freenode/#lisp - IRC Chatlog

0:28:34 tetero Is it fast? (ulisp)

0:40:04 random_numbers I don't currently have a board to test it at the moment, unfortunately.

1:57:40 Suzuran42 ** NICK Suzuran

2:20:38 aeth tetero: This is a channel for Common Lisp, often just referred to as Lisp. ##lisp is for the Lisp family of languages. The fastest Lisp you'll find, though, is probably SBCL, a CL implementation. There are also some fast Schemes, but the popular/complete Schemes aren't as fast as those.

2:21:25 aeth It looks like uLisp is very similar to Common Lisp, perhaps even a subset, but it's a subset that doesn't include many things you'd want if you wanted to write very fast CL. This appears to be all of it: http://www.ulisp.com/show?3L

2:22:03 |3b| ACTION suspects not having things is good if you are writing for arduino :p

2:22:05 aeth (It's not quite a subset because it has a different loop.)

2:22:20 aeth Notably missing are arrays, type declarations, and multiple return values.

2:22:30 aeth And defstruct.

2:22:45 aeth If I wanted a fast CL subset, I'd include all of these.

2:22:53 |3b| not on a system with 2k of ram

2:23:32 |3b| (well, maybe you would, but i'd probably just go straight to memory access instead of full CL arrays)

2:24:56 aeth Oh, I forgot to mention that SBCL is fast at the expense of RAM (64 bit SBCL takes about 100 MB as a minimum on startup). And the style I use generally prioritizes CPU at the expense of RAM, too.

2:25:23 aeth But a lot of the principles are in common, and so a slim CL would have many shared features with what you'd want in a fast SBCL or CCL program.

2:30:37 |3b| ulisp looks nicer than the last microcontroller lisp i looked at, may have to consider that next time i think about doing hardware stuff

2:33:08 Bike missing type declarations seems weird. is it not cross compiled?

2:33:45 Bike maybe there just aren't bignums?

2:40:00 aeth Bike: ah, good point

2:40:39 Bike "slim CL" is kind of an oxymoron though

2:40:47 aeth I guess its numbers are just a signed 16 bit integer, and that's it?

2:40:55 aeth I'm guessing this because it says its max is 32767

2:41:07 Bike oh, i missed that

2:41:19 aeth I did, too, until you said it.

2:41:34 |3b| yeah, 16 bit int, symbol or cons seem to be only types

2:41:48 aeth "slim CL" as in, it looks mostly like a subset of CL, instead of being Scheme-style or radically different and just parenthesized.

2:41:57 random_numbers The only one I know of that's cross-compiled is ChickenScheme. Then again I only started looking today.

2:42:03 aeth Some "Lisps" really are just random languages in s-expression form

2:42:19 random_numbers Any other one you'd recommend? (I discount Hy and Clojure. They're interpreted, not compiled.)

2:42:40 random_numbers Fuzzier to say that with Clojure but whatever.

2:42:54 aeth Bike: it looks like it even has funcall, which suggests Lisp-2

2:43:31 aeth On the surface, it looks very CL-style, although it's disappointing that it doesn't have arrays or structs.

2:43:44 Bike clojure isn't compiled?

2:44:01 aeth Bike: I'm guessing random_numbers is talking about AOT native compilation

2:44:09 |3b| from their benchmarks seems to be ~500-1000 times less efficient than sbcl by instruction count

2:44:33 random_numbers aeth: Bike: Yes. I'm disregarding VMs because those aren't quite practical on small SoCs.

2:44:48 Bike i mean, i thought it was compiled to JVMese

2:44:58 |3b| hard to say how much of that is from running heavily recursive but otherwise simple benchmarks

2:44:59 random_numbers It is. It's never truly interpreted.

2:45:08 Bike kay, makes sense.

2:45:36 random_numbers Hy is transformed into Python AST and executed the same as native code so it depends on circumstances. Still all runs on Python VM.

2:45:38 Bike not having arrays is kind of weird. structs would be more difficult since you'd probably want something more like c structs. or at least with typed slots.

2:45:55 Bike if you don't have typed slots a struct is just an array with newtype, who gives a fuck

2:46:07 |3b| ACTION is more disappointed by lack of macros than structs or arrays :p

2:46:10 random_numbers Being used to Clojure I'd say that if maps aren't too heavy/slow, they can do pretty well in a pinch.

2:46:28 Bike whoa, whoa, no macros

2:46:30 Bike fucked uuuuup

2:46:47 aeth Bike: I use typed slots in my structs in CL, and I assume that they do something in SBCL.

2:46:53 Bike no point then. might as well use a macroassembler

2:47:02 Bike aeth: they do

2:47:07 random_numbers Really? Well. That's kind of a significant change that doesn't make that much sense. It's not as though macros are ever executed at run-time.

2:48:32 |3b| ACTION also doesn't see memory access anywhere :(

2:55:03 random_numbers That's a shame.

2:55:15 random_numbers Are there any Lisps that transpile nicely to C?

2:55:33 Quadresce What does "transpile nicely" mean

2:55:49 Quadresce The abstractions of lisps exceed C a lot

2:56:19 random_numbers Mostly that it does interop properly and compiles. I'm not asking for something that generates legible/maintainable code.

2:56:43 Quadresce ECL for Common Lisp

2:56:52 Quadresce Chicken Scheme for scheme

2:57:10 random_numbers Heard of Chicken, didn't know ECL actually transpiled too.

2:57:44 Quadresce I think MOCL too?

2:58:09 random_numbers Huh. Mobile Lisp. That's nice and def something I'll look up.

2:58:29 random_numbers Commercial though. Shame.

2:58:53 Quadresce It targets commercial platforms. Seems reasonable.

2:59:11 random_numbers Perhaps, but it means I can't get in.

2:59:46 random_numbers Clojure's a fair bit easier to use on mobile in comparison.

3:06:41 vtomole I can't hack clojure cause the arguments for procedures are represented as vectors instead of lists. (square brackets, uh)

3:07:42 random_numbers It has its idiomatics pecularities, yeah.

3:08:03 random_numbers I picked it up as a pretty-looking functional-first language.

3:12:02 vtomole haha yeah it's cool. Im just petty.

3:20:39 tetero aeth: Only reason I asked was because someone just mentioned it ;)

3:21:50 random_numbers tetero: Yeah, I found it and wondered what people thought of it and if they'd heard of it. Some people dug into it far more than I expected and found it was somewhat lacking.

3:22:17 tetero Ah

3:34:51 akkad Quadresce: actually for mobile LambdaNative uses gambit and is much more performant than chicken.

3:51:03 random_numbers Having read through some of ECL's docs, I like what they did with the implementation.

3:52:44 random_numbers akkad: That's rather neat.

3:57:39 random_numbers LambdaNative is very nice.

4:36:40 aeth tetero: sorry, my brain's parser must be broken. I didn't parse the "ulisp" within the URL further up and thought that you were the first mention

5:09:05 beach Good morning everyone!

5:12:02 akkad gambit is amazingly fast

5:21:58 tetero aeth: np

5:33:44 reinuseslisp does any library offer a portable way to get cpu threads?

5:34:19 reinuseslisp something like a trivial-cpu

5:39:53 akkad linux yes.

5:40:53 random_numbers akkad: Gambit-C?

5:45:27 akkad yes.

5:53:32 mfiano_ Am I correct to assume static-vectors works for sbcl on windows?

5:53:45 mfiano_ By the way, this is axion on another computer

6:19:49 Suzuran42 ** NICK Suzuran

6:22:26 shka reinuseslisp: system threads\

6:22:45 shka reinuseslisp: https://common-lisp.net/project/bordeaux-threads/

6:26:10 reinuseslisp shka: bt is for (logical?) threads, I mean something like unix's sysconf(_SC_NPROCESSORS_ONLN)

6:26:38 reinuseslisp or C++'s std::thread::hardware_concurrency()

6:29:14 shka rigth

6:29:20 shka never cared about it

6:40:32 jackdaniel reinuseslisp: bordeaux threads is just a portability layer which maps into threads in common lisp implementation

6:40:40 jackdaniel most implementations use native threads

6:41:00 reinuseslisp yes, I know that

6:42:27 reinuseslisp I just didn't find a word to differentiate CPU cores/threads and application (bordeaux-) threads

6:51:21 jackdaniel one way to do that with cffi is

6:51:27 jackdaniel (cffi:defcfun "sysconf" :long (name :int))

6:51:32 jackdaniel (sysconf 84)

6:52:16 jackdaniel you can define 84 as constant to give it a meaningful name

6:56:44 reinuseslisp jackdaniel: yes, but win32 seems not to offer that API

6:56:51 reinuseslisp https://stackoverflow.com/questions/150355/programmatically-find-the-number-of-cores-on-a-machine

6:57:16 reinuseslisp #+windows may work

7:11:46 jackdaniel there are bindings to openmp for CL, if that's not an overkill, you could use that

7:12:48 jackdaniel but I can't find the link, so maybe I'm wrong

7:16:10 Suzuran42 ** NICK Suzuran

7:50:03 schweers does anyone have an idea why `prin1' or `format' with "~S" would emit a linefeed after printing a symbol? It seems to me that this only happens when printing on a socket stream.

7:53:21 jackdaniel they don't: (with-output-to-string (s) (prin1 'a s)) ;-> "A"

7:53:32 jackdaniel it might be return value in REPL what calls terpri

7:54:16 schweers that’s the weird thing: when I output to a string all is well

7:54:34 schweers I could swear that the code in question worked before, I’m really stumped why this happens now

7:55:22 schweers also I’m not doing anything weird with the readtable, at least for this test

7:56:12 jackdaniel some minimal test case to reproduce?

7:56:29 schweers working on it

7:59:26 jdz Printing has nothing to do with readtables.

8:00:11 schweers damn, on the minimal test all is well

8:00:26 schweers jdz: readtable-case has influence on how symbols are printed

8:01:12 jdz True that. But in this case I suspect the problem is in pretty-printing.

8:01:49 schweers well, I have to change the readtable-case as I’m interacting with a scheme, but that’s disabled for now

8:03:59 schweers wait, pp should not be turned on, that’s why I’m using prin1 and/or format with "~S"

8:05:24 schweers well, I can’t reproduce it outside of my program, which I find quite disconcerting. Oh well, more debugging it is.

8:05:28 jdz PRIN1 only binds *PRINT-ESCAPE* to true, nothing about layout control.

8:05:48 schweers that’s exactly what I too thought.

8:06:26 schweers Well, if no one else has an idea, I suspect it really is a problem with some other code I’ve written, i.e. the problem is not me misunderstanding `prin1'

8:07:15 jdz Have you tried binding *PRINT-PRETTY* to NIL?

8:08:10 schweers no, I’l try it

8:08:29 schweers ahhh, thank you so much, that did it!

8:08:55 schweers somehow that possibility completely escaped me

8:10:05 jackdaniel (setf *print-ugly* t) ;)

8:10:23 schweers hehe

8:10:40 schweers pff, my expressions should be pretty enough ;)

9:15:04 nimiux_ ** NICK nimiux

11:04:34 phoe Is there any function that, given a cons and a value, destructively updates that cons in a way similar to PUSH?

11:05:05 phoe Like - I can write it myself, it's simple, but is there anything like that out in the wild already?

11:07:08 _death it depends on the place you give to push

11:29:06 phoe Is (defgeneric foo (bar bar bar)) valid?

11:30:33 Bike uh, probably not

11:30:36 TMA phoe: you mean something like (lambda (cons value) (push (cdr cons) (car cons)) (setf (car cons) value)) ?

11:30:42 phoe TMA: yes.

11:30:47 phoe Bike: neither SBCL nor CCL complain about it.

11:31:12 TMA clhs defgeneric

11:31:12 Bike i still say it's weird

11:31:12 specbot http://www.lispworks.com/reference/HyperSpec/Body/m_defgen.htm

11:31:41 Bike the parameter names in the defgeneric are irrelevant except for debugging, so i can imagine it working and an implementation not bothering ot check

11:31:44 Bike but don't

11:32:08 phoe ECL chokes when I try to DEFMETHOD on that generic function

11:32:14 phoe but SBCL allows this

11:32:19 phoe and the method works