#clojure – 2009-07-16

2009-07-16

00:30	Knekk	bleh, using the Emacs Starter Kit messes up my arrow keys
00:44	ataggart	is there a macro somewhere that will evaluate forms until one returns logical true?
00:45	ataggart	I seem to recall it, but can't find it
00:49	ataggart	aha! (some eval ['(foo ...) '(bar ...) '(baz ...)])
00:50	skalnik	Learning Clojure has given me a headache :D
00:50	ataggart	thats the feeling of new pathways being burrowed through your brain.
00:50	skalnik	yep!
00:51	skalnik	I've given up on Project Euler problem #2 for now :)
00:51	ataggart	do you have a fib function?
00:52	skalnik	yep, that works :)
00:53	skalnik	I got stuck on trying to create a set of all the fib values under 4 million
00:53	ataggart	(reduce + (filter even? (take-while #(<= % 4e6) (fib-seq))))
00:53	ataggart	booyah
00:54	skalnik	Haha
00:54	skalnik	Why reduce instead of apply?
00:54	skalnik	What's the difference
00:54	skalnik	(if any)
00:55	ataggart	lazy
00:55	skalnik	ah
00:55	ataggart	I think
00:55	ataggart	hmm
00:55	ataggart	no cant be about laziness
00:56	ataggart	since you have to run thru the whole list
00:56	ataggart	but "reduce" is more desciptive
00:56	skalnik	Now I just need to generate fib-seq
00:56	skalnik	I agree :)
00:57	ataggart	ah ok, apply would work but only because + can take any number of args
00:58	ataggart	reduce is when you have a function that takes two args
00:58	ataggart	and crepeatedly applies it
00:58	skalnik	ah ha
00:58	skalnik	So what I want to do for fib-seq is (map #(fib %) (range 10)), but replace 10 with a very large number
00:59	skalnik	However, that seems wasteful
00:59	ataggart	hmm not sure how you're trying to apprach it. doesn't sound right
01:00	skalnik	so I have a fib function, so if you call (fib 3) it gives the third fibonacci number
01:00	skalnik	The next step I wanted to do is create a set of fibonacci numbers less than 4 million
01:00	hiredman	someone should just write a canonical fib function so everyone can get on with their lives
01:01	ataggart	no one that needs to write one is doing for any reason than to learn. be gentle.
01:02	ataggart	skalnik, the thing is you don't want a function that takes an index
01:02	ataggart	you want a function that will return a sequence of all values
01:03	skalnik	I suppose
01:03	skalnik	I haven't learned that much though, and I'm quite tired, plus I have surgery tomorrow, so I should be heading to bed :)
01:03	skalnik	Thanks for the help ataggart
01:04	skalnik	also, my method works, just seems improper to me :)
01:04	ataggart	your intuition is correct ;)
01:09	skalnik	Very cool that no matter what large number I plug in, execution time is still roughly the same :)
01:09	skalnik	now, I really should be off to bed.
03:08	hiredman	clojurebot: CA
03:08	clojurebot	CA is Contributor Agreement: http://clojure.org/contributing
03:27	ataggart	mailed mine today
05:27	Jomyoot	(with-query-results rs ["select * from blogs"]
05:27	Jomyoot	Does that have lisp syntax for select statement?
05:27	Jomyoot	http://en.wikibooks.org/wiki/Clojure_Programming/Examples/JDBC_Examples shows lips syntax for other commmadns but not select
05:55	Jomyoot	How do I output Clojure's vector of int from Java?
05:56	Jomyoot	I am in Java. I want to output Vector of int, that is usable by Clojure
05:56	jdz_	use iterators?
05:56	jdz	or you can convert a vector to array
05:57	Jomyoot	Clojure has [1 2 3 4]
05:57	Jomyoot	vector
05:58	Jomyoot	I want to output [1 2 3 4] from Java object. that will work as [1 2 3 4] in Clojure with no conversion
05:58	Jomyoot	Is that possible?
05:58	jdz	what do you mean by "output"?
05:59	Jomyoot	return
05:59	Jomyoot	return from Java method
06:00	jdz	Clojure vectors are Java objects (as any value in Clojure)
06:00	jdz	you can pass them and retourn to any Java methods you want
06:00	Jomyoot	hmm
06:00	jdz	I don't really understand what you want
06:01	Jomyoot	Vector<Integer> vect= new Vector<Integer>(100); return vect;
06:01	Jomyoot	How would Clojure see vect?
06:02	jdz	as an instance of Vector class
06:02	jdz	Java Vector class
06:02	Jomyoot	hmm
06:03	Jomyoot	What would I output from Java then If I want Clojure to see as native [1 2 3 4]
06:03	Jomyoot	Clojure's native
06:03	rottcodd	Jomyoot: I think you want to create a clojure.lang.PersistentVector
06:03	jdz	,(class [1])
06:03	clojurebot	clojure.lang.LazilyPersistentVector
06:05	Jomyoot	Is it a standard practive to create a LazilyPersistentVector in Java. then return to Clojure?
06:05	Jomyoot	As means of interop between java/clojure?
06:06	jdz	k
06:06	jdz	sorry
06:06	jdz	(that was C-x k in the wrong window)
06:07	jdz	Jomyoot: you can use all the classes and interfaces in src/jvm/clojure/lang from Java
06:09	Jomyoot	Arrays
06:09	Jomyoot	Clojure supports the creation, reading and modification of Java arrays. It is recommended that you limit use of arrays to interop with Java libraries that require them as arguments or use them as return values.
06:09	Jomyoot	Why ? what's wrong with using arrays?
06:11	jdz	they are not persistent
06:11	jdz	and there is nothing wrong with them
06:11	jdz	you just have to be very careful if you are using them
06:11	jdz	that's why the rule of thumb of using them only for interop holds
06:13	Jomyoot	Is there easy ways to quickly convert array of int or vector of int to Clojure's native data?
06:16	rottcodd	vec
06:16	cark	or seq
06:17	Jomyoot	seq
06:17	Jomyoot	i want to make seq out of java array of vector
06:17	Jomyoot	possible?
06:17	Jomyoot	must be done manually?
06:18	Jomyoot	or is there built in method
06:18	cark	just use the seq function on your array
06:19	Jomyoot	seq also works on Strings, native Java arrays (of reference types)
06:19	Jomyoot	what is reference types?
06:19	Jomyoot	is int = reference types?
06:19	jdz	,(map (partial + 1) (seq (into-array [1 2 3 4 5])))
06:19	clojurebot	(2 3 4 5 6)
06:20	cark	i think in this context reference type means objects as opposed to primitive values
06:20	Jomyoot	hence array of integer will not work
06:20	jdz	i just showed you that it works
06:21	cark	,(seq (make-array Integer/TYPE 10))
06:21	clojurebot	(0 0 0 0 0 0 0 0 0 0)
06:21	jdz	in the example above a native Java array is created from Clojure vector, and then back to a sequence
06:21	Jomyoot	is 1 2 3 4 in Clojure represented as int or Integer in Java?
06:21	cark	it depends
06:22	cark	these are boxed before doing function calls, but could be native integers inside a function
06:22	cark	also you can use native integers in java array
06:23	cark	so there would not be any boxing on function calls i guess
06:23	ChrisPS	,#'in-ns
06:23	clojurebot	#'clojure.core/in-ns
06:30	cark	Jomyoot : native integer arrays do work with seq ...but the seq will contain boxed integer in the end
06:30	cark	,[(into-array [1 2 3 4 5]) (make-array Integer/TYPE 10)]
06:30	clojurebot	[#<Integer[] [Ljava.lang.Integer;@1b206f> #<int[] [I@ee90e9>]
06:31	Jomyoot	well then
06:31	cark	as you can see the second form creates an array of native ints
06:32	Jomyoot	does it matter in clojure wheter it's native or boxed int?
06:32	cark	no more than it matters to java
06:32	Jomyoot	do clojure functions that require ints care?
06:32	Jomyoot	well then it does matter then
06:32	cark	the only difference to my knowledge is performances
06:32	Jomyoot	some methods expect boxed some expect native
06:33	Jomyoot	i would assume that most clojure functions expect either native or boxed but not both
06:33	cark	have you an example of a java function that expects native ?
06:33	cark	i'm not knowledgable at all on the java side of things, looks like something interesting to test
06:34	jdz	many methods that accept characters also accept ints
06:34	Jomyoot	more methods expect int[] than Integer[]
06:34	jdz	like, Character/isDigit or something
06:34	Jomyoot	more methods expect int than Integer as well
06:34	Jomyoot	and all number literals in Java are native
06:34	Jomyoot	not boxed
06:35	cark	,(Character/isDigit (new Integer 3))
06:35	clojurebot	false
06:35	cark	that's a boxed integer i'm passing there
06:36	cark	it is my beleif that the vm takes care of the conversion
06:36	Jomyoot	,(Character/isDigit 1)
06:36	clojurebot	false
06:37	jdz	,(Character/isDigit (int \5))
06:37	clojurebot	true
06:37	cark	though i don't think that applies to arrays
06:37	keca	,(Character/isDigit (int \5))
06:37	clojurebot	true
06:37	cark	,(Character/isDigit (Integer. (int \5)))
06:37	clojurebot	true
06:37	grrrt	,(int \5)
06:37	clojurebot	53
06:37	jdz	5 is the number :)
06:37	keca	,\5
06:37	clojurebot	\5
06:39	grrrt	isDigit has two implementations: one with a char and one with an int
06:39	grrrt	the int version assumes the int you pass in is a unicode codepoint
06:39	cark	anyways the easy way to get an int[] array from a vector is this : (into-array Integer/TYPE [1 2 3 4 5])
06:43	mikem`	Hi, i'm trying to use a class from a jar file at the REPL in vimclojure, but I keep getting FileNotFoundException. I'm new to clojure and Java, can someone help me get started?
06:44	grrrt	if you get a FileNotFoundException, you're trying to load the class as a file?
06:45	grrrt	it should be on your java classpath, then you can use 'import' to use it in your namespace
06:45	cark	you need to start java with the path to your jar added to the classpath parameter
06:46	mikem`	grrrt: I'm trying to load HttpPost from org.apache.http.client.methods (for example). I try (use 'org.apache.http.client) and get the FileNotFoundException
06:46	mikem`	ah, import... ok, let me give that a shot
06:46	grrrt	yeah use is for clojure namespaces
06:48	mikem`	grrrt: ok so I need to specify the .jar file on the classpath when I execute the nailgun server (for vimclojure). do I have to specify every .jar file or can I specify a directory with .jar files? the relevant jar files are in my pwd
06:49	grrrt	every jar file
06:49	cark	you can use wildcards for jvm 1.6
06:49	grrrt	ah yes, that's true
06:49	grrrt	heh. spent too long on old jvms :(
06:50	cark	here is how i start java on my current project : java -server -Xmx300m -XX:+UseParallelGC -cp c:/home/cara/clojure/WebTL/lib/*;c:/home/cara/clojure/WebTL/src/ clojure.main
06:50	mikem`	grrrt: ok, that's what I needed: use import and specify all jars on the command line. thanks!
06:50	grrrt	no worries, have fun!
06:50	cark	note that there is only one star
06:50	mikem`	cark: the wildcard doesn't seem to work for me
06:51	cark	what's your jvm version ?
06:51	grrrt	mikem`: if you do 'java -version', what do you get?
06:51	mikem`	1.6
06:51	grrrt	hm
06:51	cark	i swear it works =P
06:51	mikem`	1.6.0_14
06:51	mikem`	cark: I'm probably doing it wrong...
06:51	cark	i have like 8 libraries in there in use for this project
06:52	grrrt	you'd normally start your stuff from a shell script, ant file or maven file anyways
06:52	grrrt	you wouldn't use a plain "java ............ clojure.main" command every time
06:52	cark	ahwell the end product is an executable jar
06:53	cark	so there's no scrip, and no customer mess up =)
06:53	grrrt	heh
06:54	mikem`	cark: ok, with this command, everything works: http://dpaste.com/67796/
06:55	cark	great then =)
06:55	cark	though it might be worth working out the wildcard thing to keep things simple
06:57	mikem`	ok, will try that after some food. thanks for the help so far
07:03	grrrt	hmm there was a trick to using names that were defined as private, but what was it? I'm trying to access a def that is private from a test-is testcase (in a different file)
07:07	grrrt	Ah, got it. to answer my own question: http://groups.google.com/group/clojure/browse_thread/thread/3835e5405ab930f6/
07:50	Jomyoot	Can file a.clj (:use b) and b.clj (;use a)?
07:51	cark	nope
07:52	cark	that's bad practice anyways
07:52	Jomyoot	seriously
07:53	Jomyoot	hmmm
07:55	cark	what you could do is this : load a, load b, both loaded from c ... c tells a where to find b's functions it needs, c tells b where to find a's functions it needs
07:57	cark	for instance in a you would have (def my-proxy-to-b-hello) and from c you could (set! my-proxy-to-b-hello hello)
07:57	cark	(untested)
08:01	cark	or better yet, make A use B, then when A calls a function in B use a function parameter for the call B needs to do in A. you could use a map of functions if you want to pass several functions
08:22	maacl	Using Emacs/Slime how do I identify the Clojure source line/function that resulted in a specific runtime exception? The traceback only seems to provide the Clojure .java file that failed (in this case the LazySeq.java)
08:23	durka42	try clojure.stacktrace/print-cause-trace
08:23	dysinger	I noticed clojure master on github has 'backtrace' too
08:23	dysinger	(not 1.0)
08:26	maacl	That gives me java.lang.ClassNotFoundException: clojure.stacktrace (NO_SOURCE_FILE:0)
08:33	maacl	durka42: Should I do "clojure.stacktrace/print-cause-trace" at the Repl?
08:34	durka42	do (use 'clojure.stacktrace) first
08:34	clojurebot	clojure is like life: you make trade-offs
08:34	durka42	unless you don't have the latest clojure, in which case it might still be in clojure.contrib.stacktrace
08:36	maacl	durka42: ah, was in contrib
08:38	maacl	durka42: clojure.contrib.stacktrace/print-cause-trace -> #<stacktrace$print_cause_trace__4400 clojure.contrib.stacktrace$print_cause_trace__4400@5be2753e>
08:38	Chouser	whoa. I hadn't noticed that stacktrace came in with gtic
08:38	durka42	apparently
08:39	Chouser	that would explain that repl startup error I keep seeing but hadn't investigated yet...
08:39	durka42	:p
08:40	maacl	durka42: How is it supposed to be used ?
08:41	durka42	well, you cause an exception and then you can do (print-cause-trace *e)
08:50	maacl	durka42: does this have to be from the repl or can I do it from within a source file?
08:52	maacl	durka42: from within the source file it gives a nullpointer exception
08:57	durka42	you could catch the exception and run print-cause-trace on it
09:11	Jomyoot	i know there isn't for loop
09:11	Jomyoot	but how would i efficiently emulate for loop. where i loop over 0..n
09:12	Jomyoot	omg there is (loop)
09:12	Chouser	the key question is why are you looping: producing side-effects? producing and returning a lazy seq? producing and returning some other kind of collection?
09:13	Chouser	,(for [i (range 10)] (+ i i))
09:13	clojurebot	(0 2 4 6 8 10 12 14 16 18)
09:13	Jomyoot	producing side effect
09:13	Jomyoot	inserting into database
09:13	ChrisPS	,(repeat 7)
09:13	clojurebot	Eval-in-box threw an exception:java.lang.OutOfMemoryError: Java heap space
09:13	Jomyoot	where I need to change index value from 0 to n
09:13	ChrisPS	ok
09:13	Chouser	ok, so you probably want doseq
09:14	Chouser	(doseq [i (range 10)] (insert-into-db i))
09:14	Chouser	that'll return nil after callking insert-into-db 10 times
09:15	Chouser	actually, dotimes is even better
09:15	Chouser	(dotimes [i 3] (prn i))
09:15	Chouser	,(dotimes [i 3] (prn i))
09:15	clojurebot	0 1 2
09:17	Jomyoot	,(range 20)
09:17	clojurebot	(0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19)
09:18	Jomyoot	is there sin for using (for)?
09:19	Chouser	Jomyoot: nope, 'for' is good. But it's lazy, which means if you're producing side-effects you can't be sure when (or if) they will ever actually happen.
09:20	Jomyoot	i have to read into laziness
09:24	Chouser	,(do (for [x [1 2 3]] (prn x)) "all done?")
09:24	clojurebot	"all done?"
09:25	Chouser	note the numbers never got printed. So it's best not to use 'for' if you want side effects. Instead use doseq (or dotimes, or loop/recur, etc.)
09:25	Chouser	,(do (doseq [x [1 2 3]] (prn x)) "all done?")
09:25	clojurebot	"all done?"
09:25	clojurebot	1 2 3
09:25	jdz	,(do (prn "foo") "done")
09:25	clojurebot	"done"
09:25	clojurebot	"foo"
09:25	Chouser	heh
09:27	jdz	clojurebot should prefix the lines depending on what kind of output it is (standard output/error, return value)
09:27	jdz	i'm not volunteering ;0
09:28	jdz	although i imagine the change would not be complicated
09:59	artagnon	Different CL implementations on different platforms have different strengths and weaknesses. Why tie Clojure down to an implementation?
10:00	Chousuke	What do you mean?
10:00	jdz	artagnon: there's clojure-clr, did you know?
10:00	clojurebot	"[Clojure ...] feels like a general-purpose language beamed back from the near future."
10:00	artagnon	There are tons of Scheme implmentations too. Gambit compiles directly to C and can be really really quick. PLT is the most widely used flavor of Scheme- it's excellent for beginners.
10:00	jdz	artagnon: anyway, what's your point?
10:01	jdz	artagnon: considering that Clojure is mainly a one-man project, and very young?
10:01	Chousuke	artagnon: The "official" clojure is "tied" to the JVM platform because that gives a multitude of benefits.
10:02	artagnon	jdz: I don't have a point. I'm asking the reason for the design decision
10:02	Chousuke	artagnon: nothing is going to stop you from implementing a flavour of Clojure for other host platforms though, but they will likely be incompatible, at least if host interop is used.
10:02	Chousuke	~rationale
10:02	clojurebot	rationale is http://clojure.org/rationale
10:02	Chouser	Early Clojure supported both JVM and CLR
10:03	Chousuke	I suppose perhaps in the future there might be a Clojure specification, if someone bothers to write one.
10:03	Chousuke	but I think being implementation-defined allows Clojure to progress faster, which is important.
10:04	Chouser	I think the two main factors that caused the CLR support to be dropped were (a) the design of Clojure is not complete, and (b) one mature implementation is better than 2 immature implementations
10:04	artagnon	hm. So there's no specification draft? Just a couple of implementations to date.
10:04	artagnon	Chouser: I see.
10:04	Chousuke	artagnon: yes. The JVM Clojure implementation defines Clojure.
10:05	Chouser	there is one implementation, and it's structure and design is still evolving, though several of the key core concepts are quite stable and unlikely to change.
10:05	Chousuke	There are also plans to rewrite the Clojure compiler in Clojure, which ought to make Clojure more "portable"
10:05	Chouser	then there are a couple incomplete ports trying to follow along
10:06	cark	at the end of the day, the one implementation does the job, and we finally get to use a lisp in real life
10:06	artagnon	What's the current Clojure currently written in? Java?
10:06	Chouser	artagnon: yes, mostly Java plus some Clojure
10:06	Chouser	artagnon: http://blog.n01se.net/?p=41
10:07	Chousuke	In addition to that, the core libraries use some java interop
10:07	cemerick	I've never quite understood the desire to have a specification -- seems like that's a great way to cast a language in amber, but exactly what you don't want to foster progress/innovation shrug
10:08	artagnon	Clojure-in-Clojure is fine, but you need a Lisp environment first to eval those sexps.
10:08	Chouser	cemerick: I agree.
10:08	Chouser	artagnon: yep! And we have one that's nearly good enough
10:08	artagnon	Chouser: we do?
10:08	cark	artagnon : you only need those primitive forms
10:09	artagnon	cark: ofcourse.
10:09	artagnon	to be able to understand the Clojure that defines the rest of Clojure.
10:09	cark	as opposed to all the data structures and "standard library"
10:09	artagnon	Like the inbuilts in Elisp (written in C)
10:09	Chouser	In a closed-source world, a language spec is fine -- necessary, I suppose. But I'm entirely content to have a single open-source implementation instead.
10:10	Chouser	artagnon: Clojure as it stands is nearly good enough to write Clojure-in-clojure well.
10:10	Raynes	What would be the best way to make a sequence of the result of 5 read-line calls without typing (read-line) 5 times?
10:10	jdz	instead of writing language specification i'd rather like to see conformance tests being written
10:10	artagnon	Raynes: write a macro that unfolds the read-line 5 times?
10:10	Chouser	Raynes: (take 5 (repeatedly #(read-line x)))
10:11	Chousuke	beware: lazy seq :)
10:11	Chouser	Raynes: (vec (take 5 (repeatedly #(read-line x))))
10:11	Chouser	:-)
10:11	Raynes	Thanks. :D
10:12	artagnon	Do equivalent Java programs perform worse? ie. I'm asking if we have an efficient compiler (the Clojure bytecode compiler ie.)
10:13	Chouser	the byte-code produced by Clojure is efficient
10:13	cark	well to be honest clojure is slower
10:14	Chousuke	Clojure is typically slower than java, if you pay no attention your tight loops
10:14	artagnon	I see.
10:14	cark	boxing of primitive values, and function call semantics
10:14	Chouser	but that doesn't have to do with the bytecode produced by the compiler
10:14	cark	but inside a function you can get as fast as java
10:14	Chousuke	I mean, (reduce + (range 100000)) will not beat the equivalent java loop, but it's a lot easier to write :P
10:14	artagnon	Chouser: Then what does it have to do with?
10:15	artagnon	Chousuke: Why? What's the reason behind this?
10:15	jdz	Chousuke: and better yet, (reduce + (range 1000000000000))
10:15	Chousuke	artagnon: well, + is a function.
10:15	cark	artagnon : and range produces a lazy sequence
10:15	cark	that's overhead for convenience tradeoff
10:16	artagnon	Ah, I see.
10:16	cark	still you have the choice to make it as fast as java, by doing the same ugly thing as java
10:16	rhickey	Clojure calls to methods are precisely the same bytecode as Java's, so there's no general truth to 'Clojure is slower than Java'. Clojure's primitive numerics are also as fast as Java's. But, Clojure uses (Java's) boxed numbers by default and (Java's) boxed numbers are slower than (Java's) primitives.
10:17	Chouser	the biggest common speed hit is when the code provides insufficient type hints such that reflection is done at runtime. This is very convenient for the programmer, but huge slowdown
10:17	scgilardi	along the lines of readline before, this works for me: (take 5 (repeatedly read-line)) (I don't see read-line taking an arg)
10:17	cark	rhickey : well calling a clojure function from clojure is slower than java, and primitives are always boxed on function boundaries
10:17	rhickey	cark: actually fewer ugly things, as CLojure infers much more given fewer hints
10:17	cark	agreed
10:18	cark	hey i'm sold on clojure ! just trying to be honest =)
10:18	artagnon	Another thing I'd like to adress here: Many people hate Java. One of the reasons is that it's this whole shebang dump of this isolated box: Java has its own database (some JDBC or some such rubbish), graphical toolkit (Swig, right?) and other things.
10:19	Chousuke	artagnon: JDBC is just an API for accessing databases, not a database in itself. And the graphical toolkit is Swing. But there are also Qt bindings for java if you'd like to use that instead :)
10:19	artagnon	Chouser: Right, I suppose there are some optimization features that trade off convinience for speed to use in specific bottlenecks in Clojure.
10:19	rhickey	well, it's the generalities that are the problem. We can talk about anything specific, but people read those general statements and learn nothing about the truth of how to make things fast and what are the bottleneck. For instance, many have concluded that calling is slowing when it was primitive arg boxing holding things up
10:20	rhickey	So tight-loop benchmarks that pass primitives as args suffer, but many others don't at all
10:20	artagnon	rhickey: so the slowdown has to do with the language's design (relying on boxed types for example) and not the design of the bytecode compiler itself.
10:21	artagnon	Chousuke: Frankly, I was unaware. Sorry, I never really bothered to learn any Java.
10:21	rhickey	artagnon: the JVM bytecode is so simple, and most optimizations are actually performed by HotSpot, so Clojure gets them
10:21	rhickey	free
10:21	artagnon	ah, I see.
10:22	rhickey	It is a JVM limitation that one can't write fixnums as would be common in a dynamic lang
10:22	Chousuke	artagnon: well, I haven't bothered much with Java either. But Clojure is fun, and it helps to know the java libraries if you work with Clojure :)
10:22	rhickey	not a language design issue - if there were unboxed fixnums in the JVM Clojure would use them
10:23	artagnon	Chousuke: Right :)
10:23	rhickey	The JVM won't let a single type hold either a reference or a non-heap entity
10:23	Chouser	artagnon: the Java insular universe is actually related to my biggest annoyances with Clojure. But then again, the engineering work that has gone into that Java universe (JVM performance, library availability, etc.) is one of the things I like the most.
10:23	Chousuke	Is it possible to add fixnum support to the JVM without breaking backwards compatibility? :/
10:23	artagnon	rhickey: So it's more about the JVM not being tailor-made for Clojure.
10:24	rhickey	artagnon: fixnums could be used by any dynamic lang on the JVM
10:24	rhickey	http://blogs.sun.com/jrose/entry/fixnums_in_the_vm
10:24	cark	i think there's a push for better dynamic language support on the jvm ... we might see some progress
10:24	artagnon	Chousuke: Hey, nobody has actually explained this to me: But why is Java so isolated from the rest of the world?
10:24	Chousuke	artagnon: fortunately, there seems to be a lot of work going into making the JVM better for dynamic languages lately. :)
10:25	artagnon	rhickey: anything else apart from the fixnums that would help?
10:25	Chousuke	artagnon: Hmm. I don't think it's any more isolated than other languages utilising virtual machines.
10:25	rhickey	artagnon: tail call optimization
10:25	cark	artagnon : not any more isolated than CL
10:25	artagnon	rhickey: It doesn't have tail call optimization!?
10:25	rhickey	artagnon: nope
10:26	artagnon	cark: I thought CL was isolated because it's difficult and generally unpopular, no other reason :p
10:26	Chousuke	I actually prefer recur over true tail calls. it throws an exception if I get it wrong :P
10:26	artagnon	hm, that's terrible. SBCL has nice tail call optimization.
10:26	artagnon	Chousuke: ofcourse, that's for testing.
10:27	artagnon	In production, I'd let SBCL optimize it.
10:27	Chousuke	artagnon: no, I mean the recur special form in clojure, which is a workaround for the lack of TCO :)
10:27	artagnon	er, sorry. I'm unaware of this form.
10:27	rhickey	artagnon: that's not the same at all - real TCO is a promise to eliminate, not an optional optimization. Only with the former can you design programs using calls as control flow without worrying about the stack
10:28	cark	lack of TCO kept me away from clojure for a long time, than i tried it anyways, and i'm very glad i did =)
10:28	rhickey	I'm talking about Scheme-style TCO
10:29	Chousuke	artagnon: basically, if you have a tail-recursive function you'd call like (foo x y), instead of having (foo x y) in the tail position in the definition, you just have (recur x y)
10:29	cark	we have recur and trampolines to make it up, so that's really a non issue
10:29	Chousuke	artagnon: it can't do mutual recursion though. For that, you need a trampoline :/
10:29	artagnon	Chousuke: oh I see!
10:29	artagnon	that's interesting though.
10:30	cark	artagnon : are you thinking about giving clojure a try ?
10:30	Chousuke	it's not as elegant as it would be if the JVM promised TCO, but it's not too bad.
10:30	Chouser	cark: and lazy seqs
10:30	artagnon	cark: ofcourse!
10:30	cark	chouser : ah yes indeed
10:31	artagnon	I'm just downloading it and setting up Emacs :)
10:31	Chousuke	artagnon: clojure-mode can do that for you, with clojure-install
10:31	cark	artagnon : great =) you'll find plenty of help in this channel to get you started (i abused it quite a bunch)
10:31	Chousuke	artagnon: also, http://java.ociweb.com/mark/clojure/
10:32	artagnon	Chousuke: clojure-install doesn't work anymore :(
10:32	Chousuke	oh? :(
10:32	artagnon	Chousuke: the git repo mentioned in the code is gone
10:32	artagnon	I'll tell you which line, wait
10:32	Chousuke	hmm
10:32	Chousuke	maybe you have some old version of clojure-mode?
10:33	artagnon	I got it from jochu's GitHub head.
10:33	artagnon	and besides, I don't like auto-things anyway. Especially not auto-installing things. They freak me out a bit :p
10:33	Chousuke	hmm, yeah, that seems to be a bit outdated
10:34	Chousuke	http://github.com/technomancy/clojure-mode/tree/master this fork of it is a bit better maintained :)
10:34	artagnon	I want to be able to switch swank backends. I'm not going to stop writing CL or anything :p
10:34	Chousuke	heh
10:34	artagnon	Chousuke: ah, thanks! :)
10:38	Chousuke	rhickey: how's the chunks branch coming along, btw?
10:39	rhickey	Chousuke: I'd like to see some contribs for chunked for/take/drop etc
10:39	cemerick	heh, and I was going to ask about new-new. :-P
10:41	Chouser	c-in-c could be written today: reader and compiler are blocking on nothing and the data structures could use proxy.
10:41	rhickey	cemerick: newnew touches deeply on what I want Clojure to be. I'd be pretty happy to have it not have most of genclass, and have people always have to define interfaces
10:41	rhickey	I really don't want to embed Java in Clojure
10:41	Chouser	this would create a slower-clojure-in-clojure that may be a more pleasant language in which to write new-new?
10:43	rhickey	Chouser: not sure about that question - it would be as fast as Java, but not support some things, e.g. mutable fields
10:43	cemerick	rhickey: that's 95% of what I want, and the other 5% I can proxy.
10:44	Chouser	cemerick: when would you use proxy?
10:44	rhickey	cemerick: proxy with proxy?
10:44	rhickey	or did you mean stub somehow?
10:44	cemerick	There are times when one needs to produce a subclass, etc.
10:45	cemerick	rhickey: Maybe new-new would support that as well, but there's no telling, esp. the way you talk about interfaces.
10:45	rhickey	cemerick: so, subclasses are a key consideration right now. In fresh designs, I'm not in favor, but 2 things mitigate: many Java libs require concrete derivation, and: Clojure's impl makes use of abstract classes
10:46	cemerick	I remember your first shot at proxy (maybe a year ago, named differently) only worked with interfaces.
10:46	rhickey	It's about defining names classes more than deriving
10:47	cemerick	I think not being able to subclass would be a significant blow to interop and adoption. proxy would have to stick around in some form.
10:47	rhickey	so if you can derive from and abstract class, but not create one...
10:48	rhickey	I'm really concentrating on the cinc story - how to move the data structures over, what to do with the abstract bases - define a better way to do reuse-via-composition?
10:49	rhickey	or, a genclass variant with the same direct-call code gen as newnew?
10:50	artagnon	Is there a plan to port the Clojure compiler to say, LLVM? Would it serve any purpose?
10:50	cemerick	well, it seems that avoiding the indirection is critical for performance, so I'd assume anything genclass-esque is out
10:50	rhickey	newnew really needs fn + proxy, but half that code is in Java and the other half is in Clojure
10:51	rhickey	cemerick: there's no reason there couldn't be a genclass variant that had the method bodies inline
10:51	cemerick	well, in that case :-)
10:52	rhickey	artagnon: where are the libraries for LLVM?
10:53	artagnon	Oh right. Not only would you have to re-invent things like GC, but all the Java libraries that you took for granted too ~!
10:54	Chouser	rhickey: so it would be actually beneficial to have today's compiler feature set implemented in Clojure? Or am I reading too much into that.
10:55	cemerick	rhickey: couldn't an abstract base instance just be another field in the subclass object? supermethods would be directly callable...
10:55	artagnon	It took some time for me to grasp the fact that the JVM is so generic that it should be called HLVM instead.
10:56	Chouser	artagnon: or maybe OOVM
10:56	rhickey	Chouser: I think you may be reading into that, I'm still thinking about what it needs to do, not how to do it, other than it looking like a lot of work, either way it is done
10:56	artagnon	Chouser: it's object-oriented?
10:56	artagnon	what does OO have to do with it?
10:58	Chouser	artagnon: yes, as far as I know object, methods, inheritence, etc. are baked pretty deeply into both the JVM bytecode spec and the implementation(s).
10:58	rhickey	cemerick: It's about this - what's the recipe for writing Clojure in Clojure, especially the data structures? Does one rewrite the interfaces using gen-interface and write the impls using newnew? That leaves a hole around the abstract classes
10:58	artagnon	I see. I haven't looked into that at all actually.
10:59	rhickey	there are also a very few cases where even one other object creation impacts perf in a way that matters
11:00	rhickey	any composition strategy has issues there
11:00	artagnon	Another (probably very stupid) question: What kind of an environment does say, SBCL provide? It's very Lisp-specific and doesn't have the unnecessary features of the JVM. Does it lack something?
11:02	rhickey	Chouser: I have looked at the compiler stuff you've started - I guess I'm loathe to do a rewrite and not improve things. As I've said before, I wouldn't use dynamic binding similarly again, would pass environments explicitly, also the reflection is woven throughout and the use of class instances might be better with names
11:04	Chouser	yeah, the code I have tries to improve things it shouldn't ("instance" instead of fn) and fails to improve things it should (set! side effects all over the place)
11:04	rhickey	artagnon: any CL has only a tiny fraction of the libraries of Java, and also a tiny fraction of the users of those libraries, thus less testing etc
11:05	cemerick	rhickey: I guess I'd say that one should be able to produce abstract classes with newnew as well (though calling them that is a little odd -- they'd just be usages of newnew that don't have impls for all methods). The subclasses can refer to them by var-name, which (at least in my head) would allow you to pick up what the actual generated base classname is at runtime, and declare that base class as the superclass.
11:05	rhickey	Chouser: I understand the strategy of 'copy the Java into Clojure' in order to move forward in CLojure, might mean writing the compiler twice
11:05	cemerick	Doing anything else would seem to require developing a parallel object system or some other approach.
11:05	scgilardi	artagnon: and amazing toos for profiling, visibility into the workings of the VM
11:06	scgilardi	tools
11:06	rhickey	cemerick: that kind of dynamic use of class names is a definite no go. Java class names are static and I'm not going to pretend anymore, it's a disaster for OSGi and other modularity systems
11:07	Chouser	rhickey: that was the only strategy I could follow without understanding the inner workings of the current compiler better. except for the poorly-chosen attept to implement "instance".
11:07	artagnon	rhickey: Right. Got it.
11:07	danlarkin	artagnon: SBCL doesn't use a VM (as far as I know, at least). It does native code generation
11:08	cemerick	rhickey: Sounds fine by me. In that case, (new my.class.name [bases] [s-args] ...)?
11:08	rhickey	cemerick: in line with that, I think we need a distinct static phase of compilation, for interface defs at least
11:08	rhickey	cemerick: my.class.name is defined where?
11:08	artagnon	danlarkin: at all? Are you sure? I never thought stuff like hotpatching would be possible without a VM.
11:09	cemerick	rhickey: that's a symbol that controls the name of the class generated by the compiler when it loads the newnew form
11:09	rhickey	cemerick: for use by what other code?
11:10	rhickey	that's the fundamental problem, you can't dynamically define named JAva classes without a ton of problems, with classloaders, redefinition etc
11:10	cemerick	in the context of cinc, other newnew usages that define concrete impls (new c.l.SomePersistentDS [my.class.name other-interface] ...)
11:11	Chouser	counld abstract base classes be replaced by a sort of mix-in idea? new-new could generate classes that provide implementation from some other place when not specified in the form.
11:11	cemerick	that's mostly what I'm getting at, without having the benefit of intimate knowledge of the bytecode involved :-)
11:11	Chouser	would be a pain for Java code trying to re-use the abstract "class", but who wants to write java?
11:12	rhickey	Chouser: yes, exactly, that's the kind of thing I'm thinking about
11:12	artagnon	Thanks for everything, all. I'll give Clojure a spin.
11:13	rhickey	Chouser, cemerick: if not composition, then we're talking about some sort of mixin/macro thing
11:13	clojurebot	Who??
11:14	rhickey	might be as simple as - use these defs if no others are provided
11:14	cemerick	that makes super somewhat more difficult, no?
11:14	rhickey	I could probably have done most of the data structures that way
11:15	cemerick	rhickey: this is where my lack of understanding of the details kills me. If there's only one "mixin", isn't that just an abstract class?
11:15	rhickey	cemerick: super wasn't that important for me, it's almost always a bad sign
11:16	cemerick	rhickey: critical (super-critical!) for interop, though. If you're going to care about things like osgi, then you've got to have an abstract class extension story.
11:17	rhickey	there are only 5 super uses (that aren;t ctor calls) in clojure.lang, 3 are forced on my by Java bases classes
11:19	rhickey	cemerick: we're are talking about replacing the abstract bases in clojure.lang, that's a separate question from being able to derive from abstract bases. Even if the latter is possible, doesn't mean we want to have to define abstract bases in clojure
11:19	rhickey	consuming vs producing
11:20	rhickey	in all likelihood newnew will retain proxy's ability to derive from concrete classes, and improve the super access story
11:20	rhickey	but that's not the same as providing tools for defining abstract classes (other than existing genclass)
11:21	rhickey	i.e. if newnew is the only way to define the highest-performing classes, how do you replace abstract bases in your Clojure designs?
11:21	rhickey	given newnew defines an anonymous class
11:22	cemerick	rhickey: well, see, I underestimated your ambitions again :-)
11:24	cemerick	rhickey: FWIW, I think traits are the best thing about Scala. Clojure could do worse than be influenced by them.
11:24	rhickey	a concrete code reuse mechanism other than inheritance is a promising idea, gets out of naming and dynamic redefinition troubles
11:24	Chouser	you could still do super with mixins if you want to
11:25	Raynes	Traits are nice.
11:25	cemerick	obviously a lot of "stuff" there related to the type system, but the general model and workflow is nice.
11:25	rhickey	the dynamism is the trick, these things can't become names Java classes
11:25	rhickey	named
11:25	cemerick	Chouser: agreed.
11:25	rhickey	cuper is bad design
11:26	rhickey	super
11:26	cemerick	rhickey: neither are scala traits, actually. They produce an interface of the same name as the trait, though (along with some generated name that scalac picks up).
11:26	Chouser	the set of fns to define the abstract base should be static things anyway, right? so a special form ... well, ok. We don't have to support it, but nothing about the broad design for mixins as described prevents them from working.
11:27	Chouser	hm, actually, I make kind of mixin macro for proxy recently.
11:27	Chouser	made
11:30	Chouser	I guess I mentioned it before -- allowed me to define a new anonymous class that inherited from APersistentMap, mixed in some calls to a delegate instance of PersistentHashMap, and then provide implementations for methods from another interface.
11:30	Chouser	new anonymous class via proxy
11:31	Chouser	so, um, it worked fine. :-)
11:32	rhickey	dynamic anonymous is fine, dynamic named (class or interface) is the problem
11:33	Chouser	right. so these mixins would be named and therefore not be a class or interface
11:33	rhickey	so the code we want to reuse has to be either a) staitc (e.g. genclass) or b)unnamed (e.g. not named by class/interface name)
11:36	rhickey	What's interesting about all of the abstract bases in Clojure is that there is/are corresponding interface(s) (as there should be)
11:41	rhickey	so if a newnew instance, acting as a prototype for an abstract base, sitting in some var, had an accessible (though anonymous) name (as suggested by cemerick earlier), then a 'mixing' newnew class could just declare it implements the same interfaces. Constructors are a problem though
11:42	rhickey	newnew hides ctors
11:43	Chouser	the mixing class's undefined methods would call methods of the mixed instance? wouldn't the mixing instance have to be passed in somehow?
11:43	cemerick	there's still one under the covers, though. The mixing class just provides super-args that align with those defined by the referenced newnew instance...?
11:44	cemerick	Chouser: implicit first 'this' arg?
11:45	Chouser	cemerick: but that would be a different method signature than defined by the interface its supposedly implementing!
11:47	cemerick	I really do mean implicit -- you'd use the 'this' symbol in the method impl, but the compiler would have to drop in the reference to the callee.
11:47	Chouser	I was assuming a mixed thing would have to be something like a class full of static methods whose names and signatures were similar (but not identical) to the instance methods defined by an interface.
11:48	cemerick	I don't think the resulting newnew concrete instance would be usable from a Java interface standpoint in that case.
11:49	cemerick	(not without a proxy object in front setting up the calls properly)
11:49	Chouser	cemerick: but drop that reference in where? if the method signature is defined by a public named interface, how do you pass in anything other than the "template" instance for the mixed class?
11:50	yason	Q: is there byte-seq that is analogous to line-seq ?
11:50	Chouser	yes, it would be a pain to try to use mixable class from java code
11:51	cemerick	my understanding is that the "template" instance wouldn't be used at all -- only its anonymous yet named class, with method impls, would be used.
11:51	cemerick	The template instance is just a handle that allows newnew to get a hold of the abstract base/mixin class.
11:51	cemerick	(at least how I'm reading rhickey's last msg)
11:52	Chousuke	urgh, for is a complex beast :P
11:52	cemerick	yason: how would such a thing segment the byte stream?
11:53	Chouser	ahh... so when given a mixable, newnew would ... copy the method implementations directly into the mixing class def?
11:53	rhickey	Chouser: the point I was trying to make about interfaces was that if the mixing-in class only fulfills the same contract as implied by the interfaces shared with the mixin, then the mixin can make blind calls to those methods, knowing they will call the most-derived versions
11:53	Chouser	oh!
11:54	rhickey	this is the non-macro-like solution, a macro-like one would eschew inheritance completely, just insert code when needed. There, as I said, I think calling super is bad and unnecessary
11:54	yason	cemerick: what do you mean?
11:55	cemerick	yason: line-seq breaks up text by line breaks -- how would you want to break up your InputStream?
11:55	yason	cemerick: I'm thinking of a lazy sequence of bytes read off from some reader
11:55	cemerick	or, InputStream's bytes
11:56	Chouser	so a mixing newnew class would actually inherit from the anonymous class of the mixable newnew?
11:56	rhickey	Chouser: right
11:57	rhickey	the tricks are: making newnew generate an abstract class when needed, constructors
11:57	rhickey	a newnew class gets a ctor, (as does a fn) that takes as args all closed-over locals, but Clojure completely hides that
11:58	cemerick	rhickey: and my point about ctors is that, the mixable class has ctors (or factory fn, whatever), and the mixing class can just pass super-args as one would with proxy today
11:58	rhickey	were the newnew class be used as a base, there's have to be some access to that ctor
11:59	rhickey	cemerick: factory fn is not a ctor, can't be used to init a base
11:59	rhickey	exposing the implicit ctor gets icky quickly
11:59	cemerick	right, right
12:00	cemerick	how so?
12:01	rhickey	because some aspects of the mixin prototype would have to be 'free' names
12:01	Chouser	the fields for closed-over values could be mutated once during creation. just mark them final but change them once anyway.
12:02	rhickey	e.g. (def x (let [a 42 b 57] (newnew [Object] [] ...))
12:03	rhickey	a and b will be args to the class iff they were used in the impl, as would any other closed over things, might not be one level. No way to say, I want a to always be 42 and b to be a variable thing supplied by derivee
12:04	rhickey	Chouser: that abuse of final is in conflict with the optimizer - you might not see that new value in all threads
12:05	Chouser	rhickey: already ducked.
12:05	rhickey	just saying, it's not an option even
12:06	cemerick	Chouser: I already made rhickey blush talking about mutable fields a week or two ago ;-)
12:07	rhickey	mutable fields unlikely too, the only arg for them is the multiple allocation cost for something like AtomicReference or clojure.lang.Box
12:08	rhickey	but the lowest-level Java code of Clojure does use mutable fields, with all needed precautions synchronization-wise
12:09	Chouser	to specify mutable fields you'd have to give up some of the beautiful simplicity of closing over a regular local for all your state.
12:09	cemerick	rhickey: not so. I want to stop having to think about massive intermediate garbage produced while (for example) reducing a struct through transformation. Right now, in order to avoid that, I need to manually pull out the fields of interest and loop through them, and then create a struct at the end using the transformed values.
12:10	cemerick	that said, I'm certainly not expecting there to ever be a path provided for that sort of thing in clojure.
12:10	rhickey	cemerick: heh, I just designed one yesterday
12:10	cemerick	oh?
12:11	rhickey	batch updatable persistent data structures with O(1) conversion to/from mutability
12:11	rhickey	a.k.a. the holy grail
12:12	rhickey	not a joke, I've been working on the vector all day yesterday and during this discussion today
12:12	cemerick	vector?
12:12	cemerick	nah, I'm getting punk'd.
12:13	rhickey	yes, PersistentVector replacement
12:14	Chouser	only abstract newnews would need constructor args, right?
12:14	rhickey	you can say (let [v! (mutable v)] ...) and get a non-sharable vector supporting conj!, nth!, pop!, when you are done call (immutable v!), both mutable and immutable are constant-time
12:14	cemerick	I'm lured in. If that's true, then that can be newnew's fields AFAIC
12:15	rhickey	Chouser: all newnews that close over have ctors, it's a matter of exposure, usually the enclosing factory fn calls the ctor implicitly
12:15	cemerick	rhickey: assoc as well?
12:15	cemerick	assoc!, perhaps?
12:15	rhickey	cemerick: yes assoc!
12:16	rhickey	the general design will apply to all of the data structures except lists
12:16	dysinger	interesting
12:16	Chouser	rhickey: right, I mean only abstract newnews would need to be able to expose ctor args so that a derived class could pass them in explicitly.
12:16	rhickey	Chouser: right
12:16	Chousuke	rhickey: non-sharable meaning what? is it allowed to leave the scope it's created in? :)
12:17	rhickey	Chousuke: meaning, if you share copies of it you are in trouble and I can't help you
12:18	rhickey	but the updating is open in general, and after an immutable call I can help you, preventing any further mutation
12:18	Chouser	so an abstract newnew would list formal ctor args (that would actually be added to the full list of args for closed-over things) and a concrete newnew would list actual values to be passed in to the base class.
12:19	Chouser	hm, I guess an abstract base class may itself want to derive from another abstract and be able to pass in ctor values.
12:19	rhickey	Chouser: yes, it gets ugly
12:19	rhickey	I'm not sure bending newnew to do this is the right thing
12:20	rhickey	Chousuke: the prototypical 'right' use case is something like into, or other birthing situations, where no one will see the interim values
12:20	Chouser	Would you guess that abstract bases will only rarely want to close over locals?
12:21	rhickey	Chouser: I was going to say that before, but I'm not sure - it's the nice way to build in constants
12:22	Chouser	because of course as soon as you're closing over things, you want something very much like newnew
12:22	rhickey	Chouser: exactly
12:25	Chouser	I really like the kind of dynamism this would allow. redef the base newnew and you actually get a new anon class, so all existing derived instances continue to behave as they had. But new derived instances will see the new base class and pick up the new behavior.
12:26	rhickey	Chouser: yes, I'd love to keep the static-ness to interfaces, which are still critical
12:27	rhickey	Chouser: another sticking point is that newnews would normally only have one ctor, many abstract bases have multiple ctors
12:30	Chouser	ew
12:31	cemerick	is that at all necessary for cinc?
12:45	beutdeuce	i'm getting an unquote is unbound illegalstateexception. What does that mean?
12:46	Chouser	it means you're using ~ outside of a `
12:46	beutdeuce	how do i epxress inequality?
12:46	beutdeuce	express*
12:46	Chouser	,(not= 2 3)
12:46	clojurebot	true
12:46	beutdeuce	k, thnx
12:50	beutdeuce	any idea why i'm getting a stack overflow? => http://pastie.org/548306
12:53	cemerick	very strange here: I've got a slot with the value: (delay (apply merge (pmap expensive-map-producing-fn [somedata]))) When I use pmap, the resulting delay contains an empty map. When I use map, I get correct results.
12:54	cemerick	beutdeuce: you're calling length recursively
12:54	beutdeuce	cemerick: it should stop when there is no more xs
12:55	maacl	beutdeuce: you are checking against nil
12:55	maacl	(= nil "")
12:55	maacl	,(= nil "")
12:55	clojurebot	false
12:56	beutdeuce	what does the , mean?
12:56	maacl	beutdeuce: it asks clojurebot to evaluate
12:56	beutdeuce	ah, cool
12:58	beutdeuce	no, i need to check if xs is not equal to zero
12:58	beutdeuce	which is not=
12:58	beutdeuce	but for some reason, it never equals zero, which it should
13:02	technomancy	beutdeuce: the empty list is not nil in clojure
13:02	technomancy	,(nil? ())
13:02	clojurebot	false
13:02	technomancy	,(empty? ())
13:02	clojurebot	true
13:02	beutdeuce	ah
13:02	beutdeuce	empty?
13:02	technomancy	beutdeuce: that's what you want instead
13:03	beutdeuce	ajd to check if it isnt empty?
13:03	beutdeuce	and*
13:03	cemerick	,(seq '(1 2))
13:03	clojurebot	(1 2)
13:03	cemerick	,(boolean (seq '(1 2)))
13:03	clojurebot	true
13:03	cemerick	,(not (empty? '(1 2)))
13:03	clojurebot	true
13:04	beutdeuce	is there a ! in clojure to signify not?
13:05	technomancy	beutdeuce: ! has different connotations in clojure; it usually signifies side-effects
13:05	beutdeuce	i see
13:05	technomancy	not is just not
13:06	beutdeuce	k
13:10	rhickey	http://java.sun.com/javaone/2009/rockstars.jsp
13:12	technomancy	rhickey: nicely done.
13:12	rhickey	it's good for Clojure to get on the radar in such a mainstream venue
13:13	hiredman	rockstars? you'll need to beadazzle your jeans
13:13	technomancy	\m/
13:16	technomancy	so what's the deal with this guy who's doing his own (non-clojureCLR) C# port of Clojure? http://code.google.com/p/ajlisp/updates/list
13:23	beutdeuce	lol, now i get a null pointer
13:26	Chouser	thats a whole lot of rockstars.
13:30	Chouser	does the JVM disallow operations in ctor before calls to super's ctor? Or is that a Java-level thing?
13:33	liebke	rhickey: I was just reading your JavaOne slides (linked on the rock stars page), do you have a copy of them on a public server that I can point people to (without registering with the SDN)?
13:34	rhickey	technomancy: I really wish he wouldn't call that Clojure
13:34	rhickey	I also pointed him to Clojure CLR, but he's persisted
13:34	Chouser	You asked him to stop using the name and he's refusing?
13:35	rhickey	No, I haven't asked him to stop yet, will do
13:35	technomancy	I hate that I'm saying this, but it might be wise to register the trademark if you haven't. =\
13:36	rhickey	technomancy: yes, I've been looking at what Python does with theirs, seems sensible
13:37	rhickey	there are some issues with trademarks and some distros, no?
13:38	technomancy	yes, but IIRC only trademarked artwork
13:38	rhickey	unfortunately just more work and expense for me
13:38	technomancy	it's a shame that it's necessary
13:38	technomancy	the free software community has a great legal framework for dealing with copyright, but nobody has really got anything satisfying for trademarks
13:39	Chouser	it may not be necessary. his project is more likely to get long-term traction under its own name than if its generally known as a bad-will knock-off.
13:44	technomancy	python's guidelines seem pretty sane
13:45	Lau_of_DK	rhickey: Alternatively if you want to navigate around a potential legal battle, you could rename Clojure to .. oh I dont know.. Clabango? :)
13:46	danlarkin	rimshot
13:47	cemerick	The Clojure Foundation is not far away, I guess.
13:47	Chouser	bleh
13:49	Chouser	budgets board meetings elections lawyers
13:49	beutdeuce_	technomancy: ruby's are even more :)
13:50	cemerick	Chouser: the ways of the world
13:50	clojurebot	Why are you asking him
13:50	Lau_of_DK	hehe
13:50	Lau_of_DK	nasty bot
13:53	beutdeuce_	how would one go about creating custom types/type classes in clojure?
13:55	Chouser	beutdeuce_: most of the time it's best to just provide a bunch of functions that operate on a map or vector of a particular shape
13:56	beutdeuce_	Chouser: i see, so clojure takes a different approach than haskell.
13:57	dnolen	beatdeuce_: if you need something complex/fancy look at multimethods, add-hoc hierarchies, and metadata.
13:58	Chouser	and if you need to provide an interface that can be used by regular Java code, look at gen-interface and proxy
14:02	cemerick	this map/pmap behaviour is really strange -- only manifests itself when the pmap call is made on the AWT event thread
14:20	hiredman	cemerick: please continue
14:21	cemerick	hiredman: not sure what else to say beyond what I've said already. I'm pretty baffled.
14:21	hiredman	must have missed the initial explanation
14:22	cemerick	very strange here: I've got a slot with the value: (delay (apply merge (pmap expensive-map-producing-fn [somedata]))) When I use pmap, the resulting delay contains an empty map. When I use map, I get correct results
14:22	hiredman	~def delay
14:22	hiredman	eep
14:22	hiredman	thats not right
14:23	hiredman	~def clojure.lang.Delay
14:24	hiredman	~def pmap
14:30	hiredman	hmmm
14:32	hiredman	have you tried just wrapping in a closure?
14:33	cemerick	hiredman: not yet, but I'll throw that dart at the wall next :-/
14:33	hiredman	:/
14:34	cemerick	Debugging facilities leave much to be desired. Thankfully, nearly all of our fns are small and functional enough that reasonable unit testing is straightforward and almost-uniformly successful.
14:42	hiredman	,(doc merge)
14:42	clojurebot	"([& maps]); Returns a map that consists of the rest of the maps conj-ed onto the first. If a key occurs in more than one map, the mapping from the latter (left-to-right) will be the mapping in the result."
14:43	ataggart	it would be neat to have a dummy stack be created during execution that holds the clojure line numebrs etc (like clojure.test's output). just for testing/.debugging of course.
14:44	cemerick	rhickey: did you say something before about how you'd like to have reified environments?
14:45	Chouser	the java stack holds clojure line numbers
14:46	Chousuke	I guess macros complicate the line number stuff? :/
14:47	Chouser	I would think so. might even be hard to define what you'd want
14:47	Chouser	I suppose CL has addressed this
14:48	cemerick	hiredman: wrapping the pmap call (and the entire body of the delay) in a thunk had no effect
14:48	cemerick	nor did eliminating the delay
14:48	hiredman	weird
14:48	hiredman	is your function creating non empty maps? :P
14:48	ataggart	what reason, other than caching, would one use delay/force vs quote/eval
14:49	cemerick	heh
14:49	cemerick	hiredman: note that the results are perfect over the same dataset if I change pmap -> map
14:49	Chousuke	ataggart: why would you use quote/eval if delay/force works? :/
14:49	hiredman	well
14:49	cemerick	you're saying I just think it's the same dataset? :-)
14:49	ataggart	premature optimization :( I should know better, but...
14:49	hiredman	cemerick: I dunno
14:50	hiredman	it seems like a good idea to check
14:50	hiredman	put some logging in expensive-map-producing-fn
14:50	cemerick	yeah, I did that first, actually :-)
14:50	hiredman	at it is producing the expected results
14:53	cemerick	as far as I can tell.
14:53	cemerick	I'm sure I'm the one screwing up, but it's convenient to think that might not be the case for a little while.
14:54	lisppaste8	ataggart pasted "logging functions (jul, log4j, commons)" at http://paste.lisp.org/display/83697
14:54	ataggart	if anyone thinks that's worth sticking contrib, be my guest
14:54	ataggart	my CA is in the mail
15:13	mebaran151	ataggart, that actually looks really nifty
15:13	ataggart	thx
15:14	ataggart	needs better documentation and examples tho
15:15	ataggart	I just didn't want to write an actual logging system, there's enough of those around
15:18	cemerick	ataggart: yeah, that's pretty decent. Why bother with the separate commons-logging bit though, since you have separate support for log4j and java.util.logging?
15:23	ataggart	because commons-logging has its own config stuff, and some companies (like mine) use it
15:23	ataggart	my approach was not to mess with whatever someone might have already been doing in their java code
15:24	cemerick	oh, I didn't realize c-l had much of any configuration available...it's been a long time since I've touched it
15:24	ataggart	also cl supports other logging systems
15:24	ataggart	e.g. slfj
15:24	ataggart	whic I didnt get around to adding
15:26	ataggart	I did really enjoy the fact tat it was way easier to check for the presence of a lib with clojure than what the cl folks had to do in java
15:27	cemerick	is slfj used much at all?
15:27	cemerick	(I've never seen it in the wild)
16:23	Chouser	I'd really like class name aliases
16:24	Chouser	I guess this case is bad enough a macro would help...
16:29	cemerick	hiredman: and the winner goes to.....pmap + binding!
16:30	hiredman	huh
16:30	Chouser	yep, that would do it
16:31	rhickey	cemerick: what's the contest?
16:31	Chouser	binding is really quite pernicious in a thready, lazy, closury environment
16:31	cemerick	rhickey: I had a pmap call that was failing in a UI context, but not in a headless context, and it worked in the former if I switched to map.
16:32	cemerick	turns out I had a dosync floating around 50 stack frames up, so the threads used in the pmap couldn't see a variety of things.
16:32	cemerick	(only in the UI, that is)
16:33	cemerick	sorry, s/dosync/binding
16:34	maacl	Can anybody give me some hints on how to translate this: http://bit.ly/lcs_py (the backTrackAll function) from Python into Clojure? I have a pasted my attempt here: http://paste.lisp.org/+1SL7, but it doesn't work. For completeness I have included the functions required to test the traceback-all-lcs function and my traceback-lcs function (that works :-)).
16:34	cemerick	although the same issue would pop up in a transaction where some lower level used pmap, or futures, etc.
16:35	rhickey	cemerick: when something is running in a thread pool as utilized by pmap, I don't control the thread
16:35	rhickey	and it could be taking on work sent from multiple contexts
16:35	Raynes	technomancy: Is this a known bug? http://tr.im/sGhH
16:35	rhickey	so you really need a fn that captures any bindings you care about
16:35	cemerick	rhickey: yeah, I understand the dynamic, I had simply forgotten about the binding stuff that was going on in the UI
16:36	rhickey	I think there should be some with-captured-bindings macro
16:36	rhickey	yields a fn
16:37	rhickey	but generally I think people have this weird expectation that closures should know if they are going to run in other threads and do something different
16:38	cemerick	well, it's not so much that as a caller doesn't necessarily know what a fn is going to do w.r.t. parallelizing computation, and a fn can't always be sure of the environment it's going to be called in
16:40	cemerick	Just thinking about the pmap usages that I have floating around, I'm starting to get uneasy about the notion of using refs with (potentially) large transaction scopes for my cells reimpl.
16:40	technomancy	Raynes: not known by me, but I'm far from an expert on Windows
16:40	rhickey	I don't disagree that that is in conflict with dynamic binding generally, but pmap is explicit
16:40	rhickey	and pcalls, and future, and send
16:40	Raynes	technomancy: Happens here on Ubuntu Hardy as well.
16:40	Raynes	Apparently Brian Carper gets the same thing.
16:41	Chouser	does dosync use thread-local data?
16:41	Chouser	I suppose it must...
16:41	cemerick	rhickey: yeah, I'm just jabbering into the wind here. The tools are fine, I just need to apply them a little more consciously.
16:41	cemerick	yeah, transactions are thread-local until they've committed
16:41	rhickey	cemerick: not at all, I've been thinking about what would be a good alternative to dynamic binding for concurrent use
16:41	technomancy	Raynes: oh, right. I think the problem is actually that in is hooked up to the inferior-lisp buffer instead of the slime-repl buffer
16:42	Raynes	:o
16:43	Raynes	How would I go about changing that >_>
16:43	rhickey	there is a long-standing argument about inheritance of environments by sub-threads, but that is born of an old-styile use of manually launched threads. Thread pools ruin any notion of nesting or child thread
16:43	cemerick	rhickey: what would the options be in general, short of passing around environments between thread boundaries?
16:44	cemerick	s/between/across
16:44	technomancy	Raynes: buffer interaction with slime doesn't fit well into the streams paradigm. I've only used read-line etc in conjunction with sockets from slime
16:44	rhickey	cemerick: closures would have to capture definition context of some sort
16:46	rhickey	not normally what you want wrt dynamic vars, but perhaps explicitly
16:47	rhickey	if an explicit construct could do that, and re-establish, it would cover all calls made by it. You would only need to do that in the fn you pass to pmap/agents etc
16:48	rhickey	(pmap #capture-this-env(foo %) xx)
16:49	rhickey	people really expect that of the anonymous fns they pass to pmap, but not in their defns
16:49	rhickey	even though they are the same
16:50	hiredman	basically a short cut for (binding [a 1] (let [a a] (fn [] (+ a 1))))
16:50	hiredman	?
16:50	rhickey	hiredman: not, it has got to do a real rebinding, so nested calls see the dynamic vars
16:51	hiredman	oh
16:51	rhickey	so (let [a a] #(binding [a a] ...
16:51	hiredman	yeah
16:51	rhickey	definitely macro-able
16:52	rhickey	the only trick is you want args usually
16:53	technomancy	Raynes: it looks like swank-clojure doesn't rebind in at all
16:53	rhickey	(pmap (fn-with-bindings [a b c] [x] ...) xs)
16:53	rhickey	fn-capturing
16:54	rhickey	(fn-capturing [vars] [args] body)
16:54	rhickey	or something
16:54	kotarak	(defmacro bound-fn [vars args & body] (let [gens (take (count vars) (repeadetly gensym))] `(let ~(vec (interleave gens vars)) (fn ~args (binding ~(vec (interleave vars gens)) ~@body)))))
16:54	rhickey	right
16:55	cemerick	hrm
16:55	technomancy	Raynes: doesn't quite do what you want, but it might be a start: (binding [in (java.io.BufferedReader. (:reader swank.core.connection/current-connection))] (read-line))
16:55	rhickey	cemerick: I understand you may not know what vars are used by nested calls
16:55	cemerick	yeah, that's one big issue
16:56	rhickey	that is a separate problem, I might be able to make capturing all bindings cheap
16:56	rhickey	the bindings system is designed to make that possible, or at least was
16:56	rhickey	could still again, an implementation detail
16:56	cemerick	actually, that's the biggest issue, esp w.r.t. transactions. To do what I was thinking of doing, I'd need transactions to span multiple threads -- which, now that I've said it, doesn't seem like a good plan in general.
16:57	kotarak	maybe fn-with-bindings could go into c.c.core until something suitable is ready?
16:58	Chousuke	kotarak: you don't need the (take (count vars) ..) part actually.
16:58	Chousuke	interleave stops when it runs out of items in one of the seqs :)
16:58	kotarak	ah. Nice. Like map.
16:59	rhickey	cemerick: if those threads touch refs then you need help, if they just do computations then that's ok
17:00	cemerick	rhickey: so if they're doing computations on refs, they cancel each other out or something?
17:00	cemerick	;-)
17:00	cemerick	no, my current/former plan is just wrongheaded.
17:03	cemerick	futures and pmap aren't really the "problem" -- it's not strictly necessary that I parallelize at that level of granularity. The real issue is using the cells impl in conjunction with a UI (where one will inevitably want to push some computation out to another thread).
17:10	rhickey	alrighty - http://github.com/richhickey/clojure/commits/master
17:11	rhickey	if someone would please implement push/pop/get-thread-bindings, this new (clojure.lang.Var/getThreadBindings) will get you a map ready to push
17:12	rhickey	this will let you build the (preferable) capture-all-bindings
17:13	rhickey	user=> (clojure.lang.Var/getThreadBindings)
17:13	rhickey	{#'clojure.core/command-line-args nil, #'clojure.core/print-meta false, #<Var: --unnamed--> 2, #'clojure.core/ns #<Namespace user>, #'clojure.core/math-context nil, #<Var: --unnamed--> #<DynamicClassLoader clojure.lang.DynamicClassLoader@7e28388b>, #'clojure.core/2 nil, #'clojure.core/warn-on-reflection* false, #'clojure.core/3 nil, #'clojure.core/1 {#'clojure.core/command-line-args nil, #'clojure.core/print-meta false, #<
17:13	cemerick	hmm, fancy
17:14	cemerick	the nice th
17:15	kotarak	(defmacro fn-with-bindings [args & body] `(let [bindings# (clojure.lang.Var/getThreadBindings)] (fn ~args (try (clojure.lang.Var/pushThreadBindings bindings#) ~@body (finally (clojure.lang.Var/popThreadBindings))))))
17:15	kotarak	and
17:16	rhickey	binding* ?
17:16	rhickey	a finally without it's own try?
17:16	kotarak	A function doing what binding does accepting a map.
17:17	rhickey	tools are there to build that
17:17	kotarak	binding could then look like `(binding* ~(construct map here) (fn [] ~@body))
17:17	rhickey	I'd prefer not to have Var/blah - these can become proper supported fns push/pop/get-thread-bindings
17:17	kotarak	rhickey: I know. I did several times...
17:21	kotarak	rhickey: this is what I had in mind: http://paste.pocoo.org/show/129042
17:22	kotarak	But it throws.... which means it's possible to inject something.
17:22	rhickey	kotarak: I don't get it, what's the point of my-try?
17:23	rhickey	no user should be writing finally outside of try
17:23	rhickey	I think it is a non problem
17:24	kotarak	I don't know whether it is problem or not (most likely not). But if I expand a macro into a try wrapping some body I could inject accidentally some catch clause.
17:24	kotarak	What users should do and what they actually do, are two different things... But I agree that this is not really a problem.
17:25	rhickey	kotarak: I don't see how unless people are writing loose finally/catch clauses, else they would nest
17:26	rhickey	anyway, with getThreadBindings you can now have some fun
17:26	kotarak	(my-macro-expand-in-a-try (try (my-user-try)) (catch SomeException ...)) One wrong paren later.... but anyway. It is not a problem. Otherwise it would have already been mentioned...
17:34	Chouser	rhickey: has your interest in a finger-tree diminished with the vector stuff you were describing today?
17:34	clojurebot	rhickey is a T-1000 sent from the future
17:58	rhickey	Chouser: not at all, finger-trees will support fast insertion in the middle and concatenation, definitely different properties from vector
17:58	Chouser	ok
17:59	Chouser	just checking. :-)
18:02	Chouser	I guess I knew the properties were very different from a vector, but I wondered if you wanted a mutable/immutable style instead of an always-immutable persistent finger tree.
18:03	Chouser	ah, I misunderstood the api!
18:36	rhickey	Chouser: the immutable/mutable design is always based upon a pure immutable one
18:44	wavister	does anyone else have trouble building a fresh pull of clojure-contrib?
18:45	wavister	it says it can't locate walk.clj, which is part of the whole repo i just downloaded. seems like those paths should be set up correctly by default...
18:46	Chousuke	hm
18:47	Chousuke	didn't walk move to Clojure itself when c.c.test-is did?
18:48	Chousuke	that might be the cause of your breakage. walk is no longer in contrib.
18:48	Chousuke	if you're using contrib with 1.0.0, checkout the clojure-1.0-compatible branch
18:49	wavister	ahhh that must be the thing
18:49	wavister	or maybe i should get a newer clojure...?
18:51	hiredman	yeah
18:55	Chousuke	wavister: well, if you don't mind running clojure from git :P
18:56	wavister	i don't mind.
19:00	wavister	aaand it works! thanks!
19:55	rhickey	hmm, PersistentVector in batch mode slightly faster than j.u.ArrayList!
19:56	rhickey	sorry, Vector
19:57	rhickey	only slightly slow than ArrayList
19:59	rhickey	slower
20:01	schaefer_	i'm curious why the following doesn't work: (def (with-meta (symbol (str 't)) {:t "t"}) "t")
20:02	schaefer_	i get that def is a special form... is there a way to achieve the binding with a dynmically created symbol?
20:02	rhickey	schaefer_: def is a special form that doesn't evaluate its first argument
20:03	technomancy	maybe this: (eval `(def ~(with-meta (symbol (str 't)) {:t "t"}) "t")) ?
20:03	rhickey	,(doc intern)
20:03	clojurebot	DENIED
20:03	rhickey	,(doc intern)
20:03	clojurebot	DENIED
20:03	schaefer_	i'm actually going through clojure.core and trying to understand why the defn macro works the way it does
20:04	technomancy	schaefer_: that works, but I don't think it's what you want wrt metadata
20:04	rhickey	why, clojurebot?
20:04	technomancy	(doc intern)
20:04	clojurebot	DENIED
20:04	technomancy	hiiiiiiiredman!
20:04	rhickey	,(doc rest)
20:04	clojurebot	"([coll]); Returns a possibly empty seq of the items after the first. Calls seq on its argument."
20:04	rhickey	hrm
20:04	Raynes	(doc intern)
20:04	clojurebot	DENIED
20:05	schaefer_	well, what i ultimately want is a way to define a function with some meta data attached to it.
20:05	technomancy	schaefer_: metadata on the function or on the var?
20:06	schaefer_	good question. i think i want it on the var. i don't think you can attach metadata to a function (IFn does not subclass IObj)
20:06	technomancy	it's very tricky to do, in any case
20:07	schaefer_	how about this: how would you go about attaching additional metadata to an existing var?
20:08	rhickey	it is definitely runtime metadata?
20:08	schaefer_	rhickey: yes, but i'm curious as opposed to what?
20:10	rhickey	user=> (def #^{:my :meta} foo 42)
20:10	rhickey	#'user/foo
20:10	rhickey	user=> ^#'foo
20:10	rhickey	{:ns #<Namespace user>, :name foo, :file "NO_SOURCE_PATH", :line 16, :my :meta}
20:11	schaefer_	can you translate that without the reader macros? i'm just learning and slowly coming up to speed
20:11	schaefer_	i'm really trying to get the basics of all this stuff down
20:13	rhickey	schaefer_: well, the reader metadata is key, and thus the distinguishing point, getting at the meta can go like this:
20:13	rhickey	user=> (def #^{:when (java.util.Date.)} foo 42)
20:13	rhickey	#'user/foo
20:13	rhickey	user=> (meta (var foo))
20:13	rhickey	{:ns #<Namespace user>, :name foo, :file "NO_SOURCE_PATH", :line 24, :when #<Date Thu Jul 16 20:12:55 EDT 2009>}
20:14	technomancy	I'm not sure you can do the def without reader macros
20:15	rhickey	#^amap a-symbol-or-collection will read one object with attached metadata
20:15	schaefer_	ah... that could have been my failure when i was doing this on my own. i was trying to work it without the reader macro
21:06	arohner	,(name (var map))
21:06	clojurebot	java.lang.ClassCastException: clojure.lang.Var cannot be cast to clojure.lang.Named
21:07	Chouser	,(.sym (var map))
21:07	clojurebot	map
21:07	Chouser	,(.ns (var map))
21:07	clojurebot	#<Namespace clojure.core>
21:07	arohner	yeah, I knew the methods were on the var object
21:08	arohner	just found it weird that name didn't work on it
21:08	Chouser	yeah. me too.
21:08	Chouser	of course not all vars have a name
21:08	arohner	oh, right
21:08	rhickey	ArrayList
21:08	rhickey	time: 1129
21:08	rhickey	PersistentVector (batch mode)
21:08	rhickey	time: 1354
21:08	rhickey	smokin!
21:08	arohner	with-local-vars, etc
21:10	rhickey	was:
21:10	rhickey	PersistentVector
21:10	rhickey	time: 3547
21:11	Chouser	wow
21:12	Chouser	much too fast
21:12	rhickey	truly crazy since you can, anywhere in the middle of the batch, say (immutable v) and get a perfectly fine immutable vector
21:12	rhickey	and then keep going
21:12	Chouser	does the immutable thing persist like a regular PVector?
21:13	rhickey	(immutable v) being instantaneous
21:13	rhickey	Chouser: the immutable thing _is_ a pvector
21:13	Chouser	oh, same internal structure?
21:14	rhickey	the mutable thing is almost a pvector, structurally similar so it can be 'adopted' as immutable in constant time
21:14	Chouser	ah! cool.
21:14	rhickey	precisely same node structure
21:14	rhickey	just added a bit for editing
21:15	rhickey	I'm still stunned at how fast the tree structure is
21:18	rhickey	those times are for a million writes and 10 million reads of a 100k item arraylist/pvector
21:18	rhickey	in ms
21:18	Chouser	is lazilypersistentvector still useful, or might you just as well build via mutable pvector at read time?
21:18	rhickey	Chouser: dunno yet, probably not needed now
21:19	rhickey	most readable things are small, its vec/vector/into that are going to blaze now
21:20	rhickey	and the mutable mode will be available for end user code as well
21:22	rhickey	I prevent: using the mutable thing with the normal fns (conj etc), immutability of result of immutable call. Only caveat is against sharing the mutable thing - I can't help you then
21:22	rhickey	er, protect immutability
21:23	Chouser	I really can't imagine it'll be a problem
21:23	Chouser	you can use ArrayList today if you want to
21:23	rhickey	exactly, but not with these features
21:24	Chouser	and if you accidentally forget to (immutable v) before you return it, either nobody will conj and it won't matter, or the case will be caught.
21:24	Chouser	right
21:24	Chouser	one key feature being exactly that latter protection.
21:25	rhickey	and the free persistent 'copy', no copy involved
21:25	Chouser	yep. that free copy is one key way this is different from, say, google protobuf builders.
21:26	rhickey	the pvector impl is still pretty gnarly
21:26	rhickey	:)
21:26	Chouser	heh. I don't doubt it.
21:26	rhickey	revisiting it was not fun
21:26	Chouser	oh!
21:26	Chouser	you mean the way it was previously was gnarly?
21:26	rhickey	yeah
21:27	rhickey	lots of bit shifts etc
21:27	Chouser	ah, yes. I know that very well.
21:27	Chouser	Doesn't look any better in JavaScript.
21:28	rhickey	one question is, how much functionality to put on the mutable versions, right now just basic assoc!, conj!, pop!, and nth!
21:28	Chouser	Kinda doubt it will in Clojure either. ;-) what it's doing is gnarly.
21:29	rhickey	Chouser: I was trying to figure out today why that was - it's a recursive structure, but the tail etc just make it a pile of edge cases
21:29	rhickey	also it grows up, not down
21:31	rhickey	in any case, these perf tests show it will be well worth it, no compromises in using Clojure's data structures
21:32	rhickey	I think the technique will generalize to any tree
21:32	rhickey	so once someone sees the vector, maybe they can do the maps
21:36	Chouser	will be well worth what? having made this improvement to vector so that it can compete even better against plain mutable structures?
21:37	rhickey	yes
21:37	arohner	yes, I can take advantage of a 3x performance boost
21:39	rhickey	also once you're at 1129/1354 you can say there's nothing substantial to be gained by switching to a java thingy
21:39	Chouser	didn't we say that anyway? ;-)
21:40	rhickey	Yes, I thought at 2-4x it was worth writing Clojure
21:41	rhickey	that was my target
21:41	rhickey	but the question becomes, once you have a bottleneck, do you have to leave everything good behind?
21:42	rhickey	switch code radically
21:42	Chouser	yeah, that's similar to your response at the meetup to my all-too-frequent joke about speed.
21:42	Chouser	and it's a good response
21:43	rhickey	my ultimate goal is to keep people away from Java mutable data structures other than j.u.concurrent, and away from arrays
21:43	rhickey	the latter is the toughest nut to crack
21:43	Chouser	arrays are going to be ... yeah
21:44	Chouser	vectors of primitives will help a lot of the memory-concious cases.
21:44	rhickey	but all of this (incl chunks) paves the way for vectors of primitives
22:10	mebaran151	what would be the best way to test if a class is a given subclass of another class?
22:10	hiredman	well, if it's given then = should be fine
22:11	hiredman	if you just want to test if a class is a subclass of another class, you can use isa?
22:11	mebaran151	isa? sounds right
22:12	mebaran151	but (isa? Exception IllegalArgumentException) returns false
22:12	Chouser	other way 'round
22:12	skalnik	Clojure is awesome :)
22:13	Chouser	,(isa? IllegalArgumentException Exception)
22:13	clojurebot	true
22:13	mebaran151	thanks Chouser, I wish I'd put it the right way at first
22:14	Chouser	arg ordering is frequently hard to guess at
22:14	hiredman	(doc isa?)
22:14	clojurebot	"([child parent] [h child parent]); Returns true if (= child parent), or child is directly or indirectly derived from parent, either via a Java type inheritance relationship or a relationship established via derive. h must be a hierarchy obtained from make-hierarchy, if not supplied defaults to the global hierarchy"
22:15	_hrrld	It's so cool that the seq abstractions work on built-in Java objects.
22:20	Chouser	it's somewhat interesting that despite it's homoiconicity, the data structures used most commonly for runtime data are used least commonly to represent code.
22:21	_hrrld	Isn't that because most runtime data isn't code? ;) Everyone is choosing the structures most amenable to their data.
22:23	Chouser	yes, indeed! but that's rather different from older lisps.
22:23	Chouser	...where you had a list, so you generally made so with that for both code and data.
22:23	_hrrld	Where the data structures to choose from were so much more limited.
22:23	_hrrld	yes.
22:24	Chouser	I don't think clojure's lists are particularly better at representing code than clojure's vectors would be.
22:24	_hrrld	What language will take advantage of array backed hash map tries for its code representation?
22:25	Chouser	Is the main benefit of using both lists and vectors in code that they stand out visually from each other?
22:25	hiredman	I was talking to someone in #scala a few months back that wanted a homoiconic language based on maps
22:25	hiredman	but he might have just been winding me up
22:25	Chouser	hehe
22:25	_hrrld	For nondeterministic evaluation!
22:26	arohner	hiredman: there's a language like that
22:26	arohner	it's a toy
22:26	Chouser	I've thought things like that before ... though I wanted order-preserving maps.
22:26	Chouser	...and that was before I knew lisp or what homoiconicity was.
22:26	arohner	http://lambda-the-ultimate.org/node/2887
22:37	mebaran151	what would dispatch function look like that would choose a multimethod based on whether or not the selector was a parent?
22:38	mebaran151	kind of like dispatching with type except it wouldn't be strict equality
22:41	hiredman	multimethods don't dispatch on strict type equality
22:42	hiredman	it wall walk up the hier if no equality exists
22:42	mebaran151	no I know, I'm trying to come up with a function that would dispatch this way
22:43	mebaran151	essentially I'm just passing in type right now as the dispatcher, but it would be nice to catch the class children too
22:43	hiredman	erm
22:44	hiredman	lisppaste8: url?
22:44	lisppaste8	To use the lisppaste bot, visit http://paste.lisp.org/new/clojure and enter your paste.
22:45	lisppaste8	hiredman pasted "defmulti example" at http://paste.lisp.org/display/83721
22:45	hiredman	it sounds like that is what you want
22:45	mebaran151	http://clojure.pastebin.com/m48244d1b
22:46	mebaran151	but I don't know if it would catch the children too
22:46	hiredman	it will catch subclasses of those two exceptions
22:46	hiredman	but you do the same thing for both
22:47	mebaran151	right now yes, but I'll handle them differently later
22:47	mebaran151	how does multimethod determine the match?
22:47	hiredman	it checks for an exact match, then it walks the superclasses looking for a match
22:48	mebaran151	I thought it used whatever the the function returned, kind of like a dispatch table. Are the rules more nuanced?
22:48	hiredman	yes
22:48	hiredman	it is also possible to use heirarchies other then the java Class heirarchy
22:48	mebaran151	that was what I was looking for
22:49	mebaran151	so you can pass it a custom set of rules for determining what to consider a super and subclass?
22:49	hiredman	,(doc defmulti)
22:49	clojurebot	"([name docstring? attr-map? dispatch-fn & options]); Creates a new multimethod with the associated dispatch function. The docstring and attribute-map are optional. Options are key-value pairs and may be one of: :default the default dispatch value, defaults to :default :hierarchy the isa? hierarchy to use for dispatching defaults to the global hierarchy"
22:49	hiredman	no
22:49	hiredman	you pass it a hierarchy
22:49	hiredman	which it walks just like it would walk the Classes
22:50	hiredman	mebaran151: http://clojure.org/multimethods
22:51	mebaran151	do the hierarchies always work in terms of symbols?
22:52	hiredman	if you look at the link I just pasted, you will see hierarchies using keywords
22:55	skalnik	ataggart: I got that infinite fib-seq done :)
22:57	mebaran151	thanks hiredman, I think I conflated symbols and keywords
22:58	hiredman	,(derive (Object.) :bob)
22:58	clojurebot	java.lang.Exception: Assert failed: (namespace parent)
22:59	hiredman	,(doc derive)
22:59	clojurebot	"([tag parent] [h tag parent]); Establishes a parent/child relationship between parent and tag. Parent must be a namespace-qualified symbol or keyword and child can be either a namespace-qualified symbol or keyword or a class. h must be a hierarchy obtained from make-hierarchy, if not supplied defaults to, and modifies, the global hierarchy."
23:00	Raynes	(doc intern)
23:00	clojurebot	DENIED
23:06	mebaran151	anyway, thanks hiredman. I'm pleasantly surprised the default behavior of clojure
23:07	hiredman	multimethods are very cool
23:08	grrrt	speaking of which... do all functions defined with defmethod need to be of the same arity?
23:08	Chouser	nope
23:09	grrrt	spiffy
23:09	hiredman	but the dispatch function needs to be able to handle all the arities
23:09	Chouser	but if the number of args you pass in cause either the dispatch fn or the method chosen to error out, then you get an error
23:09	grrrt	a dispatch fn could then be (fn [&args] ....) i suppose?
23:09	hiredman	(comp type first list)
23:09	Chouser	yep
23:10	grrrt	I needed that a couple of days ago, can't remember what for though. but that's really cool
23:39	gko	How to key of a value in a map?
23:39	gko	how to get key of value in a map?
23:40	gko	Oh... of course, no build in function, since value can be not unique..
23:40	hiredman	,(map #(.getName %) (.getMethods (class {})))
23:40	clojurebot	("seq" "withMeta" "withMeta" "withMeta" "assoc" "assoc" "valAt" "valAt" "entryAt" "without" "assocEx" "count" "iterator" "containsKey" "capacity" "empty" "empty" "create" "equiv" "invoke" "invoke" "get" "put" "hashCode" "clear" "equals" "toString" "isEmpty" "values" "size" "entrySet" "putAll" "cons" "remove" "keySet" "containsValue" "applyTo" "call" "applyToHelper" "throwArity" "invoke" "invoke" "invoke" "invoke" "invoke"