« first day (1357 days earlier)      last day (1313 days later) » 

jimfan
jimfan
1:12 AM
I have been reading about vintage computer and calculator lately, and I attribute this "clicking" on me to APL language. For example: youtube.com/watch?v=_DTpQ4Kk2wA
To some people it might dumb to programme on a command line via a teletype writer, but it appears to me as a very pure form of computing, which is beautiful in itself.
 
Bubbler
Bubbler
@TessellatingHeckler How did you get it? I just got two versions of 12's
 
TessellatingHeckler
TessellatingHeckler
@Bubbler not sure how to answer that without giving it away; I wrote it out and got a 13 with echo"0] and then used a trick from "Tips for programming in J" from StackExchange to get 12, and then looked at it thinking "11 is impossible" and then saw a way to make it 11
 
H.PWiz
H.PWiz
In J, a composition takes the rank of it's second operand (or something like that)
 
TessellatingHeckler
TessellatingHeckler
 
Bubbler
Bubbler
Oh damn, how did I forget that
 
 
3 hours later…
Razetime
Razetime
4:29 AM
{⍵/⍨((1=≢∘∪){+/,⍵}⌺2 2)¨⍵}{3 3⍴⍵}¨↓pmat 9
Is this function correct for getting all 3x3 sturdy squares?
26
Q: Print all 3 by 3 sturdy squares

Calvin's HobbiesA sturdy square (akin to a magic square) is an arrangement of the integers 1 to \$N^2\$ on an \$N\$ by \$N\$ grid such that every 2 by 2 subgrid has the same sum. For example, for \$N = 3\$ one sturdy square is 1 5 3 9 8 7 4 2 6 because the four 2 by 2 subgrids 1 5 9 8 5 3 8 7 9 8 4 2 8 7 2 6 al...

code-golf number arithmetic number-theory grid
 
eyepatch
eyepatch
4:46 AM
p←'(()(()))' ⋄ ⌈/+\('('∘=-=∘')')p ⍝ 3 as expected
f←⌈/+\('('∘=-=∘')') ⋄ f p ⍝ Domain error?
Why is f not substitutable with its matching expression?
 
Elias Mårtenson
Elias Mårtenson
I have a probably stupid question. If I have an array, and I want to "extend" it along a new axis, what is the most concise way of doing it? Let's say I have a rank-2 array of the following form: 3 3 ⍴ ⍳9 and I want to make a rank-3 version by creating N "layers" of the same array on top of eachother.
The best I can think of is to do ((N,(⍴A)) ⍴ A. And if I want to extend along a different axis, apply the appropriate ⍉ operation afterwards.
 
Bubbler
Bubbler
@eyepatch You have a 3-train ('('∘=-=∘')'), then monadic application of +\ then ⌈/. But your f treats ⌈/ and +\ as part of an outer 3-train, so f p evaluates to (⌈/p) +\ (...)p
To make both ⌈/ and +\ into atops, you can write f←⌈/(+\'('∘=-=∘')')
@EliasMårtenson One way would be N⌿↑,⊂A I think
 
eyepatch
eyepatch
5:02 AM
Interesting. Obviously the atop solution is best here, but is there a way that I could switch back into monadic application?
 
Bubbler
Bubbler
@eyepatch No way in a pure train.
 
eyepatch
eyepatch
I would have to wrap the 3 train in a dfn or something?
 
Bubbler
Bubbler
You could add a wrapper dfn for a chain of monadic application, such as {⌈/+\⍵}'('∘=-=∘')'
 
eyepatch
eyepatch
Huh. Ok.
 
Elias Mårtenson
Elias Mårtenson
@Bubbler thanks. I'll try that.
@Bubbler hmm, that didn't work. it returns a rank-1 array with each element duplicated N times (if I set N=2, then I get an array with content 0 0 1 1 2 2 ...etc)
 
Bubbler
Bubbler
5:11 AM
@EliasMårtenson Which implementation of APL are you using?
It works in Dyalog APL at ⎕ML←1
 
Elias Mårtenson
Elias Mårtenson
5:31 AM
@Bubbler I'm testing in GNU APL
 
Bubbler
Bubbler
@EliasMårtenson Then maybe you could try N⌿⊃1⍴⊂A?
 
Elias Mårtenson
Elias Mårtenson
That worked. Thanks!
I also discovered a bug in KAP where ⌿ doesn't do the right thing.
 
Bubbler
Bubbler
Actually it can be simplified to ⊃N⍴⊂A (GNU) / ↑N⍴⊂A (Dyalog)
 
Elias Mårtenson
Elias Mårtenson
5:59 AM
@Bubbler thanks a lot
It seems to correctly. Thanks.
 
Adám
Adám
@EliasMårtenson N/[⌈ax],[ax]A where ax is fractional.
 
Bubbler
Bubbler
@Adám Oh wow.
 
Adám
Adám
6:15 AM
APLcart has (⍋⍋ax,⍳≢⍴A)⍉A⍴⍨N,⍴A which is very clever but more verbose, and slower.
 
Bubbler
Bubbler
When a task can be done cleanly with axis notation, I guess that should be on APLcart
instead of clever but obscure ones
 
Adám
Adám
Yup (and making it so now, but I suspect I got this from some idiom list.
 
Adám
Adám
6:34 AM
@Razetime Did you check the count of the result?
 
Adám
Adám
6:52 AM
@Razetime No, your ∪ is applied to matrices.
 
Razetime
Razetime
@Adám My machine is not strong enough to handle it
 
Adám
Adám
@Razetime Are you sure? How much RAM do you have?
 
Razetime
Razetime
Not about the RAM
It's fan stopped working
 
Adám
Adám
Oh. I have a 39-byter (using pmat) that can run on my machine, but not on TIO.
 
Razetime
Razetime
@Adám Just need to check if all sums are equal
 
Adám
Adám
7:01 AM
@Razetime Yes, but the sums are returned as a 2-by-2 matrix.
 
Razetime
Razetime
oh.
so ∊ then do it
{⍵/⍨((1=≢∘∪)∊{+/,⍵}⌺2 2)¨⍵}{3 3⍴⍵}¨↓pmat 9
 
Adám
Adám
Yeah, but it takes a very long time to run (I'm running something similar now) because of nested arrays. My flat version (the 39-byter) has acceptable runtime.
 
Razetime
Razetime
oh, like the other entries in the question
I just wanted to use window cause it sounded perfect for the task
 
Adám
Adám
Hm, one processor is getting hammered here. I should split this into 4 CPU threads…
Meh, the nested version isn't much shorter, since we have to split pmat's result anyway.
@Razetime Indeed. ≢r← s⌿⍨1=≢∘∪⍤1⍪+/,⍤⊢⌺1 2 2⊢s←3 3⍴⍤1pmat 9 runs in about 5.5 seconds by me.
 
Razetime
Razetime
@Adám 5 seconds, nice
is there any constraint for time defined?
 
Adám
Adám
7:12 AM
What? Where?
 
Razetime
Razetime
in the question
nope its not restricted
 
Bubbler
Bubbler
While most code golfs don't have time limits, having a solution that works fast always helps (especially when the task is nontrivial)
 
Adám
Adám
Now I really want to get that down to 2 seconds by using all 4 processor cores.
OK, I got it down to 3 seconds.
 
Razetime
Razetime
how do we verify it?
 
Adám
Adám
It has the right count.
This was s⌿⍨∊{1=≢∘∪⍤1⍪+/,⍤⊢⌺1 2 2⊢⍵}IÏ {⍵⊂[0]⍨(≢⍵)⍴1↑⍨4÷⍨≢⍵}s←3 3⍴⍤1pmat 9 using IÏ from isolate.dws (a polyfill for ∥¨).
 
Razetime
Razetime
7:24 AM
Don't remember seeing ∥ in extended
 
Adám
Adám
It isn't.
 
Razetime
Razetime
what does it do?
 
Adám
Adám
It is exactly like ¨ but runs each in a parallel CPU thread.
∥ is a monadic operator just like &, but for CPU threads instead of "green" threads.
@Razetime GIF?
 
Razetime
Razetime
@Adám if you have one, sure
 
Adám
Adám
Working on it.
 
Razetime
Razetime
7:37 AM
nice!
 
Ven
Ven
8:06 AM
Hi, I'm looking for the snippet to (re)print trains without the extraneous parentheses. I'm tried a few keywords through the search and APLCart but no dice so far.
(my code is 1∘↓(+/⌽(⌽⍳7)∘~))
I guess 1∘↓+/⌽(⌽⍳7)∘~ is a first step...
 
Adám
Adám
8:27 AM
@Ven It is a setting of ]box but it isn't 100% reliable: ]box on -trains=def
It fails to keep a parenthesis to separate to adjacent arrays, e.g. (,∘1)(1∘,)
 
Ven
Ven
mh, doesn't seem to be allowed* on TIO or TryAPL.
 
Adám
Adám
@Ven Did you forget to separate the train from its argument?
 
Ven
Ven
I had no argument, the error comes from ]box
 
xpqz
xpqz
Is there a way to persist the settings of ]box and ]rows when saving a workspace as a .dws file?
 
Adám
Adám
@Ven Try APL doesn't allow you to change ]box settings, and TIO doesn't have ]box by default.
@xpqz It is a session setting, not a workspace setting.
 
Ven
Ven
8:44 AM
@Adám Ah, I always forget about SALT, thanks. I'm unsure about that 1×⎕io though...
@Adám Could you re-pin a link to APLCart, please? Though it's also very easy to google.
 
Razetime
Razetime
@Adám Posted!
 
xpqz
xpqz
@Adám Does that save settings to a separate config file?
 
Adám
Adám
@Ven If you use DuckDuckGo, you can use the !aplcart bang.
 
Razetime
Razetime
@Adám That's convenient
 
Adám
Adám
@xpqz Yes, but you can probably deconstruct the entry to figure out how to save to your default/current file.
 
Ven
Ven
8:46 AM
Well, typing aplcart + click or !aplcart are about the same :). But good to know.
 
Bubbler
Bubbler
So ⌽⍳7 is "optimized" to an arithmetic using ⎕IO instead of costly reversing the array?
 
Ven
Ven
it's funny
 
RGS
RGS
9:06 AM
Shouldn't this add rows of the left matrix to the rows of the right matrix?tio.run/##SyzI0U2pTMzJT///…
 
Bubbler
Bubbler
@RGS What is your expected result?
 
RikedyP
RikedyP
@RGS Yeah but you need frame conformance (2≠5 and neither = 1)
 
RGS
RGS
@Bubbler something with shape s.t. ∧/ shape ∊ 2 4 5 gives 1
 
RikedyP
RikedyP
@RGS Worse than I thought - why does a 1-row matrix not extend but a vector does? hmmmm....tio.run/##SyzI0U2pTMzJT///…
 
RGS
RGS
(and s.t. ^/ 2 4 5 € shape also gives 1)
 
Adám
Adám
9:24 AM
room topic changed to The APL Orchard: Learn, teach, ask, code, golf, & discuss. See aplwiki.com/wiki/APL_Orchard for access and info, aplcart.info for simple questions. [apl] [array-manipulation] [j] [k] [tips]
@Ven How is that ↗
 
Ven
Ven
@Adám Perfect!
 
Bubbler
Bubbler
Maybe simple how-to questions?
 
Adám
Adám
@Bubbler Done.
 
RGS
RGS
9:53 AM
@RikedyP ah I get it, I expected it to do some kind of "cross-product" and add all rows of the left with all rows of the right
 
RikedyP
RikedyP
@RGS It works like scalar extension (but apparently not like singleton extension)
@RGS You can do that with multiple uses of rank ⍤j⍤k
 
 
1 hour later…
Razetime
Razetime
11:10 AM
{∧/(⍳⍵)∊(⍋⊃¨⊂)∪+/¨⊃,/(⊢,,¨)(∪⊢∨⍳)⍵} I made a delicate primes finding function
but I get a WS FULL error on very big numbers
 
Adám
Adám
Try to keep things flat for optimal CPU and memory usage.
 
Razetime
Razetime
11:37 AM
ok got it
 
 
1 hour later…
Razetime
Razetime
12:54 PM
how do I group equal runs of elements?
 
rak1507
rak1507
{⍵⊂⍨1,2≠/⍵}
Might be something cleverer
but idk
 
Marshall
Marshall
1:08 PM
If you know the runs are globally unique, you can use Unique Mask ≠ instead of 1,2≠/.
 
rak1507
rak1507
Now that is what I call something cleverer!
 
xpqz
xpqz
1:53 PM
As we're on the topic of partitioning, is there a better way than Xs {↓(⍺÷⍨≢⍵) ⍺⍴⍵} Y to group Y into vectors of length Xs?
 
rak1507
rak1507
I'm not quite sure what that does, can you give an example?
 
xpqz
xpqz
2 {↓(⍺÷⍨≢⍵) ⍺⍴⍵} ⍳6
┌→──────────────────┐
│ ┌→──┐ ┌→──┐ ┌→──┐ │
│ │0 1│ │2 3│ │4 5│ │
│ └~──┘ └~──┘ └~──┘ │
└∊──────────────────┘
3 {↓(⍺÷⍨≢⍵) ⍺⍴⍵} ⍳6
┌→────────────────┐
│ ┌→────┐ ┌→────┐ │
│ │0 1 2│ │3 4 5│ │
│ └~────┘ └~────┘ │
└∊────────────────┘
 
rak1507
rak1507
{⍵⊂⍨(≢⍵)⍴⍺↑1} maybe?
I don't know
Your one errors if it doesn't go exactly
Is that what you want?
 
xpqz
xpqz
I like yours better.
 
Adám
Adám
2:08 PM
Golfed: ⊢⊂⍨≢⍤⊢⍴⊣↑≢
 
xpqz
xpqz
Can you do this with stencil? I had a go but I clearly have a limited understanding of how it works.
 
Adám
Adám
@xpqz It has difficulty at the edges.
 
Razetime
Razetime
@rak1507 that is pretty clever
 
ngn
ngn
Jul 26 at 14:37, by ngn
@RGS -1 byte: ⍉|∘⍳∘≢⊢⌸⊢. btw, in k9 this whole thing is a primitive: ^ (the new "cut"), e.g. 2^"abcdef" -> ("ab";"cd";"ef")
(the apl train above returns a matrix, so it needs a split: ↓)
 
xpqz
xpqz
I think it also has a slightly different behaviour for non-clean divides -- it pads the last group, rather than leaving it shorter.
 
ngn
ngn
2:25 PM
yes
@xpqz there's no obvious way to design this primitve for non-clean divides. k9 just drops the last group. if you leave it shorter, or try to distribute the imbalance as evenly as possible in the result, like pre-k9 k did with its 0N 2#x there's danger of getting a length error when you do something like reduction on the result
 
Adám
Adám
@ngn My proposal is ¯1 for truncation, ¯2 for requiring exact divides, ¯3 for recycling.
 
ngn
ngn
@Adám that looks like something j would do :)
 
Adám
Adám
There's a precedent: (type,shape)⎕MAP allows up to one ¯1 in size for auto-sizing that dimension, and it truncates.
 
ngn
ngn
can't decide which is better, so to avoid responsibility, let's just put all the options in, and assign them hard-to-remember numeric constants :P
@Adám but just -1, right?
that's just a workaround for apl's lack of nulls, not j-bloat
 
Adám
Adám
Yeah, J's Cut operator does this in the extreme, so we decided on a specific subset for ⌺ ― and got it pretty wrong, I think.
@ngn Yes.
@ngn We do have a null, though.
 
ngn
ngn
2:38 PM
@Adám quad null? that's not quite what i mean by null..
you don't even use it as filler in arrays where the prototype is a namespace
 
Adám
Adám
It is a null, and could be used for such. I know of at least one case where we use it as filler in arrays where the prototype is a namespace.
 
ngn
ngn
what is 2↑# ?
 
xpqz
xpqz
A 'NONCE' whatever that is :)
 
ngn
ngn
an error, no filler is used
 
Adám
Adám
# is not a namespace, but an instance of the Root class, so it would create additional instances of the Root class ― if that was allowed.
 
xpqz
xpqz
2:41 PM
APL errors are terse. But not quite as terse as k's ...
 
Adám
Adám
      i←⎕NEW⊂'Timer'
      2↑i
 #.[Timer]  #.[Timer]
      =/2↑i
0
 
ngn
ngn
@Adám "# is not a namespace" - really??
 
Adám
Adám
@ngn Yes, really: ⋄{⎕NC⊂,'⍵'}# 9.2 is instance, 9.1 is namespace.
@Moonchild Hey, where's the bot?
 
ngn
ngn
in any case, it would be better if you had a numeric null for ⎕map's left arg
 
Adám
Adám
@ngn How about '⌈' for round-up '⌊' for round-down, and '⊢' for require-exact?
 
ngn
ngn
2:43 PM
@Adám so, if i look in your documentation, i won't find instances of # being described as a namespace?
@Adám chars?
 
Adám
Adám
@ngn I'll be the first to claim out docs have issues.
@ngn Yup, or even functions if the array model supports it.
 
ngn
ngn
@Adám a char is not a number, just like a namespace is not a number
 
xpqz
xpqz
Lookming back through my notes, it seems we discussed the general matrix case of this a while back.
Non-overlapping sub-matrices.
This was one solution proposed: 
 ⎕IO←1 ⋄ 2 2 {⊂[2×⍳≢⍴⍵]⍵⍴⍨∊⍺,¨⍨⍺÷⍨⍴⍵} 4 4 ⍴ ⍳16
┌→────────────────┐
↓ ┌→──┐   ┌→──┐   │
│ ↓1 2│   ↓3 4│   │
│ │5 6│   │7 8│   │
│ └~──┘   └~──┘   │
│ ┌→────┐ ┌→────┐ │
│ ↓ 9 10│ ↓11 12│ │
│ │13 14│ │15 16│ │
│ └~────┘ └~────┘ │
└∊────────────────┘
 
Adám
Adám
3:27 PM
@ngn So? Who cares about data types for small control arrays?
 
ngn
ngn
@Adám the implementation becomes more complex, as you need additional type checks
it's also bad taste to mix types like that
 
Adám
Adám
I would think you'd need additional checks even for special values, and taste is subjective.
 
ngn
ngn
@Adám yes, but on top of that you'd have to check for char/namespace too
 
Adám
Adám
Namespace?
 
ngn
ngn
ah, pardon.. "instance" :)
 
Adám
Adám
3:29 PM
What? Where? Nobody suggested using # as a special value here.
 
ngn
ngn
@Adám you suggested ⎕null at some point, if i understood correctly
 
Adám
Adám
⎕NULL is not an object (namespace/class instance) at all.
And, no I didn't suggest using it for ⍴. I was just commenting that we do have a null value.
 
ngn
ngn
@Adám =/⎕dr¨#⎕null whatever it is, you'd have to test for its type
 
Adám
Adám
@ngn Uh, =/⎕dr¨#('abc' 'def') so that doesn't say anything.
 
ngn
ngn
@Adám fair. what's the official way to get the type of something?
 
Adám
Adám
3:36 PM
@ngn ⎕NC but it requires a bit of fiddling because it takes a name, not a value.
 
ngn
ngn
it returns 2. so []null is a number?
 
Adám
Adám
      {⎕NC⊂,'⍵'}¨#⎕NULL('abc' 'def')⎕SE.Dyalog 42
9.2 2.1 2.1 9.1 2.1
@ngn No 2 means variable. 2.1 specifically means a normal array (not a ref).
 
ngn
ngn
i always forget these..
so how do i get the type of something then?
i thought []dr was the closest
 
Adám
Adám
Where type means what? number/character/nested/namespace/class/…?
 
ngn
ngn
@Adám yes
and it's ok to have multiple number types for 1-byte, 2-byte, 4-byte..
 
Adám
Adám
3:40 PM
It isn't really a thing though, as there are many types of namespaces, and many types of nested arrays too.
 
ngn
ngn
anyway
 
Adám
Adám
I'd say the type primitive applied to an array gives the right image, and if that doesn't work, then it is a namespace or class, in which case its type is itself, so {0::⍵ ⋄ ⊃0⍴⊂⍵}
 
ngn
ngn
why am i even telling you this. keep adding bloat and be happy.
 
Adám
Adám
What would you say is the type of 1'a'?
 
ngn
ngn
@Adám in k we call it a generic list
in apl: array of pointers, i guess?
mixed array?
 
Adám
Adám
3:44 PM
@ngn Bloat doesn't make me happy. Getting a salary doesn't really either, but it is unfortunately a necessary evil. I don't add bloat on purpose, but I do speak for increased utility to increase the chance that my salary remains.
@ngn Right, it isn't at all clear what type means, so asking what the type is isn't well-defined either.
I still think that we should make monadic ⍷ be like old monadic ∊ (i.e. type) and I wouldn't mind extending that to return self if a normal type couldn't be gotten (as in my dfn above).
This would make the type of # be # and of a class be that class, while the type of an instance would be a new clean instance of the same class.
 
dzaima
dzaima
@Adám imo there are two types of types that matter - effective/visible type (so whether argument is namespace, number, char, array, etc), and internal type (array of 8/16/32-bit ints/64/128-bit floats/references, reference, simple (probably also split into simple scalar number & char))
 
ngn
ngn
4:00 PM
@Adám here's how i see things: you're getting a salary because (a.) your friends from management believe you're doing something useful and (b.) because customer lock-in guarantees stable profits. there's no conflict between customer needs (useful functionality) and good engineering (lack of bloat).
 
Adám
Adám
4:22 PM
@ngn True, but there is a conflict between customer lock-in and removing bloat. And catering to additional customer needs with a lock-in product based on sub-optimal engineering leads to bloat.
 
ngn
ngn
@Adám yes! that conflict's name is backwards compatibility
 
Adám
Adám
Correct. But it is also the case that backwards compatibility is a core locked-in customer need. Call it a bad self-perpetuating cycle, if you want ― it still pays my salary.
@dzaima "number, char, array, etc" ― what is "array"?
 
dzaima
dzaima
@Adám "array" is a thing with depth > 0
 
ngn
ngn
@Adám well, with the risk of repeating myself: there are solutions to this. afaik you don't have access to the complete source code of your customers, but you have a very small number of important customers. you can easily talk to them. you can give them a tool that patches old sources (like python's 2to3). if the key customers agree to change their sources, you can fix old mistakes, and the rest of the customers will follow.
 
Adám
Adám
@ngn I guess we could ask the few big ones. That might be a viable plan for migration to APL'.
 
dzaima
dzaima
4:32 PM
@ngn why do that when you could not? I certainly would much prefer for updates to take precisely 0 effort, certainly much more than "removal of bloat" when the "bloat" might contain things i use
 
ngn
ngn
@Adám and not only that, if adam manages to broker such a deal among all stakeholders, he might get a promotion :)
 
dzaima
dzaima
@dzaima (both types should also have function, monadic operator and dyadic operator, forgot about those (maybe it could make sense for fn/op to be the same "internal type", but i don't know))
 
ngn
ngn
@dzaima depends on what your horizon is. if you're looking forward to a couple of months, sure, it's easier to ignore mistakes. but in the long term they cost you more.
how many times have we bumped into that ⍳,n bug.. and how much would it have cost to patch a codebase or two, and fix it once and for all
 
dzaima
dzaima
@ngn and then another couple code-bases when production goes down due to it being relied-on behavior in edge-case error-checking. They certainly would much prefer it to not have changed
 
ngn
ngn
@dzaima if you're afraid of bears don't go in the woods. if you're afraid of making and fixing bugs, don't write code.
 
dzaima
dzaima
4:47 PM
@ngn there's a difference between going in the woods by my house where i know everything and a random place somewhere on the world explicitly called dangerous
 
ngn
ngn
@dzaima if you know the bug well, you can certainly write a patching tool for it
and if the patching tool fails.. well, you didn't know the bug well in the first place
 
dzaima
dzaima
@ngn at least for ⍳ i can't imagine any solution other than replacing it with a call to a function that does what it did before working at all
and then, making (,1) + 1 2 3 error would mean replacing each and every scalar function with a wrapper
 
Adám
Adám
@dzaima Which would break in ⍣¯1 contexts, not to mention considerable slow-downs.
And replacing the + in +/ would give a different result too! That'd need another wrapper.
@ngn Dyalog did fix ⍳⍬ and it was a bit of a disaster.
 
ngn
ngn
@Adám so they'll never try fixing anything again :)
@dzaima if customers have test suites, that can be really useful here
maybe i'm too optimistic about automatic patching
 
dzaima
dzaima
@ngn though i obviously don't know, i wouldn't be surprised if enough things would still not get caught by it
@ngn APL is a pretty damn bad language for static analysis of data
 
Adám
Adám
4:58 PM
@dzaima data code?
 
ngn
ngn
@dzaima @Adám yep
 
dzaima
dzaima
@Adám code can be reasonably well parsed at compile-time and simple transformations can be automated if you're brave enough, but most of the problems are about specific shape/rank inputs, which you just can't know anything about most of the time
 
Adám
Adám
@ngn Not so, we later fixed the implied ⋄ before control words, and gave a tool to track down such cases. That went so-so.
@ngn It would maybe be more feasible if Dyalog APL code was always stored in a consistent text file format. Remember that code can be source text files, workspaces, component files, external workspaces,…
 
dzaima
dzaima
@dzaima (remove "shape/rank"; Just "specific inputs" is good enough)
 
ngn
ngn
@Adám there was a certain guy at a certain company (no names for privacy) who was writing a tool for static type/rank/shape checking of tradfns's args and results, with declarations in the comments. what happened with that effort?
 
Adám
Adám
5:07 PM
Dunno.
 
Elias Mårtenson
Elias Mårtenson
5:20 PM
Does Dyalog implement monadic equals? I.e. is =4 valid?
 
ngn
ngn
@EliasMårtenson Try it online!
 
Elias Mårtenson
Elias Mårtenson
Thanks. So it doesn't. It seems like such an obvious character to give a function. I mean ≡ has a monadic version.
 
Adám
Adám
@ngn Bad test, since it fails for ≠ (TIO is on 17.1).
@EliasMårtenson Sure, but what function?
 
Elias Mårtenson
Elias Mårtenson
Is there a monadic ≠?
@Adám I was thinking of using it to force evaluation of a lazy value
 
Adám
Adám
@EliasMårtenson Interesting. That might be a good idea if we add lazy values…
 
ngn
ngn
5:24 PM
@EliasMårtenson yes
 
Elias Mårtenson
Elias Mårtenson
Right now, the only way to do it in KAP is to use ←, because assignment forces evaluation, which is why you'll see in my code the use of foo←foo to ensure that the variable doesn't hold a lazy value.
 
ngn
ngn
@Adám dennis of tio shows no signs of coming back. what are we gonna do..
 
Adám
Adám
@EliasMårtenson Monadic ≠ was taken verbatim from A Dictionary of APL. It might make sense to take monadic = too.
@ngn I really want to set up a similar service for APL. I was in talks with a 3rd part that could contribute a secure container.
 
ngn
ngn
@Adám will it support other languages?
 
Adám
Adám
@ngn I wasn't planning on it.
@ngn I'm sure the community would be happy to take over maintenance if Dennis would give over the reins. If he stays incommunicado, someone will probably create a mirror site. All the code is online. Dyalog could probably cover the costs too, but we don't have the manpower.
 
Elias Mårtenson
Elias Mårtenson
5:34 PM
@Adám Interesting link you gave me. The discussion of the parser in this document describes the first implementation I guess? It's fascinating to see that apparently it not only evaluated from right-to-left, but it also parsed the code from right-to-left?
That's somewhat bizarre to me.
 
Adám
Adám
@EliasMårtenson No, it is a very late document, after the mainstream APL implementors broke with Iverson.
 
Elias Mårtenson
Elias Mårtenson
@Adám I see.
@Adám A right-to-left parser sounds so strange. I'm curious why he did it like that? Perhaps it alowed the implementation to evaluate while parsing, thus not having to keep a parse tree in memory?
 
dzaima
dzaima
5:55 PM
@EliasMårtenson parsing in the direction of execution indeed means you can do the two at the same time, with the only temporary structure being a simple stack of evaluated but unused things
 
 
1 hour later…
ngn
ngn
6:57 PM
@Marshall did you try this? i hit a wall with bdf2psf and i don't understand what it's complaining about
 
Marshall
Marshall
@ngn No, sorry.
 
ngn
ngn
7:38 PM
i tried psftools. it's great - easy to build and use. so, if someone gives me a set of same-shape boolean matrices, i can easily make a psf font out of them.
 
TessellatingHeckler
TessellatingHeckler
8:22 PM
Doesn't Dyalog have a way to support language changes with ⎕ML? That they could use for breaking changes, opt-in to new improved APL' by setting ⎕ML to the year/month of release.
 
Adám
Adám
@TessellatingHeckler I don't know if ⎕ML would be the right "thing", but yes, I'm also envisioning using the same interpreter (or at least a very shared code base) for a "fixed up" APL ― APL'.
 
TessellatingHeckler
TessellatingHeckler
@Adám but you already have 'extended', lol. I'm beginning to wonder if array languages are as fractured as lisp yet. Does ~everyone who meets APL leave saying "I could do that better" and it never gravitationally captures enough people to solidify into a large and popular thing?
 
ngn
ngn
9:06 PM
@Marshall i've just discovered that what i'm looking for has been in front of me all along: apt-get install psf-unifont && setfont /usr/share/consolefonts/Unifont-APL8x16.psf.gz
 
 
1 hour later…
Dyalog APL
Dyalog APL
10:23 PM
<jcowan> Does anyone know of use cases for arrays whose shape includes a zero? They have no elements, but they are valid.
 
Marshall
Marshall
jcowan: They're used constantly. If your program works with a list of something but in a particular case there aren't any, then it should use an empty list.
 
Adám
Adám
jcowan: They are often used as initial value to which more is then appended.
→ 38 messages moved to The Nineteenth Byte (Try to keep it on topic here. General code golf chat belongs in TNB.)
 
TessellatingHeckler
TessellatingHeckler
sorry; something a bit more on-topic: when I search APLCart for "/" it brings up a lot of things which have a / in the category in the TSV but aren't about reduce/compress.
 
Adám
Adám
@TessellatingHeckler APLcart's focus is not to document the language, but to answer "how-to" questions. Just type it in your interpreter and press F1 or enter ]help /
 
Dyalog APL
Dyalog APL
<moon-child> they certainly come up, generally. In j IIRC a boxed empty array (a:) gets used a lot in indexing
 
TessellatingHeckler
TessellatingHeckler
10:34 PM
@Adám I was trying to find the how-to mask and filter in a train without an intermediate variable, searching "filter", "reduce", "compress" then "/"
 
Adám
Adám
moon-child: But that's only because a: is used as a special value (well, an extra enclosure is).
 
TessellatingHeckler
TessellatingHeckler
I was looking for 5(>⊢⍤/⊢)⍳10 or similar by searching "/"
 
Adám
Adám
@TessellatingHeckler Ah, good one. How to force / etc. to be functions (not operators) in tacit.
Will add.
 
elliptic00
elliptic00
Does APL have tuple like other language?
 
Marshall
Marshall
@elliptic00 Nope, just arrays.
 
elliptic00
elliptic00
10:46 PM
I assume there is no record either?
How to you handle Object like person which has name, age, address, phone # etc?
 
Marshall
Marshall
@elliptic00 That's right. Some dialects like Dyalog have namespaces, but generally the idea is to use arrays for everything.
@elliptic00 I think the typical way would be to decide on an order for those properties and put them in an array in that order. You would have to reference them by index instead of by name. If there are many of these objects, you can instead have one array for each property, which lists the values it has for each object.
 
Dyalog APL
Dyalog APL
<jcowan> Empty vectors are one thing, since they are sequences. But empty matrices and larger ranks strike me as a kind of useless generalization.
 
Marshall
Marshall
jcowan: The preference should be for generalization instead of restriction in a case like this where the general version is simpler to describe. And if there's a property relating to elements from more than one sequence (say, number of item i that person j owns), then the natural way to store it is in an array with one axis for each sequence. If one sequence is empty, the entire array will be as well.
 
Adám
Adám
@elliptic00 Is there any practical difference between them? I.e. any reason to have both?
 
Marshall
Marshall
@Adám I understand the difference between tuples and lists/arrays in a statically-typed language (tuples have a type for each element but lists have only one type for all of them), but I don't know why they should be different in a dynamically-typed one. Python's use of lists and tuples always confuses me.
 
ngn
ngn
10:56 PM
@Marshall in python the difference is mutability
 
Adám
Adám
Right, that's what I mean. In a completely dynamic language like APL, where all* arrays are immutable, there would be no difference.
 
ngn
ngn
@Adám except dyalog namespaces :P
 
Adám
Adám
*
 
rak1507
rak1507
What is a←1 2 3 4 5 6 ⋄ a[4]←7 if it's not mutation?
 
ngn
ngn
@rak1507 copying(if refcount>1) or reuse(if refcount=1)
 
rak1507
rak1507
10:58 PM
interesting
 
Adám
Adám
@rak1507 ^^ because b←a ⋄ a[4]←7 doesn't change b.
 
ngn
ngn
but if they are dyalog namespaces.. :P *
 
Marshall
Marshall
@rak1507 It is a reassignment of a with a new value related to the old one. The difference is that a[4]←a won't result in an array that contains itself.
 
rak1507
rak1507
TIL
 
Adám
Adám
@elliptic00 Marshall mentioned namespaces, but Dyalog APL has proper OO (inspired by C# and Visual Basic) so you can do all that. You can even have objects with keyed properties. (Doesn't mean you should.)
 
ngn
ngn
11:02 PM
"proper" and "OO" in the same sentence - hm. "Visual Basic" and "inspired" in the same sentence - wow!
 
Adám
Adám
@ngn The "Doesn't mean you should" goes on both sentences.
 
Dyalog APL
Dyalog APL
<moon-child> fair enough. I think if you can imagine that a sequence just happens to be zero-length, you can imagine the same for any other axis. But I haven't ever had call to use them
 
Adám
Adám
moon-child: Let's say you have some sort of query on a 7-column table, and there are no matches. Would it not be strange to get an empty list back, rather than a 0-row 7-column table?
Let's say you then want to extract column 3 from the search result. That'd fail on an empty list, but work on the 0-row table.
 
elliptic00
elliptic00
v ← 2 3 ⍴ ⍳ 10
1 2 3
4 5 6
v[1][1]
how to access the first element of v?
v[1][1] does not work
 
cannadayr
cannadayr
@dzaima or @Marshall

reading implementation/vm:
"When the block is evaluated, its frame is initialized using any arguments passed to it, the next instruction's index is pushed onto the return stack, and execution moves to the first instruction in the block."

the "next instructions index" refers to the index of the next list of block parameters (a,b,c,d), correct?

also, (i think) this means that the return stack is initialized before the frame is evaluated, and is passed to the next frame evaluation
 
Bubbler
Bubbler
11:12 PM
v[1;1]
 
Adám
Adám
Or ⊃v
 
elliptic00
elliptic00
wow.. APL uses Visual Basic array style..LOL
 
Bubbler
Bubbler
Or in a convoluted way: v[1;][1]
 
elliptic00
elliptic00
is Visual Basic v[1;1] style?
anyone idea how to write v to a file using Dyalog APL?
 
TessellatingHeckler
TessellatingHeckler
Visual Basic uses parens for indexing, v(1)
 
Adám
Adám
11:16 PM
@elliptic00 What exactly do you expect in the file?
 
elliptic00
elliptic00
1 2 3
4 5 6
keep it like a 2 dim vector
 
Adám
Adám
(↓⍕v)⎕NPUT'path/file.ext'
 
Moonchild
Moonchild
@Adám got 500'd. Should reconnect automatically now when it gets timed out
 
elliptic00
elliptic00
what is the symbol inside (..)?
it is hard to see it..
 
Adám
Adám
@elliptic00 ⍕ converts the numeric matrix into a character matrix, and ↓ splits it into a list of lists (which then become lines in the file).
 
elliptic00
elliptic00
11:21 PM
I got Invalid file encoding type
 
Adám
Adám
Oh, stupid me. Sorry. It should be (⊂↓⍕v)⎕NPUT'path/file.ext'
@elliptic00 Because ⎕NPUT can take multiple parameters as left argument, and only the first is the actual content, so if you don't want to specify any more parameters, you've got to enclose to make the content a single element.
 
elliptic00
elliptic00
what we need Enclose?
It works.. Thanks..
 
Adám
Adám
@elliptic00 Do you know about APLcart?
 
elliptic00
elliptic00
I do try to search on APLcart, aplcart.info but I have hard time to find the simple example which write some data to a file,
This example should be in APLcart
I have searched the function for long time... and see some example on Dyalog Website, but I did not understand it....
 
Adám
Adám
@elliptic00 What did you search for?
 
elliptic00
elliptic00
11:28 PM
write to file.. write file... etc.
 
Adám
Adám
@elliptic00 OK, then you should find ⎕NPUT in that list. Only the example doesn't show how to write multiple lines from a vector of text vectors.
 
elliptic00
elliptic00
'write vector to file' ..
 
Adám
Adám
I'll amend the example.
 
elliptic00
elliptic00
Yep, I know NPUT, but it is hard to guess what the format...when writing data to file i APL... I'm pretty new to APL...
 
Adám
Adám
Would this have helped?
 
elliptic00
elliptic00
11:31 PM
it make sense to write 2 dim or 3 dim vector to file..in APL.. because APL is array based lang
 
Adám
Adám
But files are 1-dimensional. (Or at most ragged 2-dimensional if you look at them as collections of lines.)
I do agree that writing a character matrix to a file could work. I'll suggest that.
 
elliptic00
elliptic00
11:52 PM
I did look at the example in APLcart, but I have no idea how to write vector when I was looking the example...
 

« first day (1357 days earlier)      last day (1313 days later) »