#clojure – 2008-05-14

2008-05-14

08:26	cgrand	rhickey: I used gen-class to extend java.io.Writer but overloaded methods make this not so fun (too much type tests and var-bindings to access super's for some arities). Would you be ok with a way to bind a method signature to a different var?
09:31	asbjxrn	Is there a way to turn 'foo/bar into 'bar ?
09:33	asbjxrn	or alternatively: Is there a way to do ({'foo 1 'bar 2} 'gaz/bar) to return 2 and not nil?
09:35	cgrand	asbjxrn: (symbol (name 'foo/bar))
09:36	asbjxrn	Works.
09:40	asbjxrn	How about: Is it possible to make this return just bar :(do (defmacro foo [] `'bar) (foo))
09:44	asbjxrn	(This thing is happening at rather high frequencies, better to create the symbol without the namespace instead of removing it later by going by the string if possible.
09:45	cgrand	and using keywords instead of symbols?
09:46	asbjxrn	I'd prefer not to.
09:47	asbjxrn	Hmm. On the other hand... It's kind of a parser thingy, and translation to keywork would only happen once. Maybe that is a better option.
09:49	cgrand	If you persiste with symbols: (defmacro foo [] ''bar) ({'bar 2} (foo))
09:49	cgrand	-> 2
09:49	rhickey	asbjxrn: why wouldn't the symbols get resolved correctly, i.e. one of the points of using symbols is to get the resolution - so use ` instead of '
09:50	rhickey	{`foo 1 `bar 2}
09:50	rhickey	that way your foo and bar are different from mine
09:51	rhickey	and a key advantage over keywords
09:51	rhickey	cgrand: looking at Writer now so I can look at what you are talking about
09:55	rhickey	cgrand: so the issue is that providing a version of write overrides all super versions?
09:55	asbjxrn	Ok. {`foo 1 `bar 2} Sounds like a plan. I'm still fiddling with my environment map, which will cointain built-in and user specified variables. greating the built in variables in "my" namespace might be the right thing.
09:56	cgrand	rhickey: yes and I have to check the types of the arguments to know which write was called
09:57	rhickey	a type-case macro would make the type-switch easier
09:57	cgrand	I have a patch that allow me to write :methods [['append [java.lang.CharSequence] java.io.Writer :as 'append-cs]]
09:58	rhickey	append(char) still goes to super?
09:58	cgrand	yes
09:58	rhickey	hmm...
10:00	asbjxrn	This damn thing started out as some helper functions and are quickly turning into a bloody language. (Just because I couldn't manage to integrate processing and clojure.) At least I'm learning, I hope.
10:00	rhickey	one of the design objectives was to minimize the number of cases where an implementation change would require a re-gen
10:02	rhickey	cgrand: so if you later wanted to implement append(char) similarly, you would need another :methods entry
10:05	cgrand	yes -- and it would clash with the previous def if I'm using gen-and-load-class
10:05	rhickey	right re-gen means re-load means re-start :(
10:07	rhickey	so I'd prefer a dynamic convention for specialized overrrides - perhaos a two stage process - look up append var, if not found look for append-char or some thing, just for overloaded methods
10:07	cgrand	hmm...
10:12	rhickey	append-char, append-CharSequence, append-CharSequence-int-int
10:13	rhickey	some convention for array names
10:13	rhickey	<>
10:13	rhickey	write-char<>
10:15	cgrand	I was thinking of prepending method descriptors (but it would require some escaping)
10:16	cgrand	I meant appending...
10:28	cgrand	I just noticed you proposed to first look for a general handler and if not found for a specific... so as to not slow down the "normal" path, I guess.
10:28	rhickey	sometimes it's convenient to route the overloads to a single function
10:32	rhickey	the special process would only be done for overloaded methods in the first place, so the normal, non-overloaded path doesn't do a speculative lookup
10:36	cgrand	I agree: in some cases all overloads can be handled by a single function and it's better not to cripple the normal path. I was surprised because I was thinking of doing it the other way: first look for a specific handler else look for a fallback (and by default the specific var could point to the var containing the fallback function) -- but your way is certainly better in the normal overloaded path.
10:40	rhickey	Oh, I see. I'm not sure general before specific is better - it will force there to be a specific for all overloads if there is a specific for any
10:41	rhickey	I wouldn't let perf considerations dominate here - the perf will be good, the lookups aren't really lookups, just binding tests
11:21	asbjxrn	This stuff is making me feel stupid. The `foo solution presents some other problems. You guys got (a lot of) free time to pound some macro/substitution stuff through my skull?
11:21	lisppaste8	asbjxrn pasted "evaluation of symbols in macros" at http://paste.lisp.org/display/60762
11:21	asbjxrn	Kind of a writeup here:
11:26	cgrand	asbjxrn: try putting a quote before ~x : (lookup '~x env#) I think it will prevent resolution of your symbol after macroexpansion
11:27	cgrand	(foerget it: I read too quickly)
11:29	rhickey	one trick with macros is to make as little as possible a macro or backquote. If the purpose of making it a macro is just to do quoting for you, then have the macro do just that part, then call a function to do the work
11:30	asbjxrn	Not just quoting, I don't want the "width" to be evaluated until the function that the macro creates is called later.
11:31	rhickey	you can write a function that returns a function
11:31	asbjxrn	But when calling that function-creating function, the arguments will be evaluated, right?
11:32	asbjxrn	(say instead of background, I have a function (circle radius x y)
11:34	asbjxrn	And I want to call it with (circle 10 mousex mousey) where mousex and mousey are entries in the environment map that are set before the function that circle returns is called...
11:34	asbjxrn	(kinda convoluted explanation...)
11:35	asbjxrn	(wouldn't circle have to be a macro to avoid mousex and mousey to be evaluated (They do not even exist when the circle function is called.))
11:42	rhickey	right now, background expands into a fn literal, embedding the key x (or trying to). Imagine there was an ordinary function, make-bk-fn, that took an env and a key, and returned a fn of env. background could expand into a call to make-bk.
11:43	rhickey	make-bk-fn
11:47	asbjxrn	Ok. But that still leaves me confused as to how to embed the symbol. (I'll try to distill the problem a bit.)
11:50	asbjxrn	Is there a (number? x) function?
11:51	rhickey	(instance? Number x)
11:55	rhickey	(defn mk-bk [key]
11:55	rhickey	(fn [env]
11:55	rhickey	(let [g (env :graphics)
11:55	rhickey	x (lookup key env)]
11:55	rhickey	(. g (setColor (new Color x x x)))
11:55	rhickey	(. g (fillRect 0 0 (lookup `width env) (lookup `height env)))
11:55	rhickey	env)))
11:55	rhickey	(defmacro background [x]
11:55	rhickey	`(mk-bk 'x))
11:57	lisppaste8	rhickey annotated #60762 with "macro rev" at http://paste.lisp.org/display/60762#1
12:01	asbjxrn	Uhm.
12:02	asbjxrn	'x should be ~x?
12:03	rhickey	oops '~x
12:15	asbjxrn	Ok, a little more work and it works pretty much the way I want. Now to understand it and what went wrong in my version so I can make the macro bigger again ;-)
12:15	jteo	yay for you. :)
12:18	lisppaste8	asbjxrn annotated #60762 with "some final work on lookup" at http://paste.lisp.org/display/60762#2
12:18	asbjxrn	In case anyone cares.
12:21	lisppaste8	rhickey annotated #60762 with "capturing resolved symbol" at http://paste.lisp.org/display/60762#3
12:22	rhickey	if you wanted to resolve the symbol at point of macro call
12:22	rhickey	(that should be easier)
12:22	asbjxrn	resolve in this case means?
12:22	asbjxrn	(cl:intern ?
12:25	asbjxrn	Ok, it basically expands 'foo -> bar/foo iff foo is a defined symbol in the namespace bar.
12:26	rhickey	if an api exposed the keys as defs, there would be corresponding ns-qualified vars, and consumers could use normal namespace tools to distinguish my/foo from your/foo, should they be combined
12:28	asbjxrn	Yea, I misread the result of my command, it doesn't return the symbol 'bar/foo, but eg. #<Var: bar/foo>
12:35	asbjxrn	I meant what I said when I said lot of time to pound... ^xv means what?
12:41	asbjxrn	Not sure if it is easier. Would have to add a test to check if x is a number and stuff. Probably easier to deal with in the function.
12:43	asbjxrn	Anyway, it's getting late and my head is spinning. Thanks for all the help.
12:55	rhickey	I meant, that should be easier than it is, not easier for what you are doing
17:41	rhickey	If you are interested in parallel computation, I've added parallel.clj, an interface to the Fork/Join framework. Just put jsr166y.jar in your classpath and off you go!
17:42	rhickey	parallel min/max/reduce/filter/map etc
17:43	rhickey	also added lazily persistent vectors
17:45	rhickey	parallel reduce 3x faster on my quad core
19:42	rhickey	I've added docs to parallel.clj