libera/commonlisp - IRC Chatlog

19:54:44 Guest74 I really need to fix my recording functions. recording this https://youtu.be/zgXHWmBH00o has brought my laptop to a crawl. a single thread running an ffmpeg process shouldn't cause this should it? I wonder if sbcl is reading all the output? Shouldn't this slowdown go away after the thread is killed?

19:55:03 Guest74 I'm going to have to reboot, it's still slow like half an hour after.

23:05:52 mister_m Does CL have anything comparable to elixir's Enum.chunk_every? https://hexdocs.pm/elixir/Enum.html#chunk_every/4

23:08:43 Guest74 Not that I know of. Doesn't look hard to write though.

23:13:28 Bike nothing standard. serapeum has the case of count = step as "batches". other utility libraries might also.

0:02:42 Guest74 so maybe not as easy as I thought.

0:19:06 Guest74 mister_m: There has to be an easier way. It's a three line loop if you don't care about adding elements or removing lists. 4 to remove lists that are less than count. And this mess is what I got for everything. https://plaster.tymoon.eu/view/2769#2769

0:29:49 rotateq ":when (not (null current))" <- ":unless (null current)"

0:37:07 rotateq hm is the nested "setf current" by intention?

0:46:41 Guest74 nested? if you mean the new lines it's so it doesn't run off the screen. I choose when or unless based on which I think reads better.

0:47:07 rotateq no because of your form in symbol-macrolet above

0:47:08 Bike as in, after macroexpansion this will be like (setf current (typecase leftover ... (cons (setf current ...))))

0:47:45 rotateq yes that

0:48:32 Guest74 it's definitely not pretty.

0:49:06 Guest74 oh I see.

0:49:41 Guest74 an error that doesn't cause errors.

0:50:06 rotateq and beware, even when current is an empty vector NULL will return NIL

0:52:21 Guest74 though never happens here.

0:52:52 rotateq that's no reason to let it be this way :)

0:53:34 Guest74 You propose I change something for something that will never happen?

0:53:45 Guest74 what do you propose?

0:55:10 rotateq that this is no argument for it when you just think be "seeing" that it shall never happen. things are also by reading intention

0:55:44 Guest74 what do you proose

0:55:48 Guest74 propose

0:56:02 rotateq i can always put more to it which "can never happen"

0:56:31 Guest74 what do you propose? or how do you see it happening?

0:56:58 rotateq by reading your lines

0:58:49 Guest74 What do you propose as a solution. Can you give an example of how this would happen?

1:00:10 rotateq i didn't say it's about anything that could happen, just the intention you give to your readers

1:00:51 Guest74 so what is the intention I'm giving with that according to you?

1:00:53 rotateq and no, i don't even have until now the right specification of what should clearly be done by the elixir thing

1:01:01 rotateq *sigh*

1:01:55 Bike http://ix.io/3Hxy here's my take

1:02:04 Bike kind of fiddly

1:02:12 rotateq that while you cover lists and vectors it wouldn't handle just lists right

1:02:46 rotateq ah you set the &optional different

1:04:28 Guest74 ah, class-of! I was wondering how to get just vector and not vector some length

1:04:46 Bike i guess interiorp doesn't need to be its own variable

1:07:45 Guest74 I kinda cheated on :discard

1:08:15 Guest74 i like default of step as count as well.

1:12:33 Bike that's what the elixir page says, right?

1:13:34 Guest74 oops. I just looked at the examples.

1:13:46 Guest74 ...and I don't know elixir.

1:14:25 Bike me neither

1:14:53 Guest74 I should have got it from looking at the examples though. they have without step.

1:19:54 Guest74 I think you're missing collecting just the short chunk if I'm reading yours correctly.

2:37:42 Bike it works with all the examples on the elixir page, at least

2:38:04 Bike (chunk '(1 2 3 4) 3 3 ()) => ((1 2 3) (4))

2:38:16 Guest74 well, I think I've already established I don't read very well :)

2:38:50 Guest74 ah, i see now.

2:39:46 Guest74 I think I used nil because I don't like typing empty lists. but that makes the code so much easier.

4:00:50 beach Good morning everyone!

4:08:32 fe[nl]ix hey beach

5:36:16 lisp123 Has the idea of (slightly after) compile-time determination of generic-function calls to the relevant specialised method been thought of in the past?

5:37:54 lisp123 In the event new classes are not added after compiliation, then each function call could be fully determined right after compile time? And if that is correct, then one could inline the actual methods vs. runtime determination

5:38:34 lisp123 Perhaps have like a marker for particular functions to say "hey dont worry about classes created at runtime, just inline these at compile time"

5:40:48 Nilby Most things in CL, including CLOS, was designed with the idea and possibility that a "sufficiently smart compiler" could make it fast. So yeah. Not that it always works out.

5:43:14 lisp123 I see. So a compiler already does something similar

5:44:07 lisp123 That would make sense. Until a new class is created during runtime, you can use the old specialisations and avoid having to recheck

5:45:11 Nilby Compilers are allowed to inline things where they can, which is why all those redefinition warnings can actually be useful, becuase if the other uses aren't recompiled it could be using the old inline version.

5:45:18 edgar-rft Classes still an be re-defined and inheritance order still can be changed, both at run-time.

5:47:24 lisp123 Makes sense!

6:00:58 beach lisp123: Yes, one idea is "sealing" as heisig explains in his paper on fast generic functions.

6:01:25 beach lisp123: Another idea is to alter the call site in some way when the generic function changes.

6:01:38 hayley lisp123: Are you thinking of (polymorphic) inline caching?

6:02:00 beach lisp123: You can look at my paper on call-site optimization for the latter.

6:02:52 hayley An inline cache avoids performing a full lookup almost all of the time, and still permits redefinition by "emptying" the caches.

6:03:39 lisp123 beach: Thanks! I will read these. I read bits of http://metamodular.com/SICL/generic-dispatch.pdf, but to be honest I skimmed to the unique number bit so maybe missed other parts

6:03:50 lisp123 Is it a different paper?

6:04:01 beach Yes. Hold on...

6:04:33 beach First entry: http://metamodular.com/SICL/

6:05:58 beach Nilby: What does the standard say about inlining generic functions? I would be surprised if it were allowed to inline a generic function in the state it is in when the caller is compiled.

6:06:26 lisp123 hayley: I was looking at Swift which basically copied many of CL's ideas and renamed them. But instead of mixins, they would inline the code directly by synthesising the different components

6:07:15 hayley ACTION waits for Swift to copy tracing garbage collection and multiple dispatch. Oh wait.

6:07:56 lisp123 beach: Thanks!

6:08:04 hayley But "mixins" are a design strategy, and inlining is a compilation strategy. You could still program using mixins, if the compiler inlined everything.

6:08:27 beach lisp123: I don't know Swift, but is it a dynamic programming language, by which I mean a language where the semantics are defined by a sequence of interactions in a global environment?

6:08:38 lisp123 hayley: Seriously, they go out of their way to rename things when its clear they took inspiration from CL. Blatant plagiarising, but then they hide a lot of their details so its a complete mess anyway

6:08:55 hayley You might want to read <https://bibliography.selflanguage.org/_static/pics.pdf> on polymorphic inline caches.

6:08:56 lisp123 beach: No, not at all. All compiled

6:09:11 beach lisp123: I am not talking compiled vs interpreted.

6:09:14 hayley That is again an impedance mismatch.

6:09:16 beach I am talking static vs dynamic.

6:09:38 hayley Swift is defined by a set of source files, i.e. static.

6:09:46 beach lisp123: In a static language, you have none of the problems you have in a dynamic language with adding methods at run time.

6:09:55 lisp123 Yes, static

6:13:02 beach lisp123: A programming language is neither compiled nor interpreted.

6:13:29 beach lisp123: That's an implementation strategy. You can compile Perl and interpret C.

6:13:46 lisp123 beach: So I was wondering the value of giving people the option to specify certain methods not to be superceded by other methods defined at run time

6:14:23 beach lisp123: It can have a great impact on performance, as heisig's technique takes advantage of.

6:15:17 lisp123 beach: Nice! Well time for me to shut up and read :)

6:16:06 beach However, heisig is now convinced that my technique for call-site optimization will largely supersede his technique.

6:19:45 beach So a "programming language" is a specification that determines the syntax and semantics of conforming programs as well as the result of some non-conforming programs.

6:21:30 lisp123 beach: Thanks, I need to remind myself to be more precise on the terminology