freenode/lisp - IRC Chatlog
Search
21:51:31
jasom
jmercouris: whether it's good or bad is up for debate, but JSON makes the assumption that "No Object" "False" and "Empty" are distinguishable
21:52:08
jasom
jmercouris: there are ergonomic values to having all of them being treated similarly, but you have less information when you cant distinguish them
21:53:08
jmercouris
I can't think of a scenario in which I've needed this distinction, but I'm sure there is one, something to think about I guess
21:57:23
jmercouris
Shinmera: One of the biggest java proponent arguments for all of this type safety bs, generics etc "Catch the problems during compilation time, save development time"
21:57:35
aeth
CL but statically typed would be the ideal programming language imo. Might need to separate '() and #f like Scheme, though
21:58:18
aeth
jmercouris: Shinmera iirc is working on a game engine. Game engines mostly deal with very well-defined types. Static typing is a huge win here.
21:58:43
Shinmera
The fact that Java sucks balls has very little to do with people finding advantages in strong typing.
21:58:57
jmercouris
Otherwise java programs should have considerably less development time over weakly typed languages, but they dont
21:59:12
jmercouris
Especially because of all of this compiler support and IDE support you get in java
21:59:37
aeth
Java sucks because it sets-in-stone 1990s C++ OOP fads while the world has moved on, even the C++ world.
22:00:16
jmercouris
there had been huge developments in JIT work at Sun, as well as Dtrace, and all sorts of facility integration that would have made java phenomenalj
22:00:31
jmercouris
I believe that Java could be so much more today if shit didn't hit the fan at sun
22:02:19
aeth
Idea? Yes. In practice? The JVM is designed for Java-like languages and the CLR is designed for C#-like languages, and actually-interesting languages can't run on either unless they were designed to make the compromises necessary to run on one or both.
22:03:10
jmercouris
aeth: Are we getting into this transpiler debate again? Last time I was disparaged as some insane heretic for suggesting the word transpiler exists :D
22:03:41
aeth
I think compiling to CL could support more interesting languages than compiling to the JVM.
22:03:58
phoe
In LOOP, can I somehow skip back to the beginning under some condition? Kind of like "continue" in C.
22:04:21
aeth
CL can recompile at runtime. Could you use this to make a JIT language that runs in a CL environment that's not JIT?
22:06:33
jasom
phoe: not really; you just need to have the rest of the loop under a continue, or put the body of the loop in a tagbody and GO to the end
22:07:06
aeth
jmercouris: I thought you needed two parts: (1) analyze the code and (2) recompile the code at run time. If #1 is sufficient, you could do #2 on a language that runs in CL
22:07:32
jmercouris
aeth: You're spot on with your definition, I'm just saying the purpose isn't to have a dynamic system, it's to make compilation quicker
22:08:04
Bike
plenty of jit systems take advantage to speed up generated code in ways that would be difficult ahead of time.
22:09:44
stylewarning
Fare: I have a little minimal example that I cannot get to work or that I do not understand.
22:09:55
Bike
you could do something dumb like (defmacro defunj (name lambda-list &body body) `(let ((counter 0)) (defun ,name ,lambda-list (when (> (incf counter) 10) (compile ',name)) ,@body))) and then you can define a function that's just interpreted at first and then compiled once it's run a lot.
22:10:55
jmercouris
have some sort of "probability" that some branch will be executed and compile or not compile different paths
22:11:19
jasom
I thought the main reason for JITs is to allow efficient compilation of languages where the types are mostly fixed at runtime, but mostly unknown at compile time
22:11:44
jmercouris
jasom: No, the types in java are always known at compile time, so there is your first counter example
22:12:59
dmiles
the reason for JVMs JIT i thought was because of limited memory, that it would be too much to compile the entire 9 million classes
22:13:02
Bike
you can do polymorphism without jit. it just makes it faster, hopefully. which is what jasom said.
22:13:03
aeth
jmercouris: I was more thinking about something like a modern JS-in-CL, which probably requires a fake JIT to support all the magic necessary to have somewhat competitive performance.
22:13:15
Bike
i suppose this means beach's gf dispatch business is jit compilation. that's kind of neat.
22:13:17
aeth
JS is a language that simply does not give enough information to produce fast code without a JIT afaik
22:14:35
dmiles
(the entire JDK and all the classes are "interpreted bytecode" ... some a few other langauges can be that way as well .. even lisp)
22:14:37
Bike
doing it for methods themselves would be a little trickier, you'd need to save method body sources, or at least enough of a source.
22:15:26
aeth
jmercouris: It's not just types, though, JS JITs take quadratic code and make it not quadratic. https://accidentallyquadratic.tumblr.com/post/142387131042/nodejs-left-pad
22:16:00
aeth
JS JITs are what you get when you take very bright minds and put them to work polishing a turd instead of creating the next generation of very high level declarative languages.
22:16:56
jmercouris
I have a lot of complaints about java, but it isn't the worst language in the world, it is very consistent for one, very structured
22:17:11
Bike
with beach's dispatch, if you call a gf with a combination of classes it's never been called with before, it compiles a new discriminating function.
22:17:14
jasom
this is why ltk is way faster to startup if you create and teardown your GUI before saving your image btw
22:17:25
jmercouris
Many JS apologists say "have you tried ES6", and the answer is no, nor do I care to
22:17:56
stylewarning
Fare: I was trying to compile an application, and I thought static-program-op might do magic to make dynamically loaded libraries load faster. I suspect this isn't true (that would be too magical).
22:18:05
jasom
Bike: oh, I missed that part; I was just aware of the part where it used first-class global environments to inline the GF dispatch function and recompile the usage points at redefinition time.
22:18:18
jmercouris
I don't understand how javascript isn't somehow a very blatant copyright infringement
22:18:33
Fare
stylewarning, wait --- by default static-program-op will still load your dynamic libraries dynamically
22:18:38
Bike
jasom: um, i haven't heard of any such part, actually. that sounds more like sbcl make-instance, actually
22:19:23
stylewarning
Fare: Yeah, I figured as much. In any case, I couldn't get this example to work. is it a non-sense example? I forgot the IN-PACKAGE form. http://codepad.org/dlY8nyFR
22:19:38
jasom
Bike: the general idea was just that you could inline functions without regard to whether or not they might be redefined; GF dispatch is just a special case of this.
22:20:34
Bike
jasom: i'm referring to http://metamodular.com/generic-dispatch.pdf which i don't believe involves environments or call sites.
22:20:39
aeth
jmercouris: In case I'm unclear, any use of a fake JIT (or would it count as a real JIT?) within CL would be for embedding JavaScript or perhaps Lua or maybe even Python in CL, ideally with competitive performance (probably somewhat slower, but hopefully acceptably slower) to modern JS engines, LuaJIT, or Pypy.
22:21:26
jmercouris
aeth: I was about to say "what would the advantage of that be?" then you qualified it at the end :D
22:21:28
Bike
jasom: which seems completely different from whatever you're talking about. i'm curious about it; i get the principle of course
22:22:00
aeth
jmercouris: Running any one of those three, or all three with a sufficiently clever engine, would be incredibly useful because of how much code is in any of those. (Lua's probably the most obscure but it does have its niches where it is the dominant scripting language, especially games)
22:22:41
jmercouris
aeth: You could just run them in a subprocess and communicate over some IO channel, it'd be much easier than maintaining the spec /runtime for all these languages
22:23:47
aeth
jmercouris: A JavaScript JIT in CL with good interoperability with the host CL would be a huge win for any web browsers written in CL. So JS would be the most useful.
22:24:10
jmercouris
aeth: For any web-browsers written in CL you say :D? You've piqued my interest :D
22:24:52
jmercouris
aeth: I mostly communicate with JSON right now, and the only way that I can get a value back from the browser is as a return from a javascript function, JS itself cannot directly invoke CL
22:25:41
aeth
The trick would be having a JS engine that can both interoperate with CL and an existing rendering engine (the rendering engine would probably be *far* too complicated to write in CL unless you have VC funding to hire a team)
22:26:13
aeth
jasom: The issue isn't JS, it's JITed JS (or Lua or Python). I know JS and Python both exist in CL, but not fast enough for all uses.
22:27:25
aeth
A fake JITed foo in the CL ecosystem should be somewhere between the actual real JIT and the non-JIT version(s) in performance. e.g. Pypy < some-cl-jit-pthon < Python, or LuaJIT < some-cl-jit-lua < Lua, or modern JS engines < some-cl-jit-js < traditional JS
22:29:27
aeth
The *extra* problem (past the performance issue) that just applies to the embedded JS is if you can also make it useful for interfacing with non-CL code, i.e. the rendering engine. However that works. I don't really have a deep understanding of browser architecture.
22:30:26
aeth
i.e. An acceptable performing JS with JIT-style enhancements through introspection + runtime recompilation could probably work in a JS-to-CL compiler. But can it be used by jmercouris?
22:30:59
aeth
The existence of cl-javascript is a great thing, btw, because now that becomes one of the benchmarks to beat.
22:32:40
aeth
You probably could actually get comparable performance to a real JS JIT engine in JS-to-CL in SBCL if you write a bunch of SBCL-specific code, potentially including assembly using SBCL's inline assembly (is "inline assembly" the right term in a sexpy language?) extension to CL.
22:35:02
aeth
One of my extremely on-the-side side projects that I come back to every few months is writing a Scheme that compiles to CL. But it would actually be very interesting to instead do something Racket-style and try to embed many languages within a CL environment, rather than just one or a few.
22:35:11
jmercouris
I can think of a couple scenarios I have wanted such a behavior, it would allow better interop with existing JS scripts on the web
22:36:36
aeth
One of my long term plans, especially for when my game engine doesn't consume almost all of my CL programming time, is to try to build up an interoperable language environment on top of CL. This exists to some limited extent to some limited other languages in implementation-specific ways... frequently just Prolog.
22:37:16
aeth
s/to some limited extent to some limited other languages/to some limited extent, restricted to a few languages,/
22:38:09
aeth
jmercouris: I have written about 2/3 of a Scheme and about 90% of a Brainfuck, which has given me ideas. JS, Lua, and Python would be nice, but they're trickier because they'd almost certainly require using advanced CL features to do JIT-like things in order to be fast enough to be interesting implementations of those languages.
22:38:38
aeth
Prolog would be nice, too. dmiles (iirc) can tell me why that won't work in existing CL implementations.
22:39:36
aeth
A lot of this might just be fast only on SBCL, ideally with a slow path for other implementations (some libraries work like this already). Popularity of even just one of those embedded languages would pressure implementations to catch up with SBCL there.
22:40:50
aeth
It would be hilariously meta if some of these languages had acceptable performance in ECL... because then you're embedding languages... in an embedded language. But that's actually what Guile was supposed to do when GNU initially pitched it as the standard GNU scripting language in the 1990s (still waiting on that)
22:41:05
dmiles
oh well think the best analogy is i cam up with is "imagine taking an assembly language program that twiddles bitmasks and using bignum math to emulate the registers of the Intel-4930k CPU. You might just see some performance differences? "
22:43:00
dmiles
do this can be an example of where one outsmarted theemselves.. they might be better of just using a bitarray
22:44:12
aeth
jmercouris: Ideally, the embedded languages system will run on SBCL, CCL, and ECL acceptably, and in that order of performance (but ECL surprises me all the time when compared to CCL, so there might be a contest for 2nd place in performance). CLISP will probably never run them performantly and ABCL wouldn't be a priority because it can interface with JVM languages.
22:44:29
aeth
But I'm going to just start with a Scheme because it doesn't need a JIT, unlike most of the other interesting scripting languages.
22:45:12
jmercouris
What is your end goal in doing this? Do you enjoy the project or do you have some product in mind?
22:46:16
aeth
My primary focus right now is making a game engine that can run in Common Lisp. A long term goal of making implementations of other, potentially more popular, languages that very easily interface with Common Lisp helps that primary objective.
22:47:48
jmercouris
I'm just saying, instead of 5 or 6 engines, if you all focused your efforts, you might have a very powerful engine
22:48:11
jmercouris
of course the different engines will have different strengths, paradigms, philosophies, etc
22:48:14
aeth
jmercouris: I started before any other game engines existed except for dto's 2D engines. Also, my approach is very different. e.g. My goal is to have a no-consing game loop so a user of the library could do a call to the GC, disable the GC, and run it between levels on loading screens or whatever, if they choose to (or they can cons in their own code if they choose to, giving them more space on the heap before problems pop up)
22:49:02
aeth
My goal for no-consing has required me to reinvent the wheel quite a bit... and I am going to have to eventually replace cl-sdl2 if I am going to reach 0 consing in my game loop. (All of my own code doesn't cons in the game loop)
22:49:53
aeth
(Note that any implementations of scripting languages will probably cons by their semantics, so the no-consing won't apply to any design of those. And the GC would probably have to be run if you use those.)
22:50:44
aeth
jmercouris: I'm not sure if I have the highest performing code, but it performs pretty fast and makes some unusual decisions in the name of no consing, e.g. I use multiple return values for vectors of size 3 and quaternions. (It'd be a bit too much to use those for 4x4 matrices, so I just declare those dynamic extent and hope for a stack allocation.)
22:51:12
aeth
i.e. during my game loop, I do almost no allocations, except the few cases where I need to do stack allocations (and when I use cl-sdl2 to handle input, which conses, unfortunately)
22:52:35
jmercouris
of course all assets beforehand, and as much pooling as possible, but nonetheless always some allocation somewhere
22:53:19
aeth
The event system is going to be tricky. I'm thinking of using a custom stream that uses a fixed-size byte buffer. So that means I'm *not* going to have the fastest engine if I do that, since there'll be a serialization/deserialization cost. But, on the plus side, that make eventually breaking the code up over the network trivial.
22:53:41
aeth
Also, I'm not sure if I cons in CCL and ECL. I don't think they provide enough profiling information as SBCL.
22:54:06
aeth
With SBCL, I can be pretty sure that if I didn't accidentally introduce new consing, all of my game loop consing right now is in the cl-sdl2 library, which I will probably have to replace eventually.
22:55:37
aeth
I'm not sure if my double-float hack will work on all implementations. In SBCL, I can just work with arrays of double float and as long as I only have constant double floats, double float values stored in an array, or coerce the double floats to something else, then those double floats don't cons! It's surprising to me, actually.
22:56:14
aeth
I don't use double floats in many places, and I can't generalize my vector math to doubles because I use multiple return values to avoid consing with single-floats, and that won't work with double floats.
22:56:43
aeth
Unless I extend my multiple return value system to have stack-allocated "registers" of double-floats
22:59:44
aeth
One other place that will be tricky, probably is text for the UI. I can't preallocate everything there. Maybe I'll have to use some buffer system.
0:18:28
pillton
jasom: Specialization-store can inline calls to "methods" if there is sufficient type information available.
0:19:59
pillton
jasom: https://github.com/markcox80/specialization-store/wiki/Tutorial-3:-Compile-Time-Support
1:27:57
jasom
anyone know if cl-sqlite is still being developed, and if so, where to report bugs? The mailing-list link on the webpage is broken...
2:53:06
myrkraverk
My very first asdf system has a name conflict whenever I re-compile it, when I load it with (ql:quickload) however, on subsequent loads, it succeeds.
2:54:23
myrkraverk
That is, I have a name conflict in :common-lisp-user and :my-package when I load it with (ql:quickload :my-package) and it needs to compile source file again.
3:34:56
Fare
asarch, he hasn't been in #lisp in many years. Not sure he uses it much if at all these days.
4:08:29
asarch
A book that you can use as a reference manual (how to open files, how to list directories, etc)
4:09:35
aeth
PCL is more of an introduction. Common Lisp Recipes (same publisher, different author) is more of a reference. I don't think it's online. The ebook was on sale for $10 on Black Friday. So... you'd have to wait almost a year
4:16:50
aeth
The core language features of Common Lisp haven't been changed in over two decades. The language extensions are a mix between just as old as the language (things that didn't get in the standard, but could have) and fairly new things. They still move fairly slowly. But everything else is just like any other programming language, and it changes over time.
4:50:09
pierpa
The best reference manual for CL is CLTLII + checking CLHS to be sure particular things haven't changed.
6:04:52
asarch
One stupid question: I start sblc and then type a few expressions, is it possible to save them into a file?
6:19:12
myrkraverk
For example I have it configured with :history #p"~/.sbcl.history" ; in my .sbclrc
7:17:17
smokeink
Is there any easy/idiomatic way to find all symbols that are fbound to some function ?
7:29:19
loke
Does anyone know of a nice library that can parse infix expressions as strings and evaluate them?
7:32:12
myrkraverk
It's basically the tutorial in most compiler books, but I don't know about a library for it.
7:33:28
myrkraverk
As in, flag1 flag2 flag3 are meant to be mutually exclusive and so far, each is a simple parameter.
7:33:48
myrkraverk
For two, I can just use (and flag1 flag2) but it gets harder when there are 3 or more.
7:35:19
myrkraverk
Of course there's something simple to do it. I was thinking about rolling my own with LOOP.
7:35:56
loke
If you have lots of flags, LOOP witll be more efficient as it will allow you to exit early.