⋄ (¯1∘↑,⍨(⍕¯2+≢),⍨1∘↑)'internationalisation'

@TessellatingHeckler i18n

Feels like there should be a shorter approach to get the first and last letters around something

@TessellatingHeckler ⊃ is first, ⊃∘⌽ or ⊢/ is last.

@TessellatingHeckler Also you can remove ⍨s in ,⍨ by flipping the order: ⋄ (1∘↑,(⍕¯2+≢),¯1∘↑)'internationalisation'

@Bubbler i18n

@Bubbler well, that's a lot tidier ⋄ (⊃,(⍕¯2+≢),⊢/)'internationalisation'

mildly annoyed I didn't think of ⊃ or ⊢/

are you guys allowed to play golf during the lock-down :P

It's socially distanced golf :P

coolio

⋄ {2/'█ '[(⌽≠⍉)⍣3⊢8 8⍴11⎕DR⍵]}'Raghu'

@Razetime Illegal code

@Adám why's APL giving me an index error?

@Razetime Works only with ⎕IO←0

          ██
  ██        ██
██  ██    ██
      ████
      ████
    ██    ██  ██
  ██        ██
    ██

weeeee

CMC: Decode a name from APL avatar

    :55378158 Yours looks pretty cool
████  ██████  ██
      ██      ██
██    ██
██        ██████
██████        ██
        ██    ██
██      ██
██  ██████  ████

@Razetime {88⎕DR,((⌽≠⍉)⍣¯1)⍣3⊢'█ '⍳8 8⍴((128⍴1 0)/⊢)⍵} might do it, if you wait for Dyalog to implement (⌽≠⍉)⍣¯1 :P

can that be undone?

I don't think ⌽≠⍉ has a unique inverse even in booleans, because summing all mod-2 equations will give a null equation, which means the coefficient matrix is not invertible.

Does every bit pattern having 4-way rotational symmetry have a corresponding string input? It's an open question too.

interesting

I mean, my actual name and my udername both yied about the same output

maybe it needs to expand in size for the size of the name

that might have an inverse

Hey lyxal

and adám

@Razetime Hi. Well, there are only 16 bits in the final output, so there will be a lot of clashes.

so we need more bits

@Bubbler That should be brute-forceable given the size of the problem space

@Razetime 42 bits: ⋄ {' ░▒▓█'[2/1+(⌽,⊢)6 3⍴5⊥⍣¯1⊢⎕UCS⍵]}'Raghu'

@Adám Illegal code

⋄ {' ░▒▓█'[2/1+(⌽,⊢)6 3⍴(3⍴5)⊤⎕UCS⍵]}'Raghu'

@Adám Illegal code

@DyalogAPL ⋄ ⎕UCS 'abc' ?

@Adám 97 98 99

⋄ {(3⍴5)⊤⎕UCS⍵}'Raghu'

@Adám Illegal code

⋄2⊤7

@Adám 1

Huh. ⋄ {' ░▒▓█'[2/1+(⌽,⊢)6 3⍴(3⍴5)⊤⎕UCS ⍵]}'Raghu'

@Adám
██▓▓▓▓▓▓▓▓██
░░████████░░
    ████
▒▒▒▒▓▓▓▓▒▒▒▒
▒▒██▓▓▓▓██▒▒
██▓▓▓▓▓▓▓▓██

@Adám 12 of the help.dyalog.com links in the AplCart tsv use http instead of https, btw

(assuming it should all be https, probably way easier to do a search/replace than a PR)

@TessellatingHeckler Yeah, every once in a while I run that search-and-replace…

ah ok :)

@TessellatingHeckler Pushed.

@DyalogAPL I could actually use this as a profile picture

But i guess the simplicity of the earlier code gives cleaner patterns

What's the easiest way to make a function (f g h) so (f g h) x is just f g h x

obviously f g h doesn't work because that makes a train :/

@rak1507 f∘g∘h

Oh yeah

You can read f∘g as f with its right argument preprocessed by g

so that f∘g∘h is f with right argument preprocessed by g∘h which preprocesses the right argument with h then g and then applies f

In this case (⌈/≢¨⍤(⍕¨)) seems to work but multiple ∘s would need more brackets

((⌈/≢¨)⍕¨) also works

Given that both g and h have each, ⌈/≢∘⍕¨ should do

great thanks

Is there a better way to do ((⌈/≢¨)↑¨⊣)? Padding to the length of the longest string

@rak1507 Like this?

Monadic ↑ (Mix) converts a vector of strings into a char matrix, adding space paddings as necessary. Then you can ↓ (Split) back into vector of strings.

That is perfect haha, so simple

Just used the rank operator correctly, feel like an APL god

(-∘⌽⍳)⌽⍤0 1∘↑(⍕∘(⍳!⊣⍴-∘1)¨1+⍳) Any obvious errors in this?

What are you trying to achieve?

I see a Pascal's triangle using ⎕IO←0, except that it isn't quite center-aligned

Pascal's triangle

Yeah

Not quite sure how to properly align it

You need to take into account the lengths of the strings

yeah :/

@rak1507 Nice, though ⌽ actually has ⍤0 1 built in.

I have some code which apparently dyalog detests - if I send a strong interrupt via RIDE the it appears to break as it should, but it's then non-responsive, and looking at top dyalog is consuming 100% cpu. This happens both on my Mac and a R-Pi. Any reason why / trick to recover from teh strong interrupt not really interrupting?

Or is it kill -9 time?

@xpqz That doesn't seem right. Are you allowed to save the ws and email it to support explaining?

It's dead -- RIDE responds to key strokes but there is no interaction with the 'terp.

@xpqz I mean, if you can repro it (after killing it).

I can see if I can cut it down to something containable.

Thanks.

I don't doubt that it's a bug in my code -- most likely a non-terminating recursion or two.

Sure, but APL shouldn't become unresponsive after a hard interrupt.

Dyalog works really well on the Pi3, btw. Roughly 8-10 times slower than my mac, but it doesn't crash.

@Adám damn, thought I'd done something clever then :(

@Adám I've got it reproducible with a tiny data set -- I've saved the workspace.

@xpqz Great.

@dzaima I compiled your dialect of APL

APLP5 tetris works but it doesn't have a kill screen

and it shows this error

No library found for APL.types.Fun

It's probably a processing 3 problem

which version exactly should I use?

@Razetime i have no idea why that's a thing, but it is and it doesn't seem harmful

@Razetime it, like the language its written in, is not completed :)

oh lol

It has a working mandelbrot set tho

what is draw supposed to do?

@Razetime mostly a test for golfiness of code

(god the updated android firefox app makes using SE chat just painful)

oh lol

all websites have taken to pestering their users to download their app

Well, draw doesn't seem to display anything

@Razetime did you try clicking&dragging in the screen?

oh so that's what it's abotu

lmaooo

@Razetime so, if you're an apl type, you should be having fun only in the library

uh what's that mean?

I like the each left nd each right symbols

@Razetime that's what the error message is implying :D

@Razetime Lbrary found for APL.types.Boring?

@dzaima pushed fix by not using import, and properly referring to an inner class with the outer class

oh lol

I just need to pull it right

@dzaima voila

@Razetime yeah. but it's largely pointless anyways

My masterpiece in draw.apl

oh lol

@dzaima how do I get the pixels of an APLImg object?

oh lol

Maybe I should figure some stuff out and pull request

@Razetime img.pixels or img.mat (i really thought i didn't implement those lol)

cool

maybe I can port some of my older processing answers

lmao

@dzaima as an example, to draw 100 lines, there are 2 options - either a 100 4-item vectors, or a 100×4 matrix (generating which is often very annoying). the rewrite does the correct thing of allowing a 4-item vector of each separate coordinate, but for some reason there are 2 separate line drawing functions and coloring things is still a complete mess (if you want to draw many different color lines, you're stuck with a loop going through each)

@dzaima (the vector of vectors approach also allows easy constants across all points - e.g. g.ln 100 (8×⍳100) 400 400 draws lines from (100;y) to (400;400) for each y in 8×⍳100)

That's what I imagined

The kind of flexibility apl has is great for processing

Are there and good APL support extensions for VS Code?

Nothing is coming up in the search for some reason

@Razetime see last row here

huh. I guess I'll have to manually install it

Is there some magic variable Dyalog APL has access to, that allows me to refer to a directory where I want to store some workspaces/.dyalog files so that I can refer to the directory with an absolute path, but making use of this (system?) variable?

maybe something relative to Dyalog APL's installation folder?

@dzaima

@RGS wouldn't you mean refer by a relative path? ]cd?

@Razetime you can use integers for colors too (the in the regular processing way of mixed 0-255 gray and 0xAARRGGBB otherwise)

you can also use 3-char color codes (i.e. 'f0f' for purple)

oh interesting

which would be best here

3 char right?

@Razetime it'd also most definitely be better to have a loop instead (or at least assigning G.stroke and G.pt to temporary functions)

hm ok

I'm a bit tired now

I'll golf it soon

also z←'0'⋄f←'f' → z f←'0f'

i have 54

@dzaima it should be simpler with netpbm

@ngn but that's boring

@dzaima (could convert to a function for 46)

@dzaima though that has made me notice that for some reason pt (Processings POINTS shape) doesn't draw the points at 100% opacity :|

@dzaima creating an image (with P5.img which takes a matrix of colors as arg) is 31

tied jelly

@dzaima same color grid?

@Razetime i just copied Jelly's output as reference for that

@Adám 'The letters from A to Z' doesn't point to the correct tio link

on aplcart

For some reason it redirects here: tio.run/##SyzI0U2pTMzJT////1Hf1EdtExSAlLOvi4J6TrGCvvr//wA

Is there any reason the function for an identity matrix on aplcart is (,⍨⍴1,⍴∘0) and not (∘.∊⍨⍳)?

17-vs-18?

Ah probably

@rak1507 Yes:

      ]runtime -c ∘.=⍨⍳100 (,⍨⍴1,⍴∘0)100

  ∘.=⍨⍳100      → 2.5E¯6 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  (,⍨⍴1,⍴∘0)100 → 9.8E¯7 | -61% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

Wow that is quite a difference

@Razetime Thanks, fixed.

@Adám i would've expected ∘.=⍨⍳ to be an idiom, but i guess requiring the obvious faster version makes sense

(≢∘⍸(∘.≥⍨⍳)) Worlds dumbest nth triangular number generator

@dzaima Lots and lots of things could be idioms. I'm not sure why some things have been made idioms and some things not, but the idioms system isn't sustainable. We need Marshall's thunks instead.

@dzaima What would you ever do where generating identity matrices actually affects the total time?

@Marshall right, that too

not really a surprise

CMC fastest-code: identity array of given shape. Identical to {1=≢¨∪¨⍳⍵}

why are dfns so much faster

@Adám yaya

Are they? @ngn

      n←100⋄ cmpx '(,⍨⍴1,⍴∘0)n' '{⍵ ⍵⍴1,⍵⍴0}n'
  (,⍨⍴1,⍴∘0)n  → 1.2E¯6 |   0% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕
  {⍵ ⍵⍴1,⍵⍴0}n → 6.9E¯7 | -44% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

Interesting

isnt ⍵⍴1 enough

also interesting

@ngn Usually the issue is the time taken to form the derived functions. If you assign the two functions in advance, the train will probably be faster.

Still slower for me

@Marshall you're right

yay dzaima/APL beating Dyalog

@rak1507 Since you're using ¨, what I said doesn't apply. In your first example the functions are created once and applied 1000 times.

thats a cool factoid

@dzaima Is this because of JAva, or more efficient interpreter?

@dzaima try ngn/k :)

 \t:1000 =100
9

what about ngn/apl

@Razetime that's in javascript (i.e. very slow)

@dzaima on my crappy laptop, yes

@ngn slower than dzaima/APL?

@dzaima why "each"? what are you solving there?

let's throw dzaima/BQN in the mix

@ngn what we (or at least i) are testing is the overhead of function evaluation

Why does '{} 1' not work? does runtime need a value?

@dzaima and why not, C# dzaima/BQN too (which doesn't support compiling to native bytecode for obvious reasons)

@dzaima (for a fair comparison, regular dzaima/BQN)

@rak1507 Yes. It compares results too.

Wait BQN is incomplete right

@dzaima Hmm, if we're on the subject of I:

$ time echo "1000.i  f  1000oi f +b1*[*0.5" | ./IO > /dev/null

real    0m0.672s
user    0m0.640s
sys     0m0.030s

@Razetime the language is still being developed, yeah

So 0.67ms per execution. Not bad!

looks pretty linear

@rak1507 isn't {⍵} an idiom?

yes

i'm still confused about what we are comparing here

@Marshall Wait, that version isn't compiling +b1*[*0.5 as a unit.

@ngn the performance of calculating the n'th triangular number by the x×(1+x)÷2 formula for each of the first 1000 natural numbers individually, nothing more :p

$ time echo "1000.i  f  1000oi f.o +b1*[*0.5" | ./IO > /dev/null

real    0m0.321s
user    0m0.300s
sys     0m0.024s

I is probably the only one that gets slower when you vectorize the arithmetic.

lool

@dzaima dzaima/apl is just the upper measurement, right? (the lower looks like dyalog)

@Marshall No, that's the wrong result. And is wasting a lot of memory management time because it keeps old results around. Last one:

$ time echo "1000.i  f  1000oi f.o +b1*[*0.5.k o +.r" | ./IO > /dev/null

real    0m0.474s
user    0m0.464s
sys     0m0.013s

(I is maybe not the most usable language out there.)

@ngn in this message? yeah, dzaima/APL uses > for repl and dyalog uses 6 spaces. some separation of different parts is still a todo for paste

@dzaima so, you're claiming 0.044ms while dyalog is 0.130ms, ngn/k is 0.127ms, and (i measured) shakti is 0.115ms

that's 3x faster if it's true

@ngn apparently. makes some sense too for it to be fast in java too since 90% of most java programs are virtual method calls :D

@ngn It's a bytecode transpiler versus three direct interpreters, right? Doesn't surprise me at all.

@Marshall dzaima/APL doesn't have any bytecode. but trains are pretty much pure recursive tree walkers

@Marshall (ngn/k does use bytecode though)

@dzaima yes, but no jit

@dzaima Oh, didn't realize that was the dzaima/APL time. If it's like dzaima/BQN it takes a lot of iterations to get the time down, so it might be doing some sort of dynamic inlining.

ok, 3x is an extraordinary claim. i'll have a closer look later. i must disappear now.

@Marshall with 1e5 iterations, it's stil 0.047ms straight from starting up. 1e3 is 0.15ms, and 10 is an awful 1.7ms

@Marshall i would not at all be surprised it's doing some magic too

but i also gotta go about now

@ngn my .044ms is still ~160 processor cycles for what's essentially 5×virtual calls + 4×object allocations + 3×pervasion logic (which is probably the most expensive thing, and also includes an interface method call which might not be as simple as a virtual call) + the actual calculations

@Marshall there is bound to be some iterations required as java requires some big number of times of execution to actually optimize/compile functions

@dzaima (there are of course a couple more things sprinkled around, such as the ∘ logic and forks checking for f g h vs A g h, but those would still be roughly free)

hm, im not using bqn vim keyb correctly.

i use pathogen, copied the directory into my pathogen path

opened foo.bqn set syntax

@dzaima BQNs pervasion logic is much worse (here from L12 returning at L19 (virtual call) → L65 returning at L72 (virtual call) → this for multiplication) yet it's 2x faster than dzaima/APL!

@Adám free to hop on a call (of the platform of your choice) whenever your free

@code_report Give me 10-15 mins.

:thumbs_up

@code_report OK, can you come to my Zoom room? ID is the result of -\758,435,8447 in APL.

@cannadayr The editor configs are designed to work when you edit a *.bqn file. For kakoune you could also manually run :set-option buffer filetype bqn; for vim I use :set keymap=bqn to get the keyboard on other filetypes.

I don't know how pathogen works so there could be an issue there; I just have all the files in editors/vim/ symlinked to the matching locations in ~/.vim/.

@cannadayr g:bqn_prefix_key is a variable you can define to override the default. the plugin doesn't define it.

@Marshall cool cool I got it working. in vim, do you know if theres a way to "lock" the apl keyboard, instead of using a prefix key?

was gonna go for a walk to take adv of this weather but i might hack together a xkeyb file later if there isn't one yet

@cannadayr this?

@cannadayr Probably not with the current system, but I expect you could tweak keymap/bqn.vim to do it somehow.

@cannadayr why would anybody do that?

@ngn use an xkeyb file or lock/unlock the prefix key?

@dzaima +1 +1 +1

@cannadayr lock the prefix key

@dzaima I can put this (adjusted to match the editor maps) in the repository. What should the filename be?

@ngn because of the way vim prefix keys work it only will modify the immediately successive character

what i normally do in X, use use the 4th+5th levels on my keyboard

@Marshall just bqn probably makes sense with a comment somewhere, as the other keymaps are plainly named too (e.g. apl is apl)

the file name is the primary way of addressing the keymap with setxkbmap

@cannadayr so you if press the "lock" key once and then type abc, ⍉⌊↓ will appear on screen?

@ngn i use right alt to go to 4th/5th levels

@cannadayr ah, so you mean a modifier key, i.e. it types alternative characters only while (not: after) alt is pressed

@ngn yes but im not sure if vim can do that, i think it just uses prefixes, but i prefer the xkeyb change

@cannadayr i think due to the way terminals work, <alt-something> comes in as <esc>something

@cannadayr Here's my keymap for XKB.

Untested, so please let me know if there are errors.

@Marshall (might want to add ‿ beside space as i forgot to)

@Marshall thx thx

@dzaima Just in the comments, right? It's there for <SPCE>.

@Marshall yep

@Marshall i merged this into my keymap file and it looks like it works perfectly ty. is there also a monospace font you use? (to get around dbl-width characters not being displayed properly in terminal)

@cannadayr I'm still using this one based on DejaVu Sans Mono but it's missing 𝕣𝕤𝕊, so BQN386 is probably better.

@Marshall (fwiw making a good BQN/APL DejaVu font is something i'd like to do at some point; i prefer DejaVu for text (alphanumerics) but prefer APL385 for the actual fancy characters)

@dzaima (the latter note being mostly unrelated, if there was any confusion. it'd still be in the DejaVu style, just more organized/complete)

@cannadayr whatever the problem, fc-cache -fr usually solves it :)

@cannadayr i was annoyed by that too, no clue what it uses to decide what's an acceptable terminal font

@ngn gnome-terminal shows a tiny subset of all fonts

it's probably expecting some marker of monospacedness about whose existence i know nothing about

@dzaima it shows up in gnome tweak tool

@dzaima but not in my gnome-terminal profile

@cannadayr exactly

much cleaner now thank you every1 for the help

gonna take adv of this weather and go for a walk but ill probably be back on later

btw @dzaima using your REPL setup was very easy on debian sid

@dzaima have you tried writing it as a dfn? :)

@ngn yeah, and i know it's absoutely awful. :P

@dzaima i don't know what the jvm is doing with your trains but it's amazing

it's a factor of ~2x on my laptop (my cpu might be missing some instructions)

i thought jit might be vectorizing the computation but it's not - it's faster without ¨

@Razetime 22 bytes: 'P3'newline3 3 1newline(⍳3)∘∊¨⍳3 3

@TessellatingHeckler yes but the second is probably described incorrectly (if "y-axis" means the imaginary part)