alternative 26

@Bubbler nice. this and that combined make 25: ⌽(⊢,⍨⊃-≢ׯ1*⊢∊⍨0⌈⊃-≢)⍣⎕⊢0

random pointless idea: in BQN modifier names start with ¯ to indicate they're superscripty

@dzaima Instead of _, you mean?

@Adám yeah

Then compositions (dyadic operators) should almost have an #x1F907 prefix :-)

Btw, there's nothing that prevents using ¯ like a digit in names. APL2, APLX, and GNU APL allow this.

Anyone here familiar with EBNF and railroad diagrams?

read the hopl paper last night, extremely informative.

Cool. Feel free to tell the authors.

maybe! i need to reread a few sections. it did a lot to clear up implementation and parsing details.

One of my favorite aspects of (learning) APL is the extensive documentation of its history

Both technical and cultural

@Adám maybe a teeny weeny

can I help?

@RGS I'm trying to specify the new array notation, and I'm almost there:

value     ::= expression | list | block | space
list      ::= "(" separator? value (separator value)* separator? ")"
block     ::= "[" separator? value (separator value)* separator? "]"
space     ::= "(" separator? (name ":" value (separator name ":" value)*)? separator? ")"
separator ::= ("⋄" | #xA | #xD | #x85 )+

@Adám ok, and what seems to be the problem? are there phrases that the grammar doesn't capture yet?

@RGS The only issue is that for list and block, there must be at least one separator, whereas the above grammar allows having none:

(btw I'd probably call the last rule separatorS instead of separator)

@Adám i don't think there's a way around it other than having multiple |ed expressions, each requiring at least one separator in at least one place

@Adám in what position(s) must/can that separator pop up?

@dzaima Hm, then maybe it is easier to simply state that a list or block without a separator is simply a strand or indexing/axis.

@dzaima (that may or may not depend on exactly what syntax that's using, as from the wikipedia page it doesn't seem like EBNF)

@RGS Any of the show. (separator value) or (value separator) or (value separator value) but (value) isn't valid.

@Adám Just put ()? around value (separator value)* separator? for those rules, right?

Wait, that's the wrong way around. Never mind.

in ANTLRish syntax a way would be to have a "(" expression ")" alternative beforehand to catch no-separators before it gets to touch them, but i'm not sure there's any formal specification of that

@dzaima Yeah, I thought about that. That's basically the trick I often use with ⎕R/⎕S.

@Adám I can't come up with something pretty that allows AB, BA and ABA without allowing B; maybe @dzaima is right and you have to go with |ing several options

@RGS Same issue as with regexes, I guess.

@RGS there definitely (probably) isn't any way if that example contains all the allowed syntax

Yeah, you pretty much need `"(" (s v? | v s | s? v (s v)+ s?) ")" with BNF.

But I don't think there's anything wrong with saying it has to contain at least one separator in the text.

@Marshall I don't think I need three choices. ABA?|BA should do, no?

list     ::= '(' ( ( value separator )+ value? | ( separator value )+ ) ')'

@Adám how about BAB (aka the basic (1 ⋄ 2))?

@dzaima Ah, you're right.

@Adám That's missing plain separator.

OK, but this should do:

list     ::= '(' ( ( value separator )+ value? | ( separator value )+ separator? ) ')'

@RGS Why? Multiple collapse to one.

@Adám oh just because the rule had the +, so it was more of a plural because of the way the rule looks and not because of how the syntax really works

I though multiple separators had some meaning; I don't know the new array notation yet

@RGS (1 ⋄⋄⋄ 2) ≡ (1 ⋄ 2)

@Adám seems correct

Yeah, not too ugly either:

Thanks, guys!

@Adám Is a lone separator for an empty list not allowed?

@Marshall No, since it doesn't specify type, and anyway, () is an empty space.

space    ::= '(' separator? ( name ':' value ( separator name ':' value )* )? separator? ')'

I guess that can't be simplified.

CMP: What would you think of multi line/statement trains? +\(-⌿'{}'∘.=⊢) could be written as ('{}'∘.=⊢ ⋄ -⌿ ⋄ +\) or

(
    '{}'∘.=⊢
    -⌿
    +\
)

Certainly be easier to parse for neophytes

They could enhance clarity, but ... it sorta makes them a third function definition syntax...

(MLTs, dfns, and tradfns)

@JeffZeitlin Where are the current trains in that list?

@Adám i think the reverse might be clearer (there was a time when i thought this was the behavior of dzaima/APL, but it wasn't)

perl's /x modifier for regexes allowed them to be broken up like so, so that you can add a comment to the right of each significant component.

@xpqz Right, the ability to comment the parts of a train would be nice.

@Adám +← 1

@dzaima though for multiline the proposed makes more sense, hmm

@dzaima The reverse is already obtainable with (…)⍤(…) e.g. (+\)⍤(-⌿)⍤('{}'∘.=⊢)

@Adám that's a lot of characters though. i guess i'm looking at it more like a monadic function fix than different programming style

Can someone help me understand what the problem with this code is, for this codegolf challenge?

@RGS f⍨⍣g → f⍣g⍨ (there might be better alternatives but this is the simple one)

oh

so what was I doing, then..? was I "resetting" the variable in some way?

@RGS Lots of golfing opportunities :-)

@RGS you were creating an outer product 2*(|l)+n-1 times, but you don't want the 2*

@dzaima oh I see

@dzaima (that "better alternative" turned out of the same length - ,⊃∘.,/((|l)+n)⍴⊂⍳k; i've often wanted a built-in for something like ⊃∘.f/N⍴⊂X)

@Adám I am sure there are :P I was just trying to get it to work correctly first D:

@dzaima (and the need for 1/ later comes from these alternatives actually being wrong & bad)

@dzaima and those are wrong and bad because 1,2 should really error but people have decided to special-case it

@dzaima It'd be really awkward to construct vectors from scalars then, no? Or would stranding always suffice?

@Adám 1) we have strands (and soon array notation); 2) have you ever actually had a case where you can't strand (or no other array creation alternative is better) but are absolutely guaranteed ≥2 items?

(using , to build arrays is only "correct" when either you have simple scalars and 2+ items, or you have a starting ⍬/''/⍮X)

@dzaima 1) Yeah, allowing scalar,scalar might well be a hold-over from before (generalised) stranding. 2) Yeah, trains.

@Adám 2) do you mean cases like this ;)

@dzaima No, like lexicographical "≤" for scalar characters: 't' (</∘⍋,) 's'

@Adám ⍮ :)

@dzaima I was just about to say that that's the general form, but without that primitive, not allowing the concatenation of scalars would be very problematic. You couldn't even define ⍮←,⍥⊂ then, but would have to resort to a dfn {⍺⍵}.

@Adám yeah, i accept there are historical reasons for it. Question is, is there absolutely any reason to keep it around in modern APL

@dzaima Modern∊dzaima Extended?

@Adám - Good point; arguably, I should have said fourth - trains/tacits, dfns, tradfns, and MLT/MSTs.

@Adám Extended needs backwards compatibility. Also, BQN

@JeffZeitlin I would counter: trains/tacits, dfns, MLD/MSDs, tradfns, MLT/MSTs.

@dzaima dzaima and BQN, right, not necessary. (Dunno what to do about Extended, as it only extends 17.1, not 18.0…)

@Adám unfortunate TIO doesn't have versioning. otherwise i'd definitely just change ≠ to match 18.0 and move on with that

@dzaima But my monadic ≠ is just so useful, whereas I've probably never needed my monadic ⍧

@Adám - Point taken; I didn't view dfns and MSDs as different because they both existed when I first became aware of them. But the proliferation of function definition syntaces is the issue.

One might argue that unparenthesised trains shouldn't have been allowed in the first place. I think NARS2000 prohibits them. And then the extension to multi-statement wouldn't be so controversial, I think.

@Adám Interesting distinction between powerful and useful/convenient there. Nub Sieve is powerful since it's otherwise difficult to do correctly and impossible to do efficiently, but 0≠ comes up so much more often.

@Adám - Perhaps not; I'm not entirely convinced that tacits are really a good thing to start with.

Their decomposition is counterintuitive, their syntax seems somewhat not-APLish.

@Marshall Sure, and when Extended does have nub sieve…

Along those lines I've been thinking of adding monadic ⍳∘≢ and even ≢∘⍴ (APL spellings) to BQN.

Yeah, those two are all over the place.

there have been ±5 separate times i've considered adding ⍳∘≢ to dzaima/APL; a couple ended with no symbol found, a couple with worries about ⍳∘⍴, and a couple with falling asleep

Hope it's +5 and not -5.

@Marshall initially i wrote "probably more than 5", then "~5" and then ended up with ±5. :p

@RGS Sorry for the lack of a response, I was getting some work done early Saturday morning before the grandkids arrived.

@dzaima i do have ϼ free now though, since ⎕VI is mostly stable

@Marshall Symbol suggestions for BQN: ⧣ and ⧥ (indices / overlaid on count ≠ and shape ≢)

@dzaima Somewhat related to ⍳∘≢ versus ⍳∘⍴, another problem I've been having is trying to figure out what kind of return value Indices should have. BQN2NGN keeps the inconsistent Dyalog behavior.

@Marshall i think the logical thing is to have ⍳,1 be ,⊂,0, and 0 1≡(builtin-of-⍳∘≢)2 3⍴0 might not be that bad

@Marshall Yeah, distinguish between scalar (number‽) and vector arguments.

I mean Indices, APL ⍸ or BQN /.

There you don't have the option as the argument shape is always a vector.

Oh.

@Marshall i don't think distinguishing on rank is that bad of a crime. (personally i always code with consistent, known rank everywhere so i've had no issues)

If you want Replicate (/) to be /⊸⊏ then it has to only allow a vector argument.

my ⎕VI←1 ⍸ has a much bigger problem though (gtg for bit)

@dzaima It's a way to use the same symbol for two different functions. Eventually, you will want one of the functions in a context where you can't use it.

There are some reasons to force vector indices everywhere. ⊏ would be able to do scatter-cell indexing easily: if the argument is a numeric array, its rank must be at least 1 and its rows are cell indices. If it's a vector of numeric arrays, then you take the cartesian product of rows from all index arrays.

@Marshall if i want specifically a vector of 1-item vectors of simple scalars, 1) i'm probably doing something stupid; 2) ,¨ is not that hard to use (similar logic for me special-casing f/vec)

(can't comment on ⊏/⌷, i've never had any reason to use it beyond the simple case of a vector left arg)

@dzaima That goes the other way too. If you have scalar indices, just use ⥊⎉0⊸⊏. Is the power of vector indices worth making scalar indices inconvenient?

@Marshall in my case, yes, i believe i've used ⍸ on vectors much more than any other case

unrelated, ⎶ might be a good char for ⍮ ({⍺⍵}/{,⊂⍵} duo)

@dzaima Pretty good, but I'm sticking with ≍○< for now.

@Marshall Not ∾○< ?

@Adám ≍○< works with both valences since monadic ∾ is Join/Raze.

@MortenKromberg don't worry, I had nothing clever to say; was just curious about what could've brought you to the APL Orchard!

@RGS He tries to be here when he can and remembers. (I just have the tab pinned.)

@Adám thanks to a conversation between you and dzaima I found out about the wonders of pinned tabs... now I keep a couple of tabs pinned for good measure; this one included

TA..?

TopAnswers.xyz

I should keep gmail pinned as well... currently I have google calendar&tasks, this chatroom, whatsapp web & cookie clicker :P

is TopAnswers taking off?

Some communities are, Code Golf not really. We have a closed beta APL one in the works with some amazing custom features.

@Adám an APL community in TA? cool!

(btw I'm guessing one does not simply ask for access for that closed beta?)

@RGS I guess you could, but there's not much going on there in terms of content yet. However, I can tell you the cool features:

@Adám what is the point of having cool features if there's no one to explore and use them? ⍥

@RGS Well, before we open it up, we have to decide on the scope.

@Adám of course; just let me know if you need any proofreading ;)

@Adám these sound really nice! What do you mean by "real" APL syntax colouring?

@RGS TA uses the same code as RIDE.

unlike some other low-quality syntax colouring systems..?

Yeah, like SE's.

@Adám ok I get it. The windows interpreter doesn't have much colour either, so I am used to coding in APL being a pretty monochromatic experience

@RGS Did you customise your colours?

Mine looks like this:

Or like this, when I'm feeling light:

@Adám uh oh, no I did not!

an adventure awaits me, I see

I have been meaning to ask this for a long time now; when I press CTRL+BACKSPACE the interpreter inserts this: 

Ok I cannot get the chat to display the character

why doesn't CTRL+BACKSPACE erase a whole word, like in any other text editor?

@RGS That's U+FF, I think.

yeah, when I hit CTRL+BACKSPACE to delete a whole variable name or a whole sequence of consecutive chars and I see that character instead, I do go "uff"

so I guess it adds up

@RGS Because it isn't a text editor. If you want an interface that behaves like a proper text editor, just use RIDE.

@Adám not even the "text editor" window that opens up when I type )ed someName?

There are lots of oddities that show that the IDE isn't a text editor: Pressing delete at the end of the line doesn't bring the next line up, you can position the cursor beyond the end of a line, etc. etc.

@Adám yup I did notice those...

@RGS That's a function editor, not a text editor.

ah ok, thanks for the clarification!

Doesn't mean I agree it should be like this. I'd want a more text-editor-like IDE too.

This is where readline comes in handy: "As a cross-platform library, readline allows applications on various systems to exhibit identical line-editing behavior." That's also what I like about my current lisp implementation.

@Wezl Yeah, unfortunately for Dyalog, not changing existing behaviour has the side effect of scaring off newcomers.

... but readline doesn't seem to have control backspace itself :/

I never knew about Ctrl+BS. I always use Ctrl+Shift+Left,Del.

@Wezl Nevermind, the nice features readline might offer would probably conflict with typing special characters (CTRL+E for end of line would mess with typing epsilon, for example)

@Adám what is this for?

@Wezl Eh, I never use Ctrl for APL characters (I use AltGr/RightAlt), so I have Ctrl+A for select all, Ctrl+S for save, etc.

@RGS Delete word left.

@Adám ah ok, seems so elaborate when you could just do ctrl+bs

@RGS I don't really think about it, and it is consistent with deleting a word to the right.

@Adám sure, consistency is nice

@Wezl it uses emacs bindings by default (as you probably know), but is configurable

"\b":unix-word-rubout in .inputrc