The APL Orchard

Adám

12:33

→ 1 time zone confusion moved to Trash

Adám

13:00

@RubenVerg I believe (∨/∊⍤(⍵⍷¨⍨⊢))¨ can be written 1∊¨⍷¨∘⍵¨

And (e⍉)¨(e←{⍵/⍨∨⌿0≠⍵})¨ as (e⍤⍉e←{⍵/⍨∨⌿0≠⍵})¨

And {4 4⍴⍵⊤⍨16/2}¨⍳2*16 as 4 4∘⍴¨,⌿2⊥⍣¯1⍳2*16 if you use ⎕IO←0

RubenVerg

all these are intuitively correct, I just have to wait five seconds to test it every time :)

@Adám no this one doesn't work

indices returns an empty vector

@Adám why do I need 0 io here?

Adám

Because otherwise we generate from 1 to 2*16 which takes 17 bits to represent. Your 16/2 chops the top bit.

RubenVerg

actually what's that code doing in the first place, what does ,⌿ do on a list of scalars

@Adám wouldn't your 4 4∘⍴ also chop the top bit?

oh no it chops the bottom

Adám

@RubenVerg It isn't a list of scalars, but a matrix with one number represented in each column, so ,⌿ catenates each column to a vector, giving a vector of vectors.

RubenVerg

oh sorry I thought you had an Each between inverse decode and iota

Adám

13:11

That wouldn't work, as it'd strip leading 0s.

RubenVerg

I think I don't really care if the 4 4⍴0 matrix doesn't appear, but I might care about another appearing twice

@Adám yeah that was why i was confused

Adám

@RubenVerg Oh, missing an enlist: 1∊∘∊¨⍷¨∘⍵¨ or 1∊¨∊¨⍷¨∘⍵¨

Three different epsilon eaches :-)

RubenVerg

thanks for the help, dropped 7 bytes

just need to run all the possibilities again because the order of things has changed

Adám

If I'm not mistaken, {⍵/⍨×+⌿⍵} does the same as {⍵/⍨∨⌿0≠⍵} since you stay with non-negative integers.

RubenVerg

ah yea this is one of the things aplcart taught me by heart and then i never questioned

Adám

13:39

@RubenVerg afaict, you only ever use the argument once, in ⍷¨∘⍵ which means you could convert to a tradfn body, replacing ⍵ with ⎕ and saving the outer braces.

rabbitgrowth

14:00

Trying to simplify {({(⍳∘(⌈/⍵))⍵}|+\(⍵-(+/÷≢)⍵))⌷⍺} from youtu.be/SlA_T91gS6s?t=3003 - is {⍺⌷⍨⊃⍒|+\(⊢-+/÷≢)⍵} the best you can do?

That ⍒ seems really inefficient though - should probably do {⍵⍳⌈/⍵} instead of ⊃⍒?

Oh hey look, it's a Y f g Y pattern!

Adám

Or (⊢⍳⌈/) but yes. We did have ⊃⍒ and ⊃⍋ optimised, but had to back that out because we didn't trust its quality.

@rabbitgrowth Yes, but an operator couldn't help you there because ⊢⍳Ø⌈/ would bind as ⊢(⍳Ø⌈)/

rabbitgrowth

Why would you need the ⊢ there?

Not sure how I missed ⊢⍳⌈/, kept on trying to do it as ⍳∘(⌈/)⍨

@Adám Interesting, I thought it'd just be like s/⊃⍋/{⍵⍳⌊/⍵}/ like RubenVerg said

Adám

Yes, but iiuc, it was one of a very large number of optimisations done by a single developer, several of which turned out to have faults. So we ripped out all this person's optimisations, adding them one-by-one, carefully.

rabbitgrowth

Ah I see

Better safe than sorry

Adám

14:16

As an anecdote, this explains our odd version numbering: 17.1, 18.0, 18.2. The 18.2 is actually a 17.2 because 18.0 was abandoned due to those optimisations being so widespread.

RubenVerg

@Adám this sounds like a variation of the classic cheater dilemma - if you find evidence that some person"s feats were cheated, do you claim they're all cheated and then analyze each to deicide if they're genuine or do you keep them and analyze each for cheating?

rabbitgrowth

@rabbitgrowth E488: Trailing characters: ⍵}/ :(

This optimization stuff is really interesting. Do you actually substitute APL code with faster APL code for the optimizations? I thought they'd be done in C or something

Adám

I think every instance is in C. It could be done using APL, though.

Btw, you can't just substitute ⊃⍋ with {⍵⍳⌊/⍵} as the latter ignores a left argument and can be an operand. ⊃⍋ with a number on the left picks the nth first, but cannot be used with an operator without composition/training.

RubenVerg

@Adám that evals it only once? I can never remember the side effect rules with composition

Adám

Yes, the ⎕ is evaluated once when first met.

RubenVerg

14:29

does (quad plus right) each arr call quad multiple times?

Adám

No, it gets input and then builds the train, which is then applied.

rabbitgrowth

@Adám Yeah, surely you can't just literally do a search and replace like that

Adám

Oh right, {⍵⍳⌊/⍵} has a very limited domain (reals), whereas ⊃⍋ works on almost every type of data.

RubenVerg

@Adám surely ({quad}plus right) each are asks multiple times?

Adám

Yes, that would.

RubenVerg

14:39

and that's because a lone quad is parsed as (A g h)?

Adám

Not necessarily:

      (⎕,⊢)¨1 2 3
⎕:
      -
┌────┬────┬────┐
│¯1 1│¯2 2│¯3 3│
└────┴────┴────┘

RubenVerg

oh I thought quad took only nouns

parsing modern apl looks so hard

Adám

Basically impossible:

     ⌽⎕'abc' 'def'
⎕:
      'xyz'
┌───┬───┬───┐
│def│abc│xyz│
└───┴───┴───┘
     ⌽⎕'abc' 'def'
⎕:
      ⊃
cba
     ⌽⎕'abc' 'def'
⎕:
      ¨
┌───┬───┐
│cba│fed│
└───┴───┘

RubenVerg

is there a decent way to force quad to be a noun/verb/modifier?

Adám

No, but you really shouldn't use ⎕ in production. Use ⍞ instead.

RubenVerg

14:47

decode jot (can't remember the name) only returns nouns, right?

oh I was also thinking about something, why was Circular chosen to be a dyad instead of an adverb that returns a monad?

Adám

@RubenVerg Execute ⍎ can return the same as ⎕ which is basically ⍎⍞

RubenVerg

execute is also not a normal verb??

Adám

@RubenVerg 1. Because adverbs (monadic operators) didn't exist at the time when ○ was designed. 2. because it allows useful things with outer product etc.

RubenVerg

yeah no I think I'll keep the behavior I had expected from Dyalog in my dialect

Adám

@RubenVerg No, it can do fun stuff:

      1(⍎'+2'[?2])3
4
      1(⍎'+2'[?2])3
1 2 3
      1(⍎'+2'[?2])3
4
      1(⍎'+2'[?2])3
1 2 3
      1(⍎'+2'[?2])3
1 2 3

RubenVerg

14:53

@Adám I was trying to think of when you could need to dynamically modify the left argument to Circular, I guess Table makes sense

@Adám thanks I hate this

LdBeth

The design of ○ enables doing thing like rotate an vector

Adám

@RubenVerg Me too. Super useful, though, when a variable specifies which function to call.

RubenVerg

docs for execute don't really say it can return non nouns

I guess "expression" includes verbs and modifiers

one example of that could be nice though

Adám

Maybe better not to tell people? ;-)

rabbitgrowth

@Adám What should the "correct" behaviour be?

RubenVerg

14:58

I don't have access to a standard Dyalog install on mobile, but tryapl's execute doesn't do what the docs say (though I'd expect it to be custom implemented to maintain the sandbox)

LdBeth

The "real" execute does return function trains

Adám

@rabbitgrowth This seems correct, no?

LdBeth

Meta programming!

CL-USER> (april:april "1(⍎'+2'[?2])3")
#(1 #<Anonymous Function #x302002C104EF> 3)

rabbitgrowth

Yeah it does, but you said you hated it, so I thought there was something wrong or bad with it

LdBeth

lol this is kinda expected

Adám

15:04

@RubenVerg Right, it can't output. But 1(⍎'+2'[?2])3 does work.

@rabbitgrowth I can hate something that's correct ;-)

RubenVerg

does {⍎'⍨'} not work or does it do something weird like become a dop

rabbitgrowth

:)

Adám

@RubenVerg dfns cannot return functions or operators.

RubenVerg

15:25

I'm starting to doubt everything I know :)

I can kind of accept quad allowing non-nouns (if I start thinking about it as a special name and not as a niladic function)

but the verb execute returning a conjunction is just too much

Adám

You can write your own tradfn that returns a function or monadic operator too.

RubenVerg

why only monadic operators?

Adám

Not sure. Much of this stuff seems haphazardly designed at best. Or maybe it just fell out of the implementation.

RubenVerg

to me it sounds like implementation details leaking through (egthe result of the function is whatever was assigned to the return variable)

Todos,←todo "Consider if I want to do the weird stuff Dyalog does with Execute"‿Priority.Medium

André Leria

16:53

From my search, it seems like the last conversation about this topic here was about 5 years ago, so... since then, has there been any development (or at least serious discussion) about APL/BQN and game development?

From what I've seen, the best bets right now would be using stuff like Py'n'APL with Pygame, or Lua*APL with Love2D, or some Rust + Game lib + BQN hackery... which could work but doesn't seem promising

RubenVerg

18:49

I was working on a port of olc/pge for dyalog

I had some issue I can't remember though

Adám

19:26

@AndréLeria Stormwind is the only notable entry.

André Leria

Hmmm does it still exist? I can't find anything by searching "stormwind apl", "stormwind bqn" or "stormwind gamedev"

Adám

Stormwind-Simulator @ Dyalog Usermeeting 2016/Glasgow

►

André Leria

I'll check, thanks

Transcript for