The APL Orchard

@ab5tract that's actually true, it matters a lot less than one might think initially, but the default is still wrong and users shouldn't be bothered with this kind of choice.

rak1507

I agree, I don't know why on earth they made it configurable in the first place...

Dyalog APL

1:08 AM

<klg> it should be that 0≠↑0⍴⎕IO and you have to always do ±⎕IO to convert between integers and indices ;)

ab5tract

@ngn 50+ years of backwards compat sure is a helluva lot to maintain

@rak1507 because it always has been. APL/360 had it, according to the book I'm currently reading from 1970

rak1507

But why did they choose that?

Who was the first person who thought 'I know, this should be configurable!' and more importantly, why

ngn

@rak1507 iverson himself wanted 0. some "domain experts" wanted 1. it was a solomonian solution.

rak1507

Experts in what, I assume not programming? Did any other languages at the time use ⎕IO←1?

ab5tract

There were only a handful. FORTRAN used 1. MATLAB uses 1.

Not that MATLAB was around at the time

rak1507

1:13 AM

Huh, FORTRAN used 1, didn't know that

ngn

@rak1507 "domain expert" = a person who can't program but has enough control over money to change the way programmers behave

rak1507

lol ouch

ngn

@ngn without the "control" part we would be calling them idiots

ab5tract

I think that's a bit harsh considering the times we are discussing. There were a great many lessons yet to be learned and very few examples to go by. I think mathematicians are in general more used to indexes starting at 1, or certainly were at the time

It's not exactly intuitive to go from saying "the first element of vector x" to using an index origin of 0

That said, I've just set ⎕IO to 0 in the Dyalog settings (Windows) :)

ngn

@ab5tract yes, but generally mathematicians can easily change habits. they understand starting from 1 is not essential, and in some parts of mathematics, like number theory, ⎕io←1 would actually be very unnatural. the residue classes modulo n traditionally go from 0 to n-1, not from 1 to n

ab5tract

1:20 AM

Fair, but again, this is over half a century ago. The book I have discusses operating systems as a newfangled concept

rak1507

codegolf.stackexchange.com/questions/216902/… lol people calling ngn's solution code golf

ngn

@rak1507 yeah, "unreadable" etc.. :) actually, i'm in the middle of writing a better solution, using nfa->dfa conversion

rak1507

idk what NFAs or DFAs are but cool

ngn

@rak1507 have you come across the chomsky hierarchy yet?

rak1507

Nope

ngn

1:26 AM

@rak1507 if you're studying computer science (or something equivalent), you will

rak1507

Probably won't be taught it for a few years then

ngn

NFA = non-deterministic finite automaton. DFA = deterministic

rak1507

Ah

ngn

they are like graphs that have states (the vertices) and transitions (the labelled edges)

rak1507

I think I've seen some examples before

ngn

1:28 AM

they can recognize languages (i.e. sets of strings) by consuming one char at a time and following the transitions

rak1507

right, how does this apply to the question? I can see how you could potentially make a graph with all strings or something, but I'm not sure

ab5tract

@ngn "they" here meaning meaning chomsky hierarchies?

or DFA

ngn

@rak1507 there's a (relatively easy) way to build an nfa recognizing strings withing a particular edit distance from a given input string. if it can be converted to a dfa (there's a theorem+algorithm that says this is always possible), then recognition can be very efficient. so can be generation of recognized strings.

rak1507

huh, well good luck with that, sounds interesting

ngn

@ab5tract dfa-s (or equivalently nfa-s)

ab5tract

1:33 AM

speaking of Levenhstein and languages with indexes starting at 1, Bertand Meyer just produced some videos on the topic in Eiffel: bertrandmeyer.com/2020/12/03/…

ngn

@ab5tract i believe it's easier to look it up in wikipedia than wait for me to describe it but.. the chomsky hierarchy consists of four proper nested classes of languages: regular (class 3), context-independent (2), context-sensitive (1), and general (class 0).

ab5tract

Even Chomsky uses index 0 ;D

ngn

each class has corresponding types of grammmars and some classes have corresponding types of automata that recognize them (i hope i got this right)

@ab5tract haha, nice observation :)

(i'm looking it up right now to make sure i'm not talking nonsense..)

ok, type 0 languages/grammars actually have a name: "recursively enumerable", but never mind..

we care most about the context-independent class, because that's what most programming languages are

ab5tract

@ngn I'll let it slide... but only just this once ;)

ngn

i forgot what i was going to say.. @rak1507 should i explain the levenshtein nfa construction?

(it's ok to stop me, sometimes i get too carried away writing about trivial details)

or maybe i should explain the difference between nfa and dfa?

ab5tract

1:47 AM

I'm interested in either approach. the explanation would cover the code you submitted, right?

ngn

@ab5tract the code i submitted is just bruteforce recursion

i'm still working on the clever nfa->dfa solution

ab5tract

ah , I misread what you meant about 'NFA -> DFA'

figured you already had the nfa and were working on a dfa version.

← is a compsci dropout

ngn

@ab5tract i would say i'm not 100% compsci either. it was right after the balkans' 1990s (terrible time), and i had to start work while studying at uni to make a living. somehow i managed to get through all exams.

in an nfa the transitions are from one state, consuming a particular symbol, to a set of states

in a dfa, the transitions are from one state, consuming a particular symbol, to exactly one other state

there's a famous algorithm for "determinization" of an nfa. the worst case could be O(2^n) where n is the number of nfa states, but in practice it's rarely that bad

ab5tract

@ngn I was at a free form liberal arts college where I could pursue many different topics -- and I have a lot (too many!) interests. On top of that I had already been earning paychecks as a programmer, so I figured "why would I need this?" ... youth and arrogance!

rak1507

@ngn definitely! it's fascinating

ngn

2:00 AM

@rak1507 just one note about nfa-s: under some definitions, they could have "epsilon"-transitions - transitions that don't consume an input character. nfa-s with epsilons are equivalent to nfa-s without epsilons.

rak1507

what's the definition of two things being equivalent?

ngn

@rak1507 two automata are equivalent iff they recognize the same languages

rak1507

Makes sense

ngn

@rak1507 imagine a matrix of nfa states. the height of the matrix represents the number of edits we've made. the width of the matrix represents how much of the input string we've matched.

Razetime

good morning

ab5tract

2:03 AM

happy new year Razetime. How's the future treating you?

ngn

@Razetime good year :)

rak1507

Morning

Razetime

the future looks fine idk

that will be decided when my exam results come up

any way, good year.

rak1507

Did all your exams go ahead and stuff? Mine are cancelled bc of covid

Razetime

we had them online

rak1507

2:04 AM

Ah

@ngn So when you convert your program to use the other method, will the overhead of converting it be sufficiently low that overall the other method is faster?

ngn

what does it mean to insert a char? - we move down in the matrix (number of edits increases, number of matched chars stays the same)
what does it mean to delete a char? - we move down and right in the matrix without consuming an input char (so it's an epsilon transition)
what does it mean to replace a char? we move down and right while consuming
and what does it mean to match an input char without an edit? - we just move right

this way we can build an nfa, relatively easily

then use the O(2^n) algorithm to turn it into a dfa

and then we can do whatever we want!

rak1507

O(2^n) sounds like a lot, what's the time complexity of your current algorithm?

ngn

like match strings from the language (language = set of strings within edit distance 2 from the given string)

or generate the whole language

rak1507

The entire thing

ngn

@rak1507 the upper bound is O(2^n) but due to the structure of this particular kind of nfa, it's still fast

O(2^n) is just worst-case

rak1507

2:11 AM

Cool, I assume doing it in APL would be pretty unrealistic? Sounds like it's quite scalar code

ngn

@rak1507 it's not impossible. as usual, apl (or any array language) would have some advantages and some disadvantages

rak1507

It doesn't sound like something where array operations would be ideal, maybe generating all strings could be done with some sort of outer product but I'm not sure

ngn

@rak1507 the challenge is tagged "fastest code", and it's not just basic arith on long vectors, so i'm not even trying with array languages

rak1507

But the bounty says they'll choose the 'most impressive' solution from the fastest submissions in each language, in which case maybe APL is more impressive than C? ;)

ngn

@rak1507 "most impressive" would probably imply a popularity-contest, but it was tagged fastest-code

rak1507

2:16 AM

'For each language I will look at the fastest solution. I will then choose the answer out of those that I find the most impressive.'

ngn

there is only one bounty (i don't really care about rep, of course, but it's a fun challenge for me)

rak1507

Yeah, and it looks like it's not even going to be awarded based on code speed, but on the impressiveness, which is weird

ngn

@rak1507 in that case i'd be happy to lose, as long as my algorithm works better

bruteforce is not so easy to beat, by the way

rak1507

I probably won't try, it all seems way over my head still

ngn

@rak1507 yeah, it's not too simple. better wait until you're introduced to automata theory by a professional teacher.

it's not too complicated either. i'm just not so good at explaining things.

rak1507

2:28 AM

No, you explained it well, there's just a large difference between sorta getting it conceptually and being able to use and implement it in practice

ngn

i hope i can finish my c solution in the next few days. you'd probably be able to figure out the details from there.

btw, user Anush (and now his friend fomin) has been posting variations on the same challenge for years! :)

Razetime

yeah, but who can really blame them

rak1507

Oh yeah

Must be interested in that stuff I guess

ngn

@Razetime i don't. many of these are great challenges.

Razetime

Somehow their names are not levenshtein neighbours

ngn

2:38 AM

:D

Elias Mårtenson

7:06 AM

@ab5tract Basic uses 1. Of the current popular languages, the only one I know of that still uses 1 is Lua.

@ab5tract Well, I started at the beginning of COVID, first commit is 8'th of January. I've been working on it on and off during the year in parallel with some other projects.

@ngn Maths people tend to use 1, for some reason. Probably tradition. The reason they get away with is is because you rarely actually do arithmetic on indexes the way you do in programming.

Then you have the "regular people" who cannot understand the difference between ordinals and cardinals. Except in the anglosphere, people do seem to have no problems dealing with it when it comes to centuries (19'th century being 18-something)

I also note that R uses 1-based indexing.

Dyalog APL

8:49 AM

<moon-child> julia also uses 1-based indexing

<moon-child> is quite popular for scitech (and possibly some fintech too?)

Adám

9:16 AM

Mathematica (Wolfram Language) also uses 1.

@ab5tract COBOL had 1 too. Several other languages, contemporary to APL, had 1 too, e.g. PL/1 and Smalltalk, Other examples of languages used today that have 1 are AWK and XPath/XQuery.

Razetime

9:35 AM

@Adám With this and cvzi's suggestion, it now works with the latest ngn-apl version

Adám

cvzi?

Razetime

cvzi's the person who helped fix the script on SO

Ah, cuzi.

you know them?

No.

9:38 AM

alright

so the userscript should update when I push to github right

Adám

Yes, but now you're using ngn/apl instead of TryAPL?

Razetime

nono

this one's the aplChat script

the one which executes stuff outside the textbox

Adám

Ah, ok, sorry. Still early here.

Razetime

the tryAPL script is still the same

Adám

For that, you may want to switch to the ngn/apl hosted on my GitHub account. ngn said he'd take down his (today?)

Razetime

9:41 AM

oh lmao

just a moment

done

Elias Mårtenson

I have a parsing question. Consider the following toplevel definition: foo ← { ⍵+1 }

Then, the following second toplevel definition: b ← { foo ← {⍵+3} ⋄ foo ⍵ }

I'm implementing support for this syntax in Kap right now (for familiarity of users).

Now, I end up with a difficult-to-solve parsing issue here.

In the second line, the parser first sees foo and rightfully determines it's a function, so it attempts to parse everything to the right as the arguments to foo. The next thing is sees is ← so it bombs out with an error because there is nothing on the left side of the arrow,

How does the Dyalog parser handle this?

And for that matter, any other parser.

Dyalog APL

<moon-child> ← should have higher precedence than function application, so should get parsed first

<moon-child> stranding > ← > operator app > function app

Adám

I don't think that's right.

In the second line, the assignment to foo is local to b, and so the toplevel def is ignored/shadowed.

Elias Mårtenson

@Adám True, but since parsing is left-to-right (which is the case in Dyalog too, if I'm not mistaken) foo has been parsed as a function application before the arrow is seen.

Adám

No, Dyalog's parsing is token based, not syntactic role based.

In Dyalog APL, you cannot know the syntactic role of a name until run time.

Dyalog APL

9:54 AM

<moon-child> parsing can't go left-to-right. Otherwise how would you handle 5+x←2

Elias Mårtenson

Kap is a strict left-to-right parser, and it has no problems handling 5+x←2. What is the question?

Adám

So KAP looks ahead to the right?

Elias Mårtenson

It's a left-to-right recursive descent parser. When it finds the leftarrow it looks at the left-side arguments (which is simply a variable reference to x. It unwraps that reference to find the symbol name x and replaces it with an assignment).

Dyalog APL

<moon-child> bad example. Better is: x+x←2. When you see 'x' on the left, you don't know its syntactic class until you get to the 'x←2'

Elias Mårtenson

@Adám Yes, but that is not important in the assignment case.

This is where the left arguments to assignment is processed: github.com/lokedhs/array/blob/master/array/src/commonMain/…

Dyalog APL

9:57 AM

<moon-child> @EliasMårtenson take a look at jsoftware.com/help/dictionary/dicte.htm. (In that example it evaluates right away instead of building a parse tree for later, but you don't have to do that to use the same technique)

Elias Mårtenson

I have to read the article in detail, but at first glance that looks like it processes all the tokens first, then parses it right-to-left?

KAP only ever does a single token lookahead.

Adám

@EliasMårtenson That would be the same as in Dyalog.

Elias Mårtenson

@Adám That explains things.

OK, I have to think about this. I'm considering dropping the idea of supporting this syntax, and only ever use ∇ for function assignment.

I still want a way to do declare simple functions without explicit argument names though (which is whta ∇ does).

Maybe something like: ∇ foo ← {1+⍵}

I don't know. That's ugly too. I'm actually terrible with user interface design (and syntax is a kind of user interface)

Adám

How does KAP handle x+x←2 and y[y←0 1]? (I know it does handle them, but I don't understand how it can.)

@EliasMårtenson Do you have recursion in KAP?

Dyalog APL

<moon-child> @EliasMårtenson obviously do what you think is best, but I think you'd be better served by giving up on the one-token-lookahead parser. It sounds a bit like the lexer hack in c, where a by using a technically more complicated parser (in that case, not context-free), you actually simplify matters substantially

Elias Mårtenson

10:27 AM

@Adám Yes, there is recursion of course.

@Adám It's (small) hack, actually. When parsing a strand, it keeps a list of Instruction instances (an Instruction is merely an object that can be evaluated to yield a value). When it comes across a function, it gathers up all the collected values (it's a member called leftArgs in the class APLParser), then recursively calls the parse function again to collect the right-side arguments.

When recursing, the leftArgs is again empty and it starts to collect Instruction instances again.

Now, a plain number results in an Instruction subclass called LiteralNumber.

A function invocation instruction is called FunctionCall1Arg or FunctionCall2Arg.

A variable reference is called VariableRef.

Now, if the parser sees a left-arrow, it looks at leftArgs to check what it is. If it contains a single element, and that element is of type VariableRef, then the element is removed tom leftArgs and replaced with a an Instruction instance of type AssignmentInstruction.

It's created by digging into the VariableRef instance to retrieve the name of the variable.

With this information, you should be able to follow the code that handles this: github.com/lokedhs/array/blob/master/array/src/commonMain/…

Oh, and if you're asking about LiteralScalarValue, it's what is returned from (foo). I.e. 1 is a LiteralNumber, while (1) i a LiteralScalarValue that contains a LiteralNumber.

I guess I could handle (x) ← y like x ← ⊃y. I guess that would be consistent. If I wanted to do that, I'd decide based on whether the left side argument is a LiteralScalarValue

@DyalogAPL I think making significant changes to the parser to deal with this particular case is not ideal. By forcing the use of a one-token-lookahead left-to-right parser, the parsing rules becomes very easy to understand and less likely to create confusion.

ngn

12:20 PM

@EliasMårtenson strictly speaking dyalog doesn't parse. at "fix" time (what we would normally call "compile time") it tokenizes left-to-right and does a small amount of other work, like preparing strand fragments or adding pointers between matching brackets. the final decisions about the tree of function applications are taken at runtime, when the kinds ("nameclasses") of variables are known (there's no actual ast in memory, just a conceptual "tree"). this goes right-to-left with minimal look-ahead.

(if marshall contradicts any of ^, he's probably right)

@EliasMårtenson do you build an ast in kap? if you want to infer the kinds at compile time, why don't you add a separate pass?

@EliasMårtenson do you call that bytecode or classcode/instancecode? :)

Elias Mårtenson

12:41 PM

@ngn Good question. I guess instancecode is the closest :-)

@ngn Well, the resulting instance structure is essentially a parse tree, although it's an executable tree so it's a bit more complex than just a regular parse tree.

However, optimisations can be done on this structure if needed.

ngn

@EliasMårtenson here's an interesting case. would kap be able to parse f←{g⍵} ⋄ g←{f⍵} and infer that both f and g are functions?

Adám

1:00 PM

@EliasMårtenson What is the syntax for a function to call itself?

Elias Mårtenson

1:15 PM

@Adám The name of the function: Infinite loop that calls itself: ∇ foo (x) { foo 1+x }

And no, unnamed functions cannot call themselves directly at the moment, simply because I don't have a syntax for it. I know Dyalog uses ∇ but that's not available. So either I change the syntax for function definition, or I use a different symbol.

@ngn No. It requires functions to be forward-declared.

However, you can mimic it using function values (first class function values)

xpqz

2:00 PM

@ngn is that how ngn/apl parses, too?

Romilly Cocking

@EliasMårtenson In my experience many pure mathmos count from zero, many applied mathmos from 1.

dzaima

@xpqz ngn/apl compiles ahead-of-time

xpqz

Is there a formal grammar for apl knocking about somewhere?

ngn

@xpqz no, ngn/k tries to do more work at compile time. first it builds a half-done ast in which sequences of nouns and verbs are flat. then it walks the ast and infers the kind of every node (kind:noun|verb|adverb|conjunction or if you prefer array|func|mop|dop). variables have static kinds, you can't have things like a←0⋄a←{}. so, using the inferred kinds, it goes through the tree again and determines what the tree structure of noun-verb sequences should be.

then this gets compiled to something like bytecode (apart from bytes, it might contain js function objects but that's just due to my lack of at the time..), and interpreted

*lack of experience

dzaima

@xpqz that's kind of hard to do (e.g. see 2 (⍎(?2)⊃'2+') 2)

there is a grammar of BQN though

ngn

2:10 PM

@dzaima if you allow ⍎ to return functions, all bets are off

another enemy of parsing is ⍺←⊢

dzaima

@dzaima but if you restrict yourself to normal functions/operators and strands and no variables (let's not think about assignment as that's very different in every impl), making a grammar isn't that hard (but still, rather pointless)

xpqz

One day I might try to make my own. Seems to be a rite of passage around here :)

ngn

@xpqz "what i cannot create, i do not understand" -feynman

Adám

2:34 PM

@xpqz I think the table of binding strengths fully defines the grammar (assuming you know which symbols are in which class). See also this

@xpqz I've never attempted to write an APL interpreter of any sort. It seems such an unnatural thing to me. Probably because APL is a native language to me. otoh, I've written quite a few interpreters and transpilers for other languages ― in APL.

dzaima

2:52 PM

@EliasMårtenson in case I didn't show this to you - dzaima/APL's parser debug info (which, if ignoring brackets & schizophrenic slashes, seems to act very equivalently to Dyalog's)

xpqz

@Adám I think implementing a language imparts a deep level of understanding. But first I'm going to write a beginner's intro to APL I wish I'd had when I started.

Adám

@EliasMårtenson Since the prefix for a function is the new thing, why not use ∇ for recursion and another symbol for declaration, e.g. ƒ (iirc, this is what VisualAPL did).

@xpqz Excellent. I'm looking forward to seeing that.

Elias Mårtenson

@Adám Very good point, and my thinking in the last few hours have steered towards this as well.

I've considered using defun just like Lisp. Since I've taken so many other ideas from Lisp.

Adám

@xpqz Hey look, a formal VisualAPL grammar!

ngn

@xpqz which language do you want to implement apl in?

dzaima

3:06 PM

@Adám my ip is permanently banned \o/

ngn

@Adám 56 pages! :O

Adám

@dzaima I can email it if you really want it.

dzaima

@Adám nah i got it

xpqz

3:25 PM

@ngn I'd probably reach for Racket (Scheme), for the love of everything lisp-like.

ngn

@xpqz cool :)

xpqz

One vintage language in another :)

ngn

@xpqz does racket support vectors and arithmetic on them?

xpqz

Yeah.

It has a nice nd-array module.

Not used it in anger. But said to be pretty quick.

Elias Mårtenson

@xpqz You may want to take a look at April as well, which is an APL implemented in Common Lisp.

If for no other reason to get an idea of the approaches taken there.

xpqz

3:36 PM

@EliasMårtenson yes, I've been poking around with April a bit.

Elias Mårtenson

@Adám What syntax do you suggest? Nothing that there are two separate ways of declaring functions that I want to support: one where arguments are explicitly named, currently using the following syntax:

∇ (a0;a1;a2) foo (b0;b1) { a1+a1+a2+b0+b1 }

and then I also want to support a very terse form, where arguments are ⍺ and ⍵.

Adám

@EliasMårtenson Can i get back 2 u?

Elias Mårtenson

It's the latter where I can't use foo←{text}

dzaima

@EliasMårtenson is anything preventing those just being ƒ left name right {body} and ƒ name {body}?

Elias Mårtenson

@Adám Of course. This is a hobby project. No deadlines.

Adám

3:39 PM

i have a deadline in 5

Elias Mårtenson

@dzaima Well, I have considered that. However, I also need to support niladic functions.

dzaima

@EliasMårtenson ah. my suggestion would be: don't :)

ngn

^ +1

Elias Mårtenson

@dzaima I don't right now, because the parsing rules are messy.

dzaima

@dzaima (fwiw chromium devtools format (function abc(){}) as ƒ abc(){} (and unnamed just don't have the name))

Elias Mårtenson

3:41 PM

I think the key thing I'm collecting suggestions about now is what "f" should be :-)

ngn

@EliasMårtenson if you're going to support {⍺⍵} anyway, why do you need other syntax for defining functions?

dzaima

@ngn because assigning them is impossible with ←

ngn

@dzaima well, then shouldn't that be fixed?

dzaima

@ngn 1) that's afaict hard; 2) there's not really that much benefit to reusing the syntax other than not introducing more syntax

Martin Janiczek

Hm, seems like I don't understand aspects of APL scoping rules...

dzaima.github.io/paste/…

I have a bug where assigning a value to a variable doesn't "write-through" to a dfn later

Yesterday I was talking with @Adám about dfns being lazy and tacit fns being eager, in a sense, but now dfns betray me too :D

dzaima

3:52 PM

@MartinJaniczek so something like a←0 ⋄ {a←1}0 ⋄ a?

Martin Janiczek

Probably. In my mind that would result in 1, not 0

which is not what Dyalog APL does, hmm

dzaima

@MartinJaniczek you want to do a⊢←1 then

Martin Janiczek

Can you explain this to me?

dzaima

@MartinJaniczek it creates a new local variable instead of mutating an outside one. When you're doing modified assignment, though, it's forced to mutate the value outside

@dzaima ⊢ there isn't special, it just uses the same syntax that a+←1 does. (and, of course, a←0 ⋄ {a+←1}0 ⋄ a also gives 1)

Martin Janiczek

@dzaima It indeed helped, thanks a bunch! And I learned something :)

Didn't expect this locality inside dfns

dzaima

3:57 PM

reasoning for this being that by default you don't want to accidentally be mutating outside state if variable names happen to clash

Martin Janiczek

I was almost considering doing a "clean" FP style where I just pass everything explicitly ...

Elias Mårtenson

@ngn Because Kap supports multiple arguments to a function.

dzaima

@MartinJaniczek fwiw iirc dfns were initially supposed to be a more "FP" alternative to tradfns

Elias Mårtenson

So I don't have a better syntax for something like: foo (arg1 ; arg2 ; arg3)

Martin Janiczek

@dzaima They do feel that way, there's just the rest of the language being mutation-heavy that betrays that

But as a concise lambda syntax, why not!

ngn

4:03 PM

@EliasMårtenson it could still have been something lambda-like, for instance {[arg1;arg2;arg3] ..}

i like lambdas. they are expressions. they can be used in other expressions.

Elias Mårtenson

4:19 PM

@ngn I have that. But lambdas are first-class functions and are values in their own right. you can do foo ← λ{⍵+1}

but note that foo 1 in this case is an array of two values: the function and the value 1.

ngn

@EliasMårtenson what i don't understand is why use one style for ≤2 args and completely different style for ≥3?

dzaima

@EliasMårtenson is that an actual 3-arg function or just destructuring of the right arg?

Elias Mårtenson

@dzaima A little bit of both. Yes, it's destructuring. But it's not destructuring an array. It's working on what is called a "list". It has the following syntax: a ; b ; c ; d.

So a multi-argument function works by destructuring a list, not an array.

ngn

"list"? what..

dzaima

^; i'll add with a "why"

Elias Mårtenson

4:24 PM

At the same time, the axis argument in something like foo[0;1;2] is simply a list too.

The idea is to normalise the otherwise odd syntactic array[index;index;index] in APL

dzaima

@EliasMårtenson I'd prefer to get rid of it but ¯\_(ツ)_/¯

Elias Mårtenson

if index;index;index was simply a single value, then there would be no need for special parsing.

Point is, people don't have to think about it, since default behaviour is APL compatible.

dzaima

; in [] is approximately as special as ⋄ in {}

Dyalog APL

@dzaima VALUE ERROR

dzaima

@DyalogAPL ಠ_ಠ

Elias Mårtenson

4:26 PM

@dzaima In APL yes. Not in kap

dzaima

at the very least I would not call it a "list". that is explicitly asking for confusion

Elias Mårtenson

In kap, you can do foo ← (1;2;3). The value of foo is now a scalar value which is a list (better name would be N-tuple) of 3 values.

Yeah, N-tuple is a better name I think?

dzaima

@EliasMårtenson i guess

ngn

@EliasMårtenson how is that different from a 3-vector?

can you do (1;2)+(3;4) to get (4;6)?

dzaima

@ngn just as much as a dictionary is different from a vector i guess

Elias Mårtenson

4:29 PM

@ngn No. As I said, N-tuples are scalars. You can't do much with it other than destructure it.

ngn

got it

Elias Mårtenson

But the biggest difference is in its parsing rules. It has lower precedence than any regular APL functions.

so 1+2 ; 3+4 ; 5 is a 3-tuple of values 3, 7 and 5.

This makes parsing compatible with array index notation in APL.

I thank you all for these discussions. Thanks to very astute observations from all of you, I have managed to implement several very good improvements.

You have to recall that one of the design criterions for KAP is that it should be possible to write code which is imperatibe and reasonably recognisable to people without APL experience. Sort of like a wa to get non-APL'ers interested by easiing them into it. A syntax where multi argument functions can be called with syntax foo (a;b;c;d) helps here I think.

dzaima

@EliasMårtenson that could just as well be done with foo (a b c d) and regular vector destructuring though

Elias Mårtenson

@dzaima Not quote. For example, foo (1+2 3+4) would give surprising results to the beginner.

dzaima

@EliasMårtenson I don't think that'd be that big of a problem

Elias Mårtenson

4:40 PM

@dzaima To you or me (or in fact anyone else on this channel) I agree with you.

But I have the tutorial that I intend to write in mind.

dzaima

or you could make a proper vector notation to use in place of strands

Elias Mårtenson

@dzaima Very true. But then I would be remaking R. :-)

What I really wanted to do was to fulfil three goals of mine: 1) Make an APL, 2) Change some things that I dislike about APL. ⎕IO being one of them, and 3) Make it possible for beginners in particular to write imperative code if they can't think of a nice array based solution.

The last is to encourage beginners to write actual software and not juts playing around.

dzaima

also with headers, ƒ foo (a b c d) {(a+b)-c+d} vs foo←{∇a b c d: (a+b)-c+d} isn't that bad either (and teaching that is quite simple - regular assignment, ∇ is recursion and thing before : is a sort of self-description of how to call itself)

ngn

@EliasMårtenson imperative is not the opposite of array-based

Elias Mårtenson

@ngn We may use different terminology, but I hope you see what I mean.

@dzaima You are absolutely right, and I'm open (nay, committed) to changing the function declaration syntax. And I'm absorbing all the ideas I get from you.

dzaima

4:46 PM

@dzaima (though i guess : might not work if there are guards)

ngn

@EliasMårtenson the opposite of "imperative" is "declarative". for the opposite of "array-based", i suggest the scientific term "loopy" :)

dzaima

@ngn i'd guess more specific would be "branchy"

ngn

@dzaima i'm not sure what you mean

even pure array code can make use of some branching

dzaima

@ngn in converting JS/C/Java to APL, ¨ takes care of 99% of loops. But translating ifs is often much more painful

ngn

@dzaima because apl has guards instead of the more general cond?

dzaima

4:53 PM

@ngn that mostly, but even cond is, imo, very ugly

ngn

right, it should be used sparingly. so we agree :)

dzaima

@ngn it should be used sparingly doesn't work if you explicitly want to allow it

Elias Mårtenson

@dzaima Kap has both if and while, both implemented in the library using defsyntax. That's one of the main features I'm hoping will make simple code easier to deal with.

dzaima

as far as i'm concerned, if/else just doesn't work with APL syntax. You either just don't have it (and make it painful to write certain type of code), or have a different syntax for it (tradfn :-things, KAP's predefined macros)

Elias Mårtenson

6:03 PM

Arguably one more purpose of KAP is for me to have a platform on which to experiment with these things.

Not just me, a lot of the things I've experimented with comes from you.

user

7:52 PM

I wish you could use guards in parentheses too - (cond: foo ⋄ cond2: bar ⋄ baz)

dzaima

@user that's a nice idea, though it breaks proposed namespace literals - (key1:value1 ⋄ key2:value2)

user

@dzaima Oh, that's too bad. What are namespace literals? Are they like JS's object literals?

dzaima

@user same idea, yeah. (dzaima/APL has them)

as a short-hand alternative for m←⎕NS⍬ ⋄ m.a←2+2 ⋄ m.b←3

user

Cool!

dzaima

(part of Array Notation)

Dyalog APL

9:43 PM

<moon-child> @user my apl does support separators in parens (e.g. (foo⋄bar⋄baz) ←→ ((foo)⊢(bar)⊢(baz))). But guards don't really make sense there; what if you have a guard inside of a set of parens inside of a dfn?

<moon-child> or, for that matter, nested parens

<moon-child> (my notation also conflicts with the proposed array notation; but I prefer it this way for consistency. May reconsider later)

user

11:43 PM

@Moonchild The same as nested if-else statements in other languages, I guess

Like {⍵>3:(⍺=⍵:⍺ ⋄ ⍵) ⋄ ⍺=⍵: ⍵ ⋄ ⍺} instead of something weird like {⍵>3: ⍺{⍺=⍵:⍺ ⋄ ⍵}⍵ ⋄ ⍺=⍵: ⍵ ⋄ ⍺}

Transcript for