the k tree

I don't think arthur has ever bothered with full lexical closure. I have it in oK because it was barely any extra work when I'm implementing it in a managed language

nathan rogers

I've tested this with f at the root scope, and it works just fine

you're the OK implementer?

JohnE

that's me

nathan rogers

noice

is there a way to interact with node libraries from OK?

can I start a server, make a request, write / read from various IO?

JohnE

not out of the box. The oK interpreter itself has no IO at all. It would be possible to create a "frontend" for it which exposed features like that.

nathan rogers

2:46 AM

what about the inverse?

JohnE

invoking oK from node? That's essentially what the built in CLI repl provides

nathan rogers

so theoretically, I could do my business in node, and make calls to ok ?

set up sockets, read and pipe the data to k functions?

JohnE

if you really, really wanted to run your code on the second-slowest k interpreter in the world

rak1507

what's the slowest

nathan rogers

is slow for k slow for standard JS?

what are your benchmarks for OK vs JS

JohnE

2:50 AM

the slowest is probably internet-janitor.itch.io/applejak

assuming you consider it a K interpreter at all

rak1507

oh cool

nathan rogers

@JohnE what I'm interested in is actual work. which means I need data, and that data has to come from somewhere. Dyalog's IO is usable if inelegant, I can at least write small wrappers around my problems, but its liscenced so I'm probably not going to use it for real work. Same with any production K, so that leaves the F/OSS and other OSS implementations, I could use J perhaps

but compared to K or APL J is... inelegant?

so, which of the APLs provides what Dyalog does? .... none
so, what open source K's provide FFI for C, or some other means of external communication? .... mmmmmmmmmmmmmm, I'm trying to find out

JohnE

oK is implemented in JS and could be expected to run at least hundreds to thousands of times slower that whatever JS engine is being used to run it. Amazingly, even so, it is fast enough to run various interesting graphical demos and so forth. I consider oK a good tool for learning, and it is free software, but I wouldn't consider it robust or fast enough for any meaningful industrial application.

nathan rogers

it seems like KONA at least provides reasonable file IO, some C capabilities if not easy to find (in the user group)

but KONA also seems to be "missing?" primitives I'd expect from something like APL

like, is there even a sliding window reductioN?

I can't find it if its there

the documentation is also extremely unclear as to how to do certain things

but its looking like KONA is my only solution, or contributing to NGN K

JohnE

imo the most productive things to come out of oK are 1) people seem to like the documentation, 2) iKe proved that K can be applied to graphics programming, and 3) oK has been used by various people as a reference when writing better open-source K implementations like ngn/k

nathan rogers

2:57 AM

I've seen ngn's table of K's

are there others worth looking at?

JohnE

I believe ngn/k is presently the most actively maintained and performant open source K

nathan rogers

I reached out to Kx some 3 years ago to see if an older veresion could be procured for research, and the answer was emphatically NO

@JohnE "performant" on golfing problems that don't interact with external data sources? Or are there actual application benchmarks?

is KONA not performant?

are there benchmarks comparing KONA to ngn?

sorry to be hammering you... just this information feels like pulling teeth to find

I get that K users tend to be terse, implying abruptness, and the documentation (sometimes helpfully) reflects that, but getting fully formulated english sentences expressing fortitude of thought past the "abruptness" of the immediate point is difficult to find

And having this information up front will make it much easier to make a decision to jump ship to one impl or another for some language. What I don't like is constantly having to learn a new impl or language just to realize it doesn't do the thing I want/need to do

JohnE

kona and ngn/k are two different dialects of the language. Kona is based on k2/3, while oK and ngn/k are based on k5/6. This is why their syntax varies in small ways and their collection of primitives and overloads varies. Kx's KDB+ is based on k4, which is yet another dialect. It takes a considerable amount of work to adapt benchmarks across these linguistic gulfs, and the only people working on open-source K stuff are volunteers and hobbyists

nathan rogers

so you're saying "get some quotes".... but even then K4 is proprietary from what I've read. you can liscence Q but not K4

and shakti doesn't feel very array oriented

but then I haven't played with it for more than a few hours trying to just figure out the differences from what little I am able to do in K at present

but from what I recall i was able to pull data in from a mongo database at work in short order using Q

maybe I should just look at their actual pricing... but I can only imagine its meant for people making actual money, not for people wanting to build personal applications

or maybe I should just jump on the bandwagon of "everyone makes their own language" X(

JohnE

3:14 AM

Kx has never made a meaningful investment in open-source and hobbyist applications, which I think is slowly dooming the language. Even a fantastic tool is at best a questionable engineering choice when compared to the dozens of high-quality free and open source languages it competes against.

Last I checked you can still get free evaluation copies of kdb+; you can escape from the q prompt to get to k4, but it's kind of annoying to use, since it kicks you back into q whenever you encounter an error.

nathan rogers

There's a fortran on every platform for every purpose... CL is a fully competent and in some ways still cutting edge language, and yet I can't find an APL-family language equally capable of catering to modern programming

J is probably the best option

but I just literally can't even with J

I wrote a few little command line scripts with it a good few years back

but I just don't have any desire to write J

JohnE

I don't find J very pleasant to read or write, but it does have a fleshed out ecosystem of libraries, bindings, and documentation

nathan rogers

K at least achieves terseness and APL is elegant. J is... bats

JohnE

doesn't hurt that the official J is open source, and people working on it don't need to be afraid of getting sued

nathan rogers

yeah

rak1507

3:17 AM

J has all the right ideas in all the wrong ways

3

nathan rogers

This is depressing

JohnE

it's kinda like F#: a kitchen-sink language full of ideas and experiments, some of which turn out to be footguns

nathan rogers

there is no language I'd rather use than APL-family languages, but I can't find one worth using

outside of toy/golf situations

if I was running a business and had income from my applications I'd license Dyalog

JohnE

sometimes I think about writing another K and getting the ecosystem-facing stuff right, like standard libraries and general purpose IO interfaces, but I just haven't had the energy for it at all these days

I've been dealing with some stuff

nathan rogers

I could go off and do language research, and study other implementations and try to build my own, but that's delaying all the other stuff I want to be doing right now

I've done BEEN wanting to do since I started this dang-ol journey 4 years ago

rak1507

3:21 AM

I've thought about trying to write a K inspired language that focused heavily on integration with existing libraries in python and other languages so it could act as a drop in replacement for current code, but with my lack of knowledge that's impossible

nathan rogers

@rak1507 there's py integration and a swath of other language interop with shakti

but $$

rak1507

yeah, exactly

$$$$

nathan rogers

the model needed to change back when they were implementing J

rak1507

very few people will shell out to replace perfectly working python code with maybe slightly faster k stuff

nathan rogers

but then they made J

JohnE

3:23 AM

the problem with piggybacking on another language is you pull in all that language's baggage and dependencies, and you don't really benefit your own ecosystem by focusing on passing the buck to e.g. python

nathan rogers

I think the Clojure model has it right, you implement a LISP inside an environment that has everything you need to do business, and you keep the core of the language tightly controlled and restricted from adding things, relying more on providing library extensions to the core language

rak1507

if it was moderately painless to transition, more people would do it, relying on something like pypi or npm is a terrible idea but it should at least be possible to use stuff in the short term

nathan rogers

"short term"

rak1507

well yeah realistically people would probably end up using that forever and making legacy code but I don't like to think about that

nathan rogers

If I know the software industry, people find the first way that works and never change their approach unless management forces them

JohnE

3:25 AM

think about how many languages have a crude wrapper for SDL2 or whatever and the documentation is basically "lol just read the C docs, hope you're intimately familiar with C as well as how our FFI system mangles the signatures of everything"

nathan rogers

Julia has a useable C FFI

chat.stackexchange.com/transcript/message/57280860#57280860 This though, an implementation of an array language such that I can straight call external methods as if they were native function calls is the right way to handle a language implementation in current year

the .NET "bridge" in Dyalog is nice.... to have, if you're busy wanting to turn your APL into OOPL

but clojure doesn't force any kind of OOP, you import java libs the same way you import clojure libs, and you call them like clojure functions

JohnE

the traditional approach for C interop from K is "use k.h, write some C functions that take and produce K objects, deal with whatever marshaling and memory management needs to happen on your own, and then you can load those functions from a DLL to get K-callable functions." ie you're on your own and constantly wishing to know more about the implementation of a black box.

nathan rogers

I really think an array notation is best embedded into the syntax of another language, rather than a language unto itself

JohnE

if you want things to be automatic and magical you have a serious round-peg-square-hole situation adapting the builtin types of a dynamic managed language to the type system of C

ah, the classic "what if k were like regexes" gambit

nathan rogers

?

JohnE

3:32 AM

a DSL called as a library from some general-purpose "real" language

nathan rogers

I think more what I'm referring to is if mainstream languages would bother to learn the lessons of array languages and actually provide array notation, like how they all seemed to learn the lessons of LISP and FP between 07 and 2014

What I'm really after is a universal programming notation, rather than some kind of dogmatic stubbornness about what constitutes a "real language" or an "array language"

I wish the modern APL community was as much invested in making APL truly general purpose for todays environment, rather than spending so much time worrying about different kinds of functional composition there are

if APL was only DFNS, but it had scripting, it'd be infinitely more viable of a language

if they could stop spending so much time trying to decide on glyphs for this or that kind of function composition and implement real closures, real first class functions, and maybe NORMALISING their 80 years of bloat into a few interfaces, the language would be truly modern

and how is there still a debate around whether dicts are useful? Why is there no effort around a notation for trees or dictionaries? That seems like a more useful effort than a new function composition glyph that'll save 2 or 3 bytes on a golfing problem

¨ might be faux pas but it gets a lot of work done

chrispsn

12:41 PM

@nathanrogers on that - most of my interest in bqn is seeing whether it's a better set of primitives than k

With the metric being the minimum of the product of character set count and average golf length across a standard set of problems

Double points if it results in fast(er) code

rak1507

does k have any way to do ⍸⍣¯1 in APL

chrispsn

@rak1507 do you have an example?

rak1507

f 1 3 5 -> 0 1 0 1 0 1, f 2 2 3 3 3 6 -> 0 0 2 3 0 0 1

it's the inverse of &

chrispsn

Oh right

Something like @[&1+|/x;x;1+]

Nothing built-in

Might need an 'over' with the first arg to ^ as the seed - can't test atm

rak1507

1:01 PM

thanks

ngn

4:34 PM

@nathanrogers yes, o is the function you're in, e.g. {$[x<2;1;x*o[x-1]]} /factorial

@JohnE it doesn't, because the original k doesn't. but i wish all ks had full lexical scope.

@nathanrogers "valence" = "arity" = sometimes "rank (of a function)"

@nathanrogers i showed you how to do i/o in k a couple of days ago. remember 0: ?

pretty sure apls other than dyalog had i/o too. i remember reading microapl's docs and the i/o squiggles were something like [←] and [→] (quad arrow)

⍇ ⍈

@nathanrogers do you have experience with ffi in any apl-like language? dyalog has ⎕na, old k-s had 2:. what do you think a good ffi should look like?

ngn/k doesn't use the standard c library, so the biggest challenge for me as implementer is writing dynamic loading from scratch

@nathanrogers what

nathan rogers

5:15 PM

@ngn File IO yes

@ngn But if you continue reading I mentioned that it is licensed and so I must pay to use it for anything business related. Another thing I mentioned was that the interface for i/o needs to be normalized because so much of it is scattered to the wind with different libraries and difficult to find documentation

@ngn I built a GLFW/GLEW parser that reads in the headers and procedurally generates ⎕NA definitions

@ngn What? We were talking yesterday about how many of the things I've come to expect from other K's, APLs and even J that have no primitives. sure they can be defined, but much of how a language enables you to think is predicated upon what intuitions you can draw from language "primitives", whether those primitives are functions, arrays, lists, macros, blocks, or what-have-you.

ngn

@nathanrogers and "shakti doesn't feel very array oriented"??

rak1507

have no primitives? huh?

nathan rogers

what?

are we doing this again?

@rak1507 that isn't what I said and you know it

rak1507

'We were talking yesterday about how many of the things I've come to expect from other K's, APLs and even J that have no primitives.'

ngn

14 hours ago, by nathan rogers

and shakti doesn't feel very array oriented

rak1507

5:21 PM

that have no primitives

what does that mean

nathan rogers

@ngn because like I said I only had a short time to play with it and I bumped into several mental bumps looking for primitives. The primitives that existmake it feel like the functions want you working with lists, and nested lists or list of lists

@rak1507 oh I see, that was a mental slur... I meant to say its missing primitives that I've come to expect, that's all

rak1507

oh ok

nathan rogers

@ngn but like I said, its just a first impression

I'm not a K wizard, so all that should be taken as though from the K-scrub that I am

I spent a good portion of the day tinkering with kona/ngn/shakti

so I was just making some first impression comparisons between the 3

ngn

i think if it can do 1 2+3, it's array-oriented

nathan rogers

I think the degree to which you're encouraged to think in arrays of any dimension is variable, and thus lies on a specturm

I'm probably way off the mark, but my first impression was that shakti is less array oriented than other K's/APL/J

ngn

5:32 PM

@nathanrogers i agree "it's a spectrum" but disagree about k9 being less array-oriented than its brethren

@nathanrogers any specific examples?

nathan rogers

where's sliding window?

no rotate?

# no longer reshapes, how do I specify matrices?

ngn

@nathanrogers sliding window and rotate are just specific patterns of indexing

nathan rogers

Right

ngn

@nathanrogers shape# can be replaced with a sequence of cuts (n^)

nathan rogers

" much of how a language enables you to think is predicated upon what intuitions you can draw from language "primitives", whether those primitives are functions, arrays, lists, macros, blocks, or what-have-you."

I can specify a function that defines indices of a "cycle" function, like 20⍴⍳4 or 20#!4 as well

but those are primitives because they're things you want to do all the time

I think an array oriented language, contrasted with perhaps a 1d list oriented language with nice formatting, is inordinately concerned with the "shape" of its data

and so being able to shape on the fly seems pretty important

but maybe that's a mental crutch

i don't know, its just my first impression

ngn

5:45 PM

in this particular case k9 trades ⍴ for ∩ (in apl-speak)

idk, you may be right that reshape is more important

on the other hand, cut (n^) already deals with the common uses of reshape

there are many subtle trade-offs to be made in a limited set of primitives, such as k's

in apl or j they would just add a new one and get a bit fatter :)

nathan rogers

nice graph

top kek

ngn

7:36 PM

@Wezl x, y, and z are implicit parameter names, like ⍺ ⍵ in apl. but unlike apl's always-ambivalent dfns, k functions have fixed valence. mentioning z makes the function triadic. mentioning y (but not z) makes it dyadic.

nathan rogers

@ngn what about more than 3 args?

Wezl

mhm

ngn

@nathanrogers {[a;b;c;d;e;f;g;h] ..}

at most 8

nathan rogers

so you have to specify the param list?

it isn't just implied to be x?

ngn

@nathanrogers for ≤3 args, you can use x y z without an explicit parameter list

for >3 you must use a parameter list

i don't think i ever needed more than 5

nathan rogers

7:39 PM

good 2 know

nathan rogers

7:53 PM

@ngn rest?

ngn

@nathanrogers you're asking if k supports rest (that json-over-http thing)?

nathan rogers

data over http thing yeh. you can send/receive csv or text or xml

coltim

@nathanrogers you can always provide an explicit argument list, e.g. {[a;b]a+b}, but as @ngn said it's somewhat optional

ngn

@nathanrogers i haven't even integrated an http server yet

nathan rogers

what needs to happen for that?

http server

coltim

7:57 PM

@nathanrogers you could destructure a list or dict input, something like {(a;b;c):x;a+b*c} or {x[`a]+x[`b]*x`c} but it's probably less idiomatic than explicit args

ngn

@nathanrogers what needs to happen is that i stop procrastinating and actually start coding :)

nathan rogers

:O

well at least you're professional

I'm just an amcrastinator myself

ngn

writing an http server isn't hard

writing a fast one is hard

i think geo had published one somewhere..

ah, there. but no licence :(

ngn

8:21 PM

@nathanrogers do you think an http server should be part of the language interpreter or something external?

nathan rogers

like a library you mean?

ngn

@nathanrogers more like a separate executable that spawns k when a request comes in, and redirects the request/response to k's stdin/stdout

nathan rogers

I have no problem with needing to import server related functions. It'd be nice to have on hand
1. load k
2. start server

but
1. load k
2. import server
3. start server

isn't an imposition so long as its mentioned somewhere

@ngn any response to that question yesterday regarding \ ?

ngn

@nathanrogers what question? remind me

nathan rogers

\ exits your k, but I type that by mistake quite a bit, and especially when I forget I'm in your k, not kona, where kona \ returns a list of all potential hellp commands

with a short descriptor

also how do I check for definitions?

ngn

8:25 PM

@nathanrogers i have no strong opinion on that. i just did what the original k did (at the time).

nathan rogers

I find the meta-help list to be indespensible

ngn

@nathanrogers would it be better if it was \\ ?

nathan rogers

I used that kona help menu to learn most everything

\\ is how to exit k4, shakti and kona

idk what k6 is/was

ngn

@nathanrogers ok, so be it, i'll change it to \\

nathan rogers

what I and... somebody, I think rak was mentioning the other day about just a bit more information than just the names is useful

kona's menu is super helpful

ngn

8:28 PM

@nathanrogers there are many things that would be useful, but only a finite amount of (unprocrastinated) time. maybe you can help?

nathan rogers

sure

how can I do that?

if I'm only an amcrastinator

rak1507

whats an amcrastinator

ngn

@nathanrogers tell me what that information should be

nathan rogers

  \
Backslash Commands:
\0        datatypes help
\+        verb help
\'        adverb help
\:        I/O verb help
\_        reserved word help
\.        assignment/amend, function, control flow help
\b [s|t]  show/set break mode (stop|trace|none)
\d [d|^]  change k directory (^=previous)
\e [n]    show/set error flag (0=off,1=on,2=exit)
\l f      load script f or f.k
\p [n]    show/set print precision (0=full)
\r [s]    show/set random seed (0=random)
\s f      step script f or f.k
\t [n]    show/set timer interval in msec (0=disable)

this is the meta help table

Verbs +-*%|&^!<>=~@?_,#$.:
Verbs work on all sensible types (ints, floats, strings, symbols, lists)
Dyadic or monadic is determined from context, default is dyadic
Add : after a verb to force the monadic form, + is plus, +: is flip
+ monadic flip. transpose a matrix (a depth-2 list)
+ dyadic plus. add numbers together
- monadic negate. invert sign
- dyadic minus. subtraction
* monadic first. first element from the list
* dyadic times. multiply two numbers
% monadic reciprocal. 1 over x

but I was thinking instead of the description perhaps just printing your unit test

demonstrating how the verb is used

something I can type in directly to see a result

like +/+2':!10 for a 2 chunk reduction because then I can take each part and play with it

if you have a table like
verb arity names unit-test(or some kind of example)
verb arity names unit-test
...

that'd be nifty

ngn

@nathanrogers it might be better to extract tests from the documentation, not the other way round. the existing unit tests were not meant for explaining what verbs do.

nathan rogers

8:38 PM

ah

ngn

they were meant to catch my own silly mistakes

nathan rogers

makes sense

unit tests often test fringe cases

ngn

if exit is \\, what should \ do? print help?

nathan rogers

that's the meta help menu

it tells me what different \'s there here

9 mins ago, by nathan rogers

  \
Backslash Commands:
\0        datatypes help
\+        verb help
\'        adverb help
\:        I/O verb help
\_        reserved word help
\.        assignment/amend, function, control flow help
\b [s|t]  show/set break mode (stop|trace|none)
\d [d|^]  change k directory (^=previous)
\e [n]    show/set error flag (0=off,1=on,2=exit)
\l f      load script f or f.k
\p [n]    show/set print precision (0=full)
\r [s]    show/set random seed (0=random)
\s f      step script f or f.k
\t [n]    show/set timer interval in msec (0=disable)

rak1507

does ngn/k support basic socket connections atm?

nathan rogers

8:42 PM

like if I want to check the workspace I use \w (which isn't here for some reason?)

but \w shows me what names are defined

so it'd be nice to know how to do things like check my workspaces/namespace/definitions

because I don't know how to do that

ngn

@rak1507 only client sockets

coltim

@ngn this whole time I could have just used "\"?? ha

ngn

@coltim what did you use?

coltim

\\

re docs: it would be good to have something available outside of the tool as well (a la the oK Manual)

ngn

@coltim i'm not articulate enough to write that much english prose :)

dzaima

8:51 PM

@ngn you could fork the oK manual and just change things that need changing

ngn

@dzaima hm.. that might work

nathan rogers

@ngn I don't think all that much is required. Shakti has a really nice minimal version

estradajke.github.io/k9-simples/k9

coltim

@nathanrogers that is definitely a valuable resource but at least by k standards I dunno if it's "minimal" =P

nathan rogers

@ngn code.jsoftware.com/wiki/Vocabulary/dot#dyadic

this isn't a lot of write-up

@coltim I mean it is minimal in terms of examples and explanation. Which tends to be hard to find unless your particular problem has already been solved

This is something I'd be happy to contribute to, I'm much more of a writer, but I don't have the domain knowledge

dzaima

@dzaima (even just cloning and replacing incorrect stuff with "todo" and opening contributions (^) would probably work well enough)

nathan rogers

8:58 PM

@dzaima Yeah, I'd even just say create some kind of html template, and write some K files, which you read to slot in the text/code into your html template

ngn

ok, this evening i'll try to think of short examples for each primitive and describe it in my own words. then @nathanrogers or anyone who wants to improve it is welcome to do so

but i can't really multitask, so i'll log out from chat

rak1507

alright cool

nathan rogers

@ngn thanks. and I'm happy to do what I can. I just don't know how much help I'll be on the implementation/design/education front

nathan rogers

9:38 PM

any non-regular syntax in K?

I don't suppose there are trains and such

coltim

@nathanrogers what do you mean by "non-regular"? $[cond;{true};{false}] has been described as a "special form" before (since it doesn't evaluate the {true} or {false} parts immediately) but I dunno if that's what you're referring to

nathan rogers

I mean what I implied, trains? syntax sugar? etc?

dzaima

@nathanrogers closest you'll get to syntax sugar would be assignment, which is syntax sugar for amend or something. Otherwise there's mostly only one way to do things in k

nathan rogers

That's good

uniformity is nice

ngn

11:06 PM

this is exhausting and it seems i haven't even described half of the cases yet

dzaima.github.io/paste/…

5

@nathanrogers as i mentioned in "the apl orchard", trains in k work as you intuitively assumed apl trains to work (but was wrong) before the odd-even rule was explained to you

1+*2% ←→ {1+*2%x}

there's no odd-even rule here. the last carriage of the train is applied to x (or to both x and y if it's dyadic), and all other carriages are applied on top of the result, following the usual rules for right-to-left expression evaluation.

compare that with apl trains, in which even-numbered carriages are applied to the argument(s) and odd-numbered are applied between the results (odd/even up to ⎕io)

nathan rogers

11:27 PM

@ngn bruh, that's excellent

Thanks←⌊/⍬

ty:>/()

rak1507

@ngn that's really great, definitely worth doing

nathan rogers

@ngn star this @rak1507

Transcript for