTYPE in Lisp will be a help.. kaufmann, but in a not common and restrictive language you could do this without pattermatching, right? ie one can dispatch on a type like: (or vector (and list (not (satisfies circular?))) ie a type which is either a vector or a non-circular list tmf, of course... all I'm saying is only tangentially related to pattern-matching itself 01:00pm kauf, yeah, but as I understand, it is somehow needed in the situations you outline? you will need an inference engine or a pattern matcher why? Fufie, ideally both - and good ones at that because what is sketched here requires more or less a small programming language and accompanying compiler/interpreter to do inferencing pattern-matching because most languages have _many_ types floating around This is about unifying functional and logic programming in a big way. so you get a meta-language pretty quickly kauf: I have a pointer to a PhD which also discusses this if you're interested ftp://progwww.vub.ac.be/dissertation/1998/ tmf, to answer your question, in a situation like the factorial definition above, you need a way for the generic function to find out which method it should dispatch. Most languages that use method-dispatch just go over the methods' parameter type lists one by one and compare them with the arguments' types, which is a very limited form of pattern-matching. thanks, Fufie well, I've got to be going now... see you all later... my email is rnedal@olimpo.com.br, in case anyone cares have fun ok, so basically what you want is to have the formal parameters as part of the function identity? bye -:- SignOff Kaufmann: #TUNES ("They say the lights are always bright on Broadway" (Genesis)) I see what you mean, I'm just trying to put it in context. function identity is somewhat ambigious at least to me basically regular languages do dispatch on simple and single types instead of more complex types. This causes code bloat in certain cases and the ability to express more clearly which function/method should be called is a Good Thing I mean, normally you talk about the function 'add' for example. But you want to distinguish between add(real, real) and add(imag,imag) 01:10pm yes that is already implemented in the language and along with this distinction comes disitions ro ambiguities about which function to pick. but I want to avoid repeating code by cut'n paste along comes ambiguity yes :( totally - cut'n'paste sucks mbiguity... can be eliminated if you have an ordering of functions - so you can allways pick the "closest" function -:- mibin [mibin@62.11.104.143] has joined #tunes yes, but defining that order is the hard part and it can't probably be done to remove all ambiguity but the thesis mentioned says that this is really ok because you are able to express much more ok, but to help you you have type hierarchies. I have type-hierarchies and I have multi-dispatch and I have EQL-dispatch can I outline the way these cases are dealth with in my oo language? can you please do it in a mail.. irc is difficult because everything is so constrained and short little time to think through as well I know. I'll send you a mail. is this address good ? stig@tunnel-44-14.vpn.uib.no no stig@tunes.org or stig@ii.uib.no ok does your language have a name or a home page? (I need a hash key for my brain :) no homepage. I call it TOOL - The Object Oriented abstraction Layer hows' that for a key a good key just think - that was that there good TOOL this guy was talking about :) can you rephrase the last sentence, please? * Fufie/#tunes puts on tool on the stereo 01:20pm i like tool. what cd is it? undertow could not find �nima right now with the pig? yes brb -:- SignOff smkl: #TUNES (Ping timeout for smkl[glubimox.yok.utu.fi]) -:- SignOff mibin: #TUNES (Ping timeout for mibin[62.11.104.143]) -:- mibin [mibin@62.11.104.143] has joined #tunes 01:30pm have you been working on some eccentric projects lately? -:- Miellaby [sgarden@ppp-44-45-3.wmar.club-internet.fr] has joined #TUNES Hello hi What's the topic today ? 'lo miel =) how are your creatures miel ? ;) as always, tunes =) 02:00pm -:- smkl [sami@glubimox.yok.utu.fi] has joined #tunes my creatures are very well. I though it would be fun if they wre sexued. heh i was referring to your AI project me too.:) ah ok =) -:- SignOff smoke: #TUNES (Ping timeout for smoke[16dyn117.delft.casema.net]) The manager could make its creatures to breed. A gene 'crossing over' would enable to create news agents with mixed properties. abi, waporware? it sounded to be excellent for genetic programming at least mibin: wish i knew abi, vaporware? mibin: bugger all, i dunno hmm, i would it even allow breeding between already existent apps =) ? s/ i // (btw i would never breed emacs and netscape) -:- smoke [smoke@16dyn117.delft.casema.net] has joined #tunes 02:10pm :). It supposes that the standardized interface of agents provides fonctions to dump parents' pseudo-genetic raw material, then to create a new instance of species with melted data. hmm, real living dna contains lotsa 'junk i think that that part of dna serves to avoid crossover points to happen in the middle of useful genes what would be the equivalent of a gene in your project? some sort of bytecode? I think. and the equivalent of proteines and enzimas? To be more simple : A binary translation of all constant rates which defines one agent behaviour. For example, the number of neurons in its network. I prefer not to speak about equivalency. (sorry for the spelling) hmm, i dont think i have understood it fully but it doesnt matter anyways, as you may have an implementation already in mind 02:20pm Well. You take every data defining an agent and concate them to build a binary sequence. Then you melt two of these sequences by cutting them at the same level and taking one part of each. just like a cross-over, yea You can also implement mutation. That is apply a little noise on the result. mutation doesnt work IRL it may wark to optimize some parameters but not to create new genes is that your opinion ? i mean, it can create different genes Mmm. I see the difference. but rarely genes for different purposes I suppose to make a real purpose to be born requires a reflexiv definition of the entity :) that would really require a lot of computing power everybody here agrees? * mibin/#tunes suspects not Reflexivity ? it will requie a decent beowulf or lots of time to make programs genetically Miellaby: wow, i agree 02:30pm All my aborted projects drive me to tunes =). Miellaby: do you mean that the creature would have to have self consciousness for that to work?