The APL Orchard

Adám

00:00

Afaik, only APL and J have IDEs where you can go back up and edit previous input (and output) for (re-)execution.

0xACE

Yeah, to my eyes it seemed to magical i didn't realise it until i saw it being used in a APL video... Makes me wonder if there are other tricks in the editor that I'm unfamiliar with

Adám

Amazing, though, that such an interface has been a mainstay of APL implementations since CRTs took over from type writer terminals.

@0xACE Did you read the tips?

0xACE

Yeah i was reading the manual in regards to CLI usage, I wasn't expecting dyalog via cli to support the extra features (im amazed of how cool ")ed myfunc" is) but I couldn't figure out how to invoke the BK (ctrl-shift-backspace) and other shortcuts...

nah, ill check it out now

Adám

@0xACE This might help.

0xACE

Aawww :'( that's so beautiful how come i couldn't find it by navigating dyalog.com

Adám

00:10

@0xACE I can find it by searching help for "linux terminal".

dzaima

@Adám (i've yet to figure out what the APL key there means)

0xACE

dzaima that's the exact problem i had, but atleast the new link resolved Cmd and CMD for me...

Adám

Well, it depends

dzaima

@Adám well, afaik, I have no way of allowing a terminal to see the windows key (and i definitely don't want to!)

0xACE

I sortof modified this gist.github.com/Russtopia/20ae5ab571d5bed73359ca956d9010a7 but for w/e reason when i swap over to it CPU spikes too 100% for a couple of seconds...

dzaima

00:14

@dzaima (but ctrl+x b works)

0xACE

dzaima, you can but it'd require that you modify your terminal and termcap for your terminal, also that applications accept it. afaik it's troublesome...

i was considering making a python script to interpret `<Keystroke> as a glyph

dzaima

@0xACE for just entering apl chars? you can use XCompose for e.g. `a → ⍺

0xACE

yes. for entering apl chars

not familiar with xcompose, let me have a look

dzaima

this in ~/.XCompose works (note that you need to restart a given program for it to recognize the updated .XCompose file)

0xACE

@dzaima do you have a full list of this for apl?

dzaima

00:20

@0xACE no

you can probably extract a list of mappings from the browser bar, the xkb apl layout, or wherever else there's an APL keyboard. gtg though

0xACE

alright, thanks :)

LdBeth

04:06

@0xACE you might want to try the Genera Lisp Machine :) or McCLIM's repl (which is a cut down version of that)

Razetime

04:40

As an idea for the APL twitter bot, it may make sense to only display an image of the output, and link to tryapl with the actual text since Twitter doesn't have code formatting

Adám

05:15

Ooh, expression + TryAPL link + picture!

AviFS

@Adám Not adding (+=)(⊸)(÷⟜≠)(⌾‿⌾), ` (≡÷≠)⊢⟨⌾,⌾⟩` & (≡÷≠)⟨˙⋄˙⟩?

Darn, I missed them. I see!

But I do still like the cuteness factor on this worm best for the main entry: (+=)(⊸)(÷⟜≠)(⌾‿⌾)

Razetime

05:41

@AviFS pull request to bqn crate

kimmolinna

@Adám FYI. I have a computer under a centralized management. So basically I can't do nothing. The main problem with Dyalog is that registry is out of question. I can start Dyalog with DYALOG_NETCORE=1 which allows me to use binary version of .NetCore 3.1. I was able to do this with Jupyter-Lab Notebooks too. I just added it to line 233 in kernel.py like this
self.dyalog_subprocess = subprocess.Popen([dyalogPath, "RIDE_SPAWNED=1", "DYALOG_NETCORE=1" , 'RIDE_INIT=SERVE::' + str(
self._port).strip(), 'LOG_FILE=nul', os.path.dirname(os.path.abspath(__file__)) + '/init.dws'])

Adám

@kimmolinna You can use a config file instead of the registry.

kimmolinna

@Adám That's true but I made a mistake and asked them to install Dyalog for all users so I can't edit the config file.

Adám

@kimmolinna But you can just add your own config file which overrride that.

kimmolinna

@Adám What is the easiest way to save all present settings to configuration file and use it? That's not so well documented...

Adám

05:53

@kimmolinna Because there's no automated way to do so. Lots of questions there: Should it also write defaults? What about settings that have been set to the default to override another non-default settings, etc.

kimmolinna

Not necessarily. The easiest way would be something like

{
Extend: "my_default_configuration.dcfg",

Settings: {
DYALOG_NETCORE: 1
}
}

AviFS

@Razetime Wdym? I meant for Adám to add to his submission.

kimmolinna

@Adám And how can I start dyalog with my own config file?

Razetime

06:17

@AviFS oh, to the submission.

AviFS

Yupadee :)

RGS

@Adám "expression" means result or original one?

kimmolinna

@Adám I found it from documentation. I can use ConfigFile="dyalog.dcfg"

Adám

06:36

@RGS Result.

RGS

So the link to TryAPL was easy twitter.com/dyalogaplbot/status/1407227956091666433

And generating the image with the result is also easy

I only need to figure out how to size the image appropriately.

rak1507

07:50

@RGS what'd you use for generating the image?

RGS

I'm using PIL

And that is already working fine.

Razetime

@RGS link's broken

RGS

I'm just having a final hurdle with figuring out a priori what is the length of the status I'm updating

@Razetime yeah I keep deleting the bot's answers and reuploading

The "stable" link is to the original tweet: twitter.com/rojergs/status/1407114811624808450

AviFS

How do we use the bot again? It's not in aplwiki.com/wiki/APL_Orchard; we should add it!

Hey @RGS, great to be back and great to see you again!

RGS

@AviFS Which one? If you are talking about the Twitter one, let me just have everything running smoothly first :P

Great to see you ○/

Btw, does anyone have suggestions for free/inexpensive hosting for the bot? (The bot was written in Python)

(I can't host it on replit because I don't have the paid plan that allows you to have your code always on)

AviFS

07:54

(P.S. I'm listening to Portuguese music; I just fell into it in the past week. I've been listening to Roberto Carlos non-stop!)

Adám

@RGS We'll host it!

AviFS

@RGS Wait no, sorry for the confusion. I meant the one that used to be here.

But then wasn't due to the pandemic, but now is again?

For executing APL

RGS

@AviFS yeah, the bot is still sick :P Hasn't been around for a long time.

AviFS

@AviFS (Let me know if you have any recommendations!)

@RGS Ouch! Really? :((

RGS

@AviFS Depends on the type of music you enjoy listening to.

AviFS

07:56

Soft rock/pop/ballads these days

I can ask my question without the bot, though. (Dyalog 18.0)

    3=/1 0 0 0 1 1
1 0 1 0

Can someone explain it please? I would've expected 0 1 0 0

RGS

@AviFS that's exactly what it returned :P

AviFS

Hahaha, that's pretty funny. You're right, let me edit that

Bubbler

It's (1=0=0)(0=0=0)(0=0=1)(0=1=1), evaluated RTL inside each paren

RGS

3=/1 0 0 0 1 1 is
(1=0=0) (0=0=0) (0=0=1) (0=1=1)

AviFS

Oh my goodness...

I know

But I was expecting...

=/

Yeah that's what I meant

But that doesn't work

RGS

08:00

except that doesn't work out of the box

AviFS

But that's exactly what I was expecting

rak1507

=/ on a boolean vector is the same as 2|+/

AviFS

@Bubbler Thanks Bubbler, very obvious in retrospect. I was hoping for (=/1 0 0)(=/0 0 0)(=/0 0 1)(=/0 1 1)

rak1507

@AviFS that is what it is

=/ does not check if all of them are the same

Bubbler

^

There's no all-equal built-in in APL

rak1507

08:01

that would be ∧/2=/ or something

AviFS

Oh yeah...

RGS

3 (1≥≢∘∪)/ 1 0 0 0 1 1

AviFS

~~Wait, that'd be ∧.=~~

Wait no

rak1507

⊃∧.=⊢

RGS

@rak1507 ⍒≡⍋

Razetime

08:02

i like how RGS feels the need to be nice to the bot when debugging the output

AviFS

What's happening?!

rak1507

@RGS actually no that doesn't work either

bc 3 f/ x is f/¨ 3,/ x not f¨ 3,/x

AviFS

I don't understand how to use either of these, nor how they would work: ⊃∧.=⊢ & ⍒≡⍋

RGS

@rak1507 right -.- it gave incorrect output to my face, but I saw some output and thought it was correct :P

AviFS

@RGS This isn't working either, it's returning 1 1 1 1

rak1507

08:03

(⊃∧.=⊢)¨3,/1 0 0 0 1 1

RGS

@AviFS yeah today my brain isn't working

Razetime

@AviFS if the ascending order matches descending order, then all elems are equal

AviFS

@RGS Not at all, just way more serious answers than I was expecting, haha

Trying to stay above water :p

Razetime

(⊃∧.=⊢)¨3,/1 0 0 0 1 1
0 1 0 0

rak1507

@rak1507 actually no it isn't, oops

@Razetime which is the expected result

Bubbler

08:05

Easier one if you allow some processing after 3 f/: 0 3∊⍨3+/1 0 0 0 1 1

AviFS

@Razetime Ah that yes, but I'm not trying to check if ALL elems are equal. I want to do it in chunks

Like 3+/ would add in 3-piece chunks

rak1507

@Bubbler yeah this is nicer for bools only

Bubbler

One byte shorter: 0=3|3+/1 0 0 0 1 1

      3(⌊/=⌈/)1 0 0 0 1 1
0 1 0 0

Adám

@AviFS For a more general sliding window, look at ⌺

Un-ninja'd.

rak1507

      1 1⍷2=/1 0 0 0 1 1
0 1 0 0 0

AviFS

08:08

Stencil is a good idea

Thanks

Adám

:58407805 You go ahead. I have things to do…

AviFS

@Bubbler Neat!

rak1507

      2∧/2=/1 0 0 0 1 1
0 1 0 0

AviFS

I'm trying to trainify this, to see if that's shorter, since I just want to do this on a boolean array

Adám

@Bubbler That's very clever.

Razetime

08:09

      {⊂(⍋≡⍒)⍵}⌺3⊢1 0 0 0 1 1
0 0 1 0 0 0

AviFS

Something like ⍱/3 0∊+/1 0 0 0 1 1

Razetime

pretty prime example for window

AviFS

Ouch. Not great for golf though

Razetime

windows needs enclosing however

Adám

@Razetime {1=≢∪⍵}⌺3

@Razetime No need to enclose simple scalars.

AviFS

08:09

Can you trainify 3 or 0 in 3+/1 0 0 0 1 1?

rak1507

@AviFS 2∧/2=/⊢

AviFS

@AviFS Like this

Adám

(0 3∊⍨3+/⊢)x or 3(+/∊0,⊣)x

Razetime

      (⍋≡⍒)⍤⊢⌺3⊢1 0 0 0 1 1
0 0 1 0 0 0

AviFS

@Adám Thanks!

It didn't occur to me this would all be so ungolfy

Does anyone have any better ideas for doing this with a boolean list?

Bubbler

08:12

0=3|3+/⊢ for Adám's first one

Adám

Nice.

@Razetime 1=≢⍤∪⍤⊢⌺3 or {1=≢∪⍵}⌺3 have same length with better performance.

Bubbler

@AviFS I think ⌊/=⌈/ is indeed shortest as a dyadic function if you pass the window size as its left arg

AviFS

I was also working on something to do with:

(3^/1 0 0 0 1 1) ∨ (3∧/~1 0 0 0 1 1)

But haven't golfed that very well yet

Bubbler

I think you mean ∨ in the middle

AviFS

Thanks, good point

Adám

08:15

@Bubbler Should be ∧/=∨/ for Booleans, no?

Bubbler

For better reading, yes

Adám

Even ⌈/=⌊/ works only for real numbers.

{1=≢∪⍵}⌺3 is the proper solution for all types and ranks.

AviFS

Thanks, I'm surprised that's the best one can do

@AviFS You don't think something like this could be golfed shorter?

Adám

@AviFS (∧/∨∧/∘~)

Isn't that the same as (∧/>∨/) or something?

Razetime

@Adám interesting

AviFS

08:18

I suppose not then! Well thank everyone, that's all really interesting

Razetime

(I forgot that windows had a left arg)

AviFS

Is anyone else surprised that it isn't easier to do, given how easy it is to n-wise reduce with other dyads?

Adám

@AviFS That's why J has n f\y for applying f on windows of size n without forcing a reduction.

AviFS

Is there a use for =/ the way it usually is?

Adám

Usually only with a left arg of 2.

AviFS

08:20

I wonder if golf-APLs could stand to benefit from implementing =/ like this, as an exception to the rule

Adám

That's why K has an operator for our ¯2 f/

AviFS

@Adám The left arg of 2 would still work the same if you did

Adám

@AviFS So =/ and {⍺=⍵}/ wouldn't give the same result? Please don't.

Bubbler

A reduction function for window size larger than 2 is impossible to design in general

Adám

@Bubbler What?

Razetime

08:21

@Adám K's ` and /` are quite insane in capability

AviFS

@Adám Mathematical purity is of course lost, but that's why I said for golf-APLs

Adám

@Razetime Try doubling your backticks!

Razetime

ah eyah meh

Adám

Not too late, mate.

Bubbler

@Adám I mean, the function f in n f/ x when you want to do some arbitrary thing n-wise

Adám

08:23

@Bubbler You mean to counteract the reduction? Sure.

Technically, it isn't too late to add J's f\

Razetime

I don't like the idea of adding additional builtins to APL, which is most of the reason why I haven't made a golfed APL yet

AviFS

It's not additional, it's instead

Assuming you mean the =/ thing

Adám

The thing is that pretty much every one of the original APL operators has issues.

Bubbler

@AviFS No, it won't be a thing ever

AviFS

Are we talking about the =/ or something else?

Ouch. This is some strong opposition

Razetime

08:26

@AviFS I'm talking about a shorter APL in general

Adám

Actual golfing languages tend to have an AllEqual built-in anyway.

AviFS

You mean a golfier APL?

Razetime

..yes

AviFS

@Adám Is extended intended for golfing?

Or is it just a convenient use?

Adám

Why not both?

Razetime

08:29

extended is kinda like a beta for what could possibly be in the next dyalog version

which also happens to be very golfy

AviFS

@Adám Just wondering if you had an 'intended audience?'

And to what extent it favors/balances mathematical elegance with practicality

Since that just came up

rak1507

well it definitely doesn't do anything crazy like redefining specific reduces or anything like that

Adám

It was intended to showcase how Dyalog APL can be extended.

AviFS

I see, so the philosophy is the same then?

Just a little more experimental and cutting-edge?

Adám

The only breaking changes are there by mistake. (Oh, and then gave a conflicting meaning to a previously unused primitive.)

@AviFS Same as?

AviFS

08:33

As current Dyalog APL

Adám

Yes.

@AviFS Very much experimental, as it is based on crazy hacking.

AviFS

Illuminating, thanks. I'd figured it was more of a recreational & golfy thing than a demoing-the-Dyalog-of-the-future

Adám

@AviFS APL′ is intended to show what Dyalog APL could be if problematic primitives were fixed.

AviFS

I like the pun here:
> The goal is to make APL more appealing to the modern audience. That is, to bring APL into its prime.

Interesting, I see. So that one includes breaking changes, but extended doesn't. Correct?

Adám

@AviFS Atop, Over, and Constant were adopted by vanilla Dyalog, straight from Extended.

AviFS

08:36

And even APL' is breaking only in the service of the math and principles , correct?

RGS

CMP: What do I do with multi-line results? twitter.com/dyalogaplbot/status/1407255985316581377 and twitter.com/rojergs/status/1407239015456292865

Adám

@AviFS Yup, almost every change in APL′ is breaking.

Razetime

@RGS i say completely remove text output

Adám

@AviFS Also utility (though you could call that "principles"). First axis functions are much more versatile than last-axis functions, and hybrids are just awkward.

@RGS What's with the trailing dots?

RGS

@Adám currently I'm trying to keep lines short enough so they don't wrap (hence the ∙) but I also have the hard limit on total character count, so I add the "..." when I hit that.

Razetime

08:40

text output just loooks bad on twitter

Adám

I agree that you should remove text output altogether if multi-line.

AviFS

@Adám Yeah, I'm calling them principles for lack of a better word. It's not to contrast it against practicality, but to say that the utility gained is from 'better math,' rather than hackish oddities a la C++.

RGS

@Razetime even if it is single-line? I figure single-line output might be nice for expressions that evaluate to simple scalars/vectors

Razetime

twitter has a 280 char limit, i think it's best to save all of it for a tryAPL url, and the image wto show output

Adám

@Razetime I think all links count as a fixed amount.

RGS

08:43

@Razetime the image doesn't take char space and the URL always takes 23, so I have 257 chars to play around with.

Razetime

URL always 23? that helps a lot then

RGS

I can use a couple to say something : "Here's a link to your expression: <tryapl link>"

then if it's single line result, add it; if it's multiline, either say something about the image or say absolutely nothing.

Razetime

maybe anything that
a) doesn't require boxing
b) is not multiline

RGS

if it requires boxing, it is multiline.

Razetime

well, 3 3⍴3 would be multiline and non boxed, no?

RGS

08:45

yes

I just mean to say that the "requires boxing" restriction adds nothing.

I just need to check the number of lines.

Razetime

ah ok

I'm considering submitting an APL notebook to a Kaggle challenge

RGS have you tried any contests on Kaggle

rak1507

that reminds me about the bioinformatics contest what were your solutions @Razetime and @RGS

Adám

@RGS Why not give it as an expression if multiline? (A: because you can't get an expression from TryAPL.)

But hey, I can maybe add that.

RGS

@Adám maybe ⎕SE.Dyalog.Utils.repObj could be whitelisted

Adám

Yeah, or simply add ]repr

Adám

09:08

Yesss, I finally managed to consistently repro an obscure bug that has been bothering me for years.

rak1507

are you allowed to say what it is

Adám

Sure. Only happens in Windows IDE. Search for some text, click inside the resulting selection, then hold shift and use a movement key to select. Rather than what should be selected, everything before the caret is selected.

Razetime

@rak1507 i did the bioinformatics contest in ruby although every competitive programmer i knew told me it was bad

rak1507

oh

ruby is bad, they are right, slow and memory intensive

closest←{i⊃⍤0 1⍨⊃⍤⍋⍤1|⍵-⍤0 1⊢⍺[i←1⌈(≢⍺)⌊0 1∘.+⍨⍺⍸⍵]}
r←''⋄{}{m a s←{⍎'¯'@('-'∘=)⊢⍵}¨1↓4↑⍵⋄r,←⍕¨⊃,/(⊂sm←m[g←⍋m]){{⍵,¨⍨g[⍵⊃⍤0 1⊢c]}⊃⍤⍋⍤1|⍉⍵-⍤1⍉(⌈/⍬)@(0>⊢)a+⍤1⊢⍺[c←⍺ closest ⍵∘.-a]}¨{⍵⊂⍨(≢⍵)⍴200↑1}s⋄4↓⍵}⍣{0≡≢⍺}1↓data

Razetime

it is not like I bothered anyway

rak1507

09:19

p2

Razetime

I got throught to sorta trying p3 in K and then err decided against it

rak1507

yeah I did a simple solution in apl but it was wayy too slow and couldn't be bothered figuring out how to do it properly

RGS

 Solver←{
     (m a t)←⍎¨¨' '∘(≠⊆⊢)¨⍵
     val←(pos←val>0)⌿val←,m∘.+a
     ord←⍋val                      ⍝ How to order possible values in ascending order.
     idx←(¯1+≢val)⌊0 1∘.+val[ord]⍸⍤1 0⊢t
     best←(⍉idx){⍺⊃⍨⊃⍋⍵}⍤1⊢|⍉val[ord][idx]-⍤1⊢t
     ⍕¨1+(≢m)(≢a)∘⊤¨(⍸pos)[ord][best]
 }

My P2

Didn't solve test case 4, only 1-3 and 5

rak1507

ah

looks fairly similar to mine also using interval index

RGS

That's the core, at least; I had a :While loop to read the test cases and feed them to the Solver

rak1507

09:22

ew, control flow

{⍺⊃⍨⊃⍋⍵}⍤1 will be slow bc ⊃⍋ will be n log n, ⊃⍤⍋ will use the idiom and be much faster for large arrays, not that I suppose it particularly matters here

m∘.+a will be the killer

RGS

@rak1507 what idiom?

rak1507

      a ← ?⍨ 100000
      ]runtime -c '⊃⍤⍋a' '⊃⍋a'

  ⊃⍤⍋a → 1.2E¯5 |     0% ⎕
  ⊃⍋a  → 4.2E¯4 | +3329% ⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕⎕

RGS

Lol then why isn't ⊃⍤⍋ coloured as an idiom?

rak1507

no idea, don't think I have that turned on anyway

Razetime

⊃⍤⍋ is also not on APLcart

RGS

09:49

Do you think this is a good Tweet example to have pinned at the top of the bot? twitter.com/dyalogaplbot/status/1407272520919797762

I'm fairly happy with what I have now, for me I'd leave it like this for a couple of days to see what happens.

Adám

@RGS Why is the image so much taller than the session display?

RGS

I give it a min width and height so that it doesn't zoom in like crazy and makes everything look bad

Your session will also have plenty of empty space when you open the interpreter :P

Adám

I see. And then it always shows the session transcript?

RGS

@Adám "always"? You mean the full thing? If it is too long it gets truncated.

Adám

No, I mean, it doesn't just show the result, it shows the transcript including input.

RGS

10:00

On top of TryAPL's default ]rows behaviour, if the full session has 101+ lines, it gets truncated to 100 lines. But if things aren't crazy-long, it shows input+output

You can give me a couple of things to try, I'll tweet them at the bot, and we can work from there.

Adám

3⍴'`'

RGS

@Adám the bot doesn't escape strings right now, and it won't for a long time. I don't feel like writing a parser

Adám

@RGS It is super easy to parse APL strings.

RGS

@Adám even with comments and all? if so, then maybe I can add that.

Adám

dyalogaplbot←5 then - @dyalogaplbot ⍳10

RGS

10:03

Right now I just use a simple regex to look for code: "`(.*?)`"

@Adám twitter.com/dyalogaplbot/status/1407278309852037123

Adám

' @dyalogaplbot ',4⍴⎕UCS 96 9075 51

RGS

@Adám hehe that's a nice one; let's see what happens

Adám

Next up: a quine

@RGS Did you forget backticks?

RGS

@Adám yes give me a sec

Adám

Huh, how comes it didn't react to its result tweet?

RGS

10:10

@Adám it did

but there's an interval

Adám

Oh, I see.

RGS

I can only query for mentions 75 times every 15 mins, so I have to pull mentions once every 12s.

So there was a delay between the mention and the reply.

Adám

CMC: Quine that includes @dyalogaplbot and two backticks with something between them.

RGS

xD

Adám

@RGS You might want to let it ignore tweets from itself.

RGS

10:18

True, that is a restriction worth considering.

Now I'll go afk for a bit, if people come up with such a quine, feel free to throw it at the bot.

(The bot might fall asleep if my computer does; still have to figure out hosting @Adám)

Adám

Jason will set that up for you.

xpqz

@rak1507 I find that depressing.

Adám

@RGS Because it isn't a single token. Rather ⍤ recognises certain operands.

RGS

10:45

@Adám ah I understand, thanks.

RGS

10:57

@Adám you were right twitter.com/rojergs/status/1407291291981991940

Adám

Blog post: News and Recommendations Regarding Dyalog Versions 17.1, 18.0 and 18.1

3

@JeffZeitlin ^

@RGS :-) Did you make the bot ignore itself?

RGS

not yet

on it, though

Razetime

ooh, recursive tryAPl bot calls lmao

Adám

@RGS How do you construct the TryAPL URLs?

RGS

What specific part do you want to know about?

Wait are you talking about the requests to execute code or the "try it online" links in the replies?

Adám

11:05

The links in the replies.

RGS

I take all the code that was ran and join it with " ⋄ ", then I do fancy url escaping with urllib.parse.quote_plus and then I plug it into "https://tryapl.org/?q=$1&run"

Where $1 is replaced by the escaped code.

Adám

Can you try {⍵}2?

RGS

@Adám twitter.com/dyalogaplbot/status/1407295282489012224 it appears to have worked

Adám

Thanks, that works.

The thing is that standards don't require %-escaping {} but if they occur in a URL, Twitter gets confused.

RGS

Alright, so now the bot avoids recursion introduced by other users, but still replies to its self-replies

so that the bot can write threads and evaluate its own code

But it won't evaluate code that appears in replies to evaluating code from others.

Razetime

11:18

very neat

how do i set default RIDe editor

RGS

11:38

Is this sensible behaviour? twitter.com/rojergs/status/1407301745328410631

Adám

I think that's very neat.

We should make TryAPL understand %0A…

RGS

Is that the newline?

Adám

Yeah.

RGS

That'd be neat.

Adám

We're planning multi-line input anyway.

Bubbler

12:16

> complete confidence

Maybe time for formal verification?

Adám

@Bubbler What do you mean? Open sourcing it?

Bubbler

@Adám No, formal verification is e.g. writing a computer-assisted proof that the given algorithm works correctly on all possible inputs

Adám

@Bubbler Hm, that's an interesting idea. Not sure how one would go about it, though.

Bubbler

12:32

@Adám This is a great textbook in that topic, if you're interested

Volumes 5 and 6 are actually new, and it looks like you can verify C programs directly now (VST project was still ongoing when I was first learning Coq)

Marek Kowalczyk

19:16

@Adám Thank you for the pointers! I'm now digesting the Legrand book.

Adám

@MarekKowalczyk Great. Feel free to ask if you have any questions or run into roadblocks.

hyper-neutrino

20:08

there was some discussion in TNB about a reduce-like operator that actually works for =/, and similarly for >/ and </

i implemented it as {∧/¨ 2⍺⍺/¨ ⍺ ,/⍵} - is there a better way of doing this (in terms of actual code style / efficiency, not golfing)?

and also, if i want this to work for both monadic and dyadic (like, ⍺ might not be provided), how would I go about doing that? I'm not very familiar with implementing custom operators

Adám

@hyper-neutrino It would only work on simple vectors. The proper coding would be {∧/¨(2⍺⍺/⊢)⌺(⍪⍺1)⊢⍵} or {(∧/2⍺⍺/⊢)⌺(⍪⍺1)⊢⍵}

@hyper-neutrino So you'd want the default to be 2?

hyper-neutrino

well, standard reduce reduces the whole thing so ideally this would work the same way

Adám

@hyper-neutrino So monadic should be identical to the current n f/ y?

hyper-neutrino

uh, I think? ¯\_(ツ)_/¯

like =? where ? is the new operator should check if a list is all equal

Adám

Ah, then without a left argument, it would be ∧/2f/⊢ but with a left argument it'd do so on windows. Got it.

hyper-neutrino

20:23

yeah, that looks right

Adám

@hyper-neutrino Using your above definition (which only works sensibly on simple vectors) it'd be really easy: {⍺←⊢ ⋄ ∧/¨ 2⍺⍺/¨ ⍺ ,/⍵}

@hyper-neutrino Using Stencil: {⍺←≢⍵ ⋄ (∧/2⍺⍺/↓)⌺(⍪⍺1)⊢⍵} but note that this preserves the number of elements by using shorter windows at the edges. Also note that ↓ replaces ⊢ from above. Otherwise we add padding rather than look at shorter windows.

hyper-neutrino

how does ⍺←⊢ work

is it that if the left argument is not defined it replaces it with right-tack so ⊢ ,/⍵ is just ident(reduce ⍵ on ,) ?

Adám

20:42

@hyper-neutrino Yes exactly. ⍺←… is basically a conditional in that the whole expression is only run if the containing function was called monadically, but skipped if it was called dyadically:

      {⍺←⎕←'monadic' ⋄ 42}⍬
monadic
42
      ⍬{⍺←⎕←'monadic' ⋄ 42}⍬
42

hyper-neutrino

oh, that's cool! so it's a special-case for ←?

Adám

It is a special case for ⍺← in that any (non-whitespace) variation is an error. E.g. ⎕←⍺← is an error and so is (⍺)←

While a function cannot take a non-array as argument, ⍺ can be assigned any type of item, even a function or operator value.

E.g. a (simplified) model for ⍥ is {⍺←{⍵ ⋄ ⍵⍵} ⋄ (⍵⍵ ⍺) ⍺⍺ (⍵⍵ ⍵)}

hyper-neutrino

ah, okay. that's very interesting

Adám

Normally, x f⍥g y is (g x) f (g y) but if ⍺ gets a "bypass" operator value, namely a monadic operator that ignores its operand and returns its argument, then the monadic call is (g{⍵ ⋄ ⍵⍵}) f (g y) making the left g be ignored and returning the result of f (g y)

(The real ⍥ is not quite this, because it has to handle shy functions, but this version is fine for modelling.)

hyper-neutrino

i'm not sure i understand here how ⍺←{⍵ ⋄ ⍵⍵} is working, specifically {⍵ ⋄ ⍵⍵}

so if it's just f⍥g y instead of x f⍥g y, then ⍺ is set to {⍵ ⋄ ⍵⍵}?

Adám

20:50

Yes.

hyper-neutrino

and so the second statement (⍵⍵ ⍺) ⍺⍺ (⍵⍵ ⍵) is (⍵⍵ {⍵ ⋄ ⍵⍵}) ⍺⍺ (⍵⍵ ⍵) from what I can tell

Adám

Oops. I made a typo.

It should say {⍵ ⋄ ⍺⍺}

Dfns and dops terminate on the first statement which is neither an assignment, a guard or an error guard, so ⍺⍺ is never reached, as ⍵ is returned, but it is the mere mention of ⍺⍺ that makes it into a monadic operator.

hyper-neutrino

ah

Adám

With ⍵⍵ it would fail because its mention would create a dyadic operator which isn't being given a right operand. Insta-fail.

hyper-neutrino

ah okay this makes a bit more sense

Adám

20:53

I'm so sorry about that.

hyper-neutrino

no worries - thanks for explaining :)

Transcript for