The APL Orchard

@EliasMårtenson What definition of X⊂Y are you aiming for? APL2? Old Dyalog? New Dyalog?

APL2.

Basically Dyalog's behaviour (with ML=2)

OK. (Though that is one very strange definition.)

@EliasMårtenson 0 0 1 1 1⊂"Hello" fails. Should give ,⊂"llo"

I mean, I'm not married to a given definition. I thought they were all mosstly the same. The ISO spec doesn't actually include it (the document seems to be in error, as it includes the monadic version of ⊂ where the dyadic form should be), so my implementation i sbased on reverse-engineered GNU APL.

@EliasMårtenson Oh they are very fundamentally different. See apl.wiki/Partition vs apl.wiki/Partitioned_Enclose.

10:03 AM

@Adám Thank you. Exactly the feedback I was hoping for. I've added your example as a testcase.

@EliasMårtenson NP.Let me know when to try again.

Yeah, that's why I was saying ML=2. I believe ⊆ is called ⊂ in that case.

Doesn't ML=2 make it behave like GNU APL in this respect?

@Adám Thank you for the testcase. I fixed the problem. It was in exactly the section of the code that I was the most concerned with.

10:19 AM

@EliasMårtenson No, in ⎕ML≤2, ⊂ is Dyalog's Partitioned Enclose and ⊆ is APL2's Partition. In ⎕ML=3, both ⊂ and ⊆ are APL2's Partition.

Oh I see.

I'm using ⊂ for APL2 partition. I will later implement ⊆ to be the same as Dyalog's ⊂.

@EliasMårtenson Oh dear, that'll be so confusing.

Why not use ⊂ for Dyalog's and ⊆ for APL2 to at least agree with one such scheme?

Or even better, leave out both definitions and instead use a more sensible definition, like partition lengths or "Group"?

@Adám I kinda agree, but I really want ⊂ to be like GNU APL. That leaves me with ⊆ to do something interesting with.

Now, until an hour ago I didn't even know about Dyalog's special version of ⊂. Are you saying it's not very useful?

@EliasMårtenson Both of them have downsides. Dyalog's cannot omit items, and APL2's cannot insert empty segments. Also, APL2's left argument is very odd. Neither is able to "group items, e.g. 1 1 2 1 2⊂'ABCDE' giving 'ABD' 'CE'

@EliasMårtenson Recommended reading: apl.wiki/Partition_representations and BQN's group.

10:37 AM

@Adám OK, I'l stop referring to APL2 here then because that sounds just weird. GNU APL returns "AB" "CD" "E", which seems reasonable.

@EliasMårtenson Maybe I shouldn't have used the ⊂ symbol in my example. Do you not find it strange that 1 1 2 1 2 and 1 1 3 1 2 and 1 1 2 2 3 and 7 3 6 2 5 are quivalent?

@Adám Not really, no. Perhaps that's because I'm used to that definition.

(Hm, I just noticed that Dyalog's ⊂ can easily be extended to allow omitting items. A ¯1 in the left argument could mean "skip".)

Hmm... Interesting. I could easily add that as well.

@EliasMårtenson What do you mean? APL2's ⊂ can already skip items.

10:43 AM

Sorry, I was thinking of reordering

Basically being able to partition "ABC" into "AB" "BC"

How would that work? Whence the duplicated item?

@EliasMårtenson That's not a partitioning.

@Adám True. I was just thinking of uses for negative values. I'm not even sure how useful that would be.

Imo, one should use / (Replicate) first for that.

@EliasMårtenson in agreement with Adám: oh god no

@EliasMårtenson imo the "start new partition on positive delta" is stupid. I don't think much would be lost if you started a new partition on any difference

10:59 AM

@dzaima That's a fair argument and I don't have any real preference either way. I'm not sure I've ever taken advantage of the "change in the positive direction only" definition.

@dzaima (recent)

11:20 AM

@EliasMårtenson and @xpqz Would you be interested in doing APL consultancy work for a large Swedish medtech company who currently has about 6 APL programmers already in the team. They accept remote work, but require APLers that know Swedish.

Flattered that you'd think I'd be good enough to do this for real.

o/

11:45 AM

Trying to use @ but to apply to an axis or the other

i.e. x@1⊢2 2⍴⍳4 is fine, but how to change the other axis?

I guess I could pass it a function

@Adám I'm flattered that you think I'm skilled enough to be able to help out. :-)

tio.run/##SyzI0U2pTMzJT///v/xR2wT1/LLUorSc/…

OK, that's a start...

thinking about that oXYoX... but maybe there's a simpler way to mirror

12:01 PM

@Ven You can either use ⍉ before and after or combine with ⍤k

@Ven Welcome back. Long time, no see.

@Adám Thanks :). It has.

well, I thought about using ⍤ but I wasn't sure how to intersperse ⍉

@Ven Use ⍉ to move the axis you want to index on so that it is the leading axis.

ah, that's what you meant. Yes, that's what I ended up doing.

@Ven E.g. ⋄ ⍉'x'@2⍉3 4⍴⎕A

@Adám
┌→───┐
↓AxCD│
│ExGH│
│IxKL│
└────┘

12:03 PM

Same as ⋄ 'x'@2⍤1⊢3 4⍴⎕A

 @Adám
┌→───┐
↓AxCD│
│ExGH│
│IxKL│
└────┘

That line is the one I'd like to change ⎕←w@p⊢⌽w@p⊢⊖w@p⊢⌽w@p⊢m@(≢w)⊢⍉m@(≢w)⊢x

@xpqz @EliasMårtenson Flatter aside ;-) are you interested? I'm sure they'd pay for upgrading your APL skills if necessary.

@Ven You can begin by removing all ⊢s, I think.

@Adám I need to change a few to ` `s, no?

@Ven Spaces? No, why?

12:05 PM

i.e. @p⊢m@

@Ven Oh, that one needs to stay.

it needs to stay in m@(≢w)⊢x too

⎕←w@p⌽w@p⊖w@p⌽w@p⊢m@(≢w)⍉m@(≢w)⊢x is fine :)

@Ven True.

I was too quick in saying "all".

⎕←{p←2/¨⍳≢⍵⋄m←⍵,1↓⌽⍵⋄m←⍵,1↓⌽⍵⋄⍵@p⌽⍵@p⊖⍵@p⌽⍵@p⊢m@(≢⍵)⍉m@(≢⍵)⊢' '⍴⍨2⍴1-⍨2×⍨≢⍵}'overflow'

@Adám emailed you, but fun as it sounds, it doesn't quite fit in with current life/work reality

12:11 PM

@xpqz Sorry, I didn't see the email until now.

mh, I don't remember how to use the bot it seems

{p←2/¨⍳≢⍵⋄m←⍵,1↓⌽⍵⋄m←⍵,1↓⌽⍵⋄⍵@p⌽⍵@p⊖⍵@p⌽⍵@p⊢m@(≢⍵)⍉m@(≢⍵)⊢' '⍴⍨2⍴1-⍨2×⍨≢⍵}'overflow'

⋄{p←2/¨⍳≢⍵⋄m←⍵,1↓⌽⍵⋄m←⍵,1↓⌽⍵⋄⍵@p⌽⍵@p⊖⍵@p⌽⍵@p⊢m@(≢⍵)⍉m@(≢⍵)⊢' '⍴⍨2⍴1-⍨2×⍨≢⍵}'overflow'

@xpqz 0x0.st/-Xs1.txt

Nice

ah, thanks

OK, now to find the TIO header again :).

12:15 PM

@Ven This?

@Adám thanks!

oh, could I use dyalog extended's "over" for what I want?

The q r → r q r part would be what I want, I think

@Ven Wait, what are q and r?

@Adám github.com/abrudz/dyalog-apl-extended/blob/master/…

where q would be ⌽, and r would be ⍵@p

@Ven That's talking about Depth.

ah, mh.

then I'm unsure.

maybe something like J's `?

12:27 PM

<kritixilithos> maybe you want to do something with ⍣ ? like perform @-substitution and rotate/flip some constant number of times

that's been suggested earlier, but I don't know how to have the rotate/flip apply inbetween

and I'm thinking maybe generate a quarter of it and mirroring the rest would be easier

Maybe there's a shorter way to mirror than ⍵,1↓⌽⍵ too

@EliasMårtenson Just got confirmed that they are willing to provide additional training if necessary.

mh, tryapl lost its examples?

@Ven Uh, no. What do you mean?

@Adám Damn it. You're putting me in a difficult situation. It's definitely something I would want to do, but I don't think it's practical. My current job is a full-time job, and I'm currently paid rather well by Swedish standards, so I wouldn't be able to quit.

12:35 PM

@Adám When I hover something in the primer I don't see examples anymore

@Ven Oh, just click.

That means that I wouldn't be able to allocate enough time for this job.

Ah!

<kritixilithos> @ven (⌽∘⍉⍵@p)⍣4(⍉m@(≢⍵))⍣2

<kritixilithos> as a first step

Thanks, will take a look

Down to 53, (⌽∘⍉⍵@(2/¨⍳≢⍵))⍣4(⍉(⍵,1↓⌽⍵)@(≢⍵))⍣2⊢' '⍴⍨2⍴1-⍨2×⍨≢⍵, now I should probably try to make it wholly tacit

12:48 PM

@EliasMårtenson "it is a well compensated role". If you are interested in finding out more, I'll email you an email address to contact for a concrete offer.

Razetime

1:04 PM

@Ven It's generally very hard to make tacit functions with @

@Razetime Only if the right operand cannot be computed from the right argument.

Razetime

I'd love to see an example

@Razetime I know, I know

1:50 PM

how do you get the values out of a json object you get from doing ⎕JSON?

aha, ⍎ works, knew something like that worked

@rak1507 for dynamic values, yeah. For constant values, just obj.key works

yeah, I knew about that, but need a dynamic value

@Adám Thanks. I'll let you know shortly.

2:33 PM

@Adám what's the incantation I have to type (in ride) in order to edit a new namespace script?

@xpqz aplcart.info?q=edit%20a%20new%20namespace

rtfm

RikedyP

3:29 PM

@Marshall Recently did a google search of "⍳" it was very cool

Announcement: 2021 round of APL Problem Solving Competition has begun. Referral awards offered.

7

@RikedyP What got me was when I looked for "APL transpose" and found the BQN docs above the fold (but APL Wiki's higher up).

@Marshall What do you think of apl.wiki/Timeline_of_array_languages?

@Adám The overlapping categories are probably pretty confusing. The idea of the timeline for influential languages is that it's more of an overview, so it should stay separate.

@Marshall Feel free to edit mercilessly.

RikedyP

3:45 PM

@Marshall what a time to be alive

@Adám wow no really?!! :P

@RikedyP But we'll know we've reached the real pinnacle of history when every search term only returns BQN docs. Soon!

2

BQN isn't mentioned in the timeline of array languages page.

@EliasMårtenson The table right now is just the one from Timeline of APL dialects. See discussion.

4:07 PM

Ouch. I just found a case where KAP is terribly slow, and I need to find out a way to deal with it. Doing ,/foo where foo is a long list (50000 elements in my case) of rather long strings.

(well, not that long, a few tens of characters)

As the reduction works its way through the list, there is a ridiculous number of allocations and reallocations of strings, and since strings in KAP are represented by lists of boxed 64-bit integers, there is a crazy amount of object allocation/deallocation happening.

@EliasMårtenson ,/ is special-cased in dzaima/APL fwiw. With immutable arrays that's like the only sane thing to do

@dzaima Yeah, I tried that, and replacing the , with a function made it much slower in Dyalog. But it's still nothing compared with the slowness of KAP here. I need a specialised string array to handle these cases. But... specialised arrays are clumsy and requires a lot of ugly special-cased code.

(and BQN has the ∾ built-in for Raze, which is APL's ⊃,/ for vectors of vectors, but also does cool things with higher rank arrays)

@EliasMårtenson shouldn't ,/ on a vector of strings be about as slow as ,/ on a vector of vectors of numbers then too?

∊ does the same thing for vectors of vectors

@dzaima (it's over a third of the whole / impl)

@EliasMårtenson A special-cased ,/ would only need to allocate one array/object

4:32 PM

@EliasMårtenson also somewhat sufficient for many cases (and reasonably simple to implement) would be to cache ASCII character instances, and then you'd only need to transfer pointers around

Razetime

4:42 PM

@dzaima Hey it's me

lol

5:06 PM

ermineii.github.io/paste/#07ZhLbuMwDIavInTVAAV4hmwM/…

some more BQN fan art, this time to advertise my animating tool :P

I'm still waiting for a BQN port that runs natively on the front of UFOs

ironically, I did the transposing in jq, because I forgot that BQN has multiline strings

CMP: although it is domain-specific, would you consider jq to be an array programming language?

@Wezl Wait how do I stop myself from pressing play every time it gets to the end?

@Wezl A few folks at work did most of last year's Advent of Code in jq.

@Wezl doesn't look like one to me

@Wezl Is it all right to put this link and the previous ones in commentary/fanart on the BQN site? I guess I have to ask since I can't really do it without reproducing them.

https://ermineii.github.io/paste/#07ZhLbuMwDIavInTVAAV4hmwM/Cujm26qQBco0gFmk@MPqYdFSZbtboJMKsFxYPK3HuRHR87ni1maq5q9Li4yVbM2exfn2blz64aWWRuNrR/6vOKezWzMpeN2Xzxh1cd3rSBZngyfJFQrQOyVD5KkGYY9NzlXApq6LQZCx0@GcHldS7SKGIM4/laLgqpMBPDlTkCdmFdTqWAqlTlxXykXSWbtNFExZAgS614vWWYmtuR5fJ5iKK11G41e3gZqd0aNfAgK1lrUINM1JWstajwPRs0Am6ixm1ErVSuoMWlmmkzJWouakGY0awO1R0WNYgQ0aw1qTFoQFrA1rEFWYR1MwVEDGzuDqsStoQ3pqsCtoS3ApnEbtD0obbTkV@FW0yawRaXGraZNpiEqLwS6uInLy2oua94QhxSdJq4GLvGmiBvA/ffAiWCXN57OEdw8Skdpuw3afh1t4T3ikXHzT8zB21PwRpE39mwChwQcNoFDAg47wIUNIxNHx4ibBnHPQlz4p8TaXeQQkMMOcgjIYR@5sNW70WDu1zHH87X2CHMQ5rDLHI4zJ@…

5:16 PM

@Wezl Mind doing same for one or both of the static ones? 0, ...

...1.

my case for jq:
it has arrays, but these are more like APL's boxes. the core of jq works on streams.
while map(filter) has to be manually called on arrays, a filter, like a function, auto-maps (it doesn't *exactly* have scalar extension, but the filter . + 1 will increment everything in the stream). It is functional, without even re-assignment, and it supports "operators" because user-defined functions/filters can take other filters as arguments.

@Marshall 0

@Marshall 1

5:31 PM

@Wezl no scalar extension disqualifies it in my mind

it still kind of does have it

@Wezl You're live!

can you link to codeberg.org/wezl instead of github?

@Wezl Got it.

thanks

5:59 PM

twitter.com/SPIDERCLlMB/status/1327111382928482304?s=20 somehow I don't think this tweet is related to BQN the programming language

6:18 PM

@rak1507 Never said it was.

But the gesture depicted is the official hand signal of the BQN programming language, in case anyone's wondering.

(you don't have to wink)

lol

6:57 PM

@dzaima Yes, and it is.

But strings is already a special-cased array of unicode codepoints. The problem is that once they are cocnatenated together, they lose that special status.

I just added an optimisation that preserves that for strings and the test case dropped from 12 seconds to less than 4.

But it's only special-cased for concatenation. If I were to, say, reverse it, the result will be a generic array with boxed values and it'll be just as slow again.

I need to come up with a more generic scheme.

@EliasMårtenson in dzaima/BQN I made a temporary mutable array class which allows copying in subarrays & stuff. Though that doesn't work for reverse, as it needs to be able to copy sequential data to be fast

7:52 PM

cmc: find a way of outputting the nth term of the sequence 4 10 4 9 19 39 79 159 319 639... without having a single number in your code (or any cheats to get a number like ≢'abc')

the 0th term is 1 if you want to use ⎕IO←0

what is this sequence?

that's part of the fun!

oeis.org/… has results, but no results when it starts with 10 or 4 10

Martin Janiczek

I mean, if I could ignore the first 4 numbers then I see a pattern :D

also "cheats to get a number" you need to get a number somehow, so what makes a way a cheating way?