The APL Orchard

Dyalog APL

2:42 AM

@Razetime Execution timed out

@rak1507 Execution timed out

Razetime

3:28 AM

@rak1507 facepalm how did i not realize that..

looks like the bot is back up

⋄'Are you alive, sir?'[5 3 16]

Dyalog APL

@Razetime yes

Adám

5:15 AM

@Razetime I've signed up, but am not allowed to comment. Maybe ask OP of you can take input in reverse (it seems like it based on the text, but doesn't hurt to ask). If so, then you've got a 1-byter.

Razetime

5:46 AM

a1-byter?

isn't ⌽ required?

@Adám hm, email verification maybe

ih nvm taking the coefficients in reverse

that doesn't sound right idk

@Adám ok, done

Did you see the drive link I sent?

moonheart08

6:30 AM

i am impressed.

jeez APL

this isn't a challenge i expected a 1-byter for to say the least

but also not surprised APL did it

Adám

6:57 AM

@Razetime I did, but didn't take a close look yet. Speaking of which, you have mult←{⍺+⍵×256} ⋄ mult / … there, which is simply 256⊥⌽…

@Razetime Hold on, you didn't actually run the code, did you?

moonheart08

TIO shows the right output

...for the test cases, so if ⊥ does something weird that won't show

Adám

@moonheart08 You're talking about your (?) code golf challenge, I presume. I was responding to the file Razetime linked.

moonheart08

ah

'yea it's mine

Razetime

7:16 AM

@Adám I did run individual parts

Adám

@Razetime I highly recommend running the code as a whole at least once. For instance, you'd discover that control structures like :If aren't allowed inside dfns {…}. :-)

Razetime

it worked on png and gif

oh.

darnit

good thing I asked

what's the syntax for running files?

Adám

There isn't really one yet. You can hack it, but for now, your best option is to make the entire thing a giant tradfn or alternatively make it into a namespace.

Razetime

ok

so basically

remove the dfn wrapper

and make that a tradfn

and then make the whole thing a tradfn

Adám

No, either one giant tradfn and remove the inner braces, or wrap everything in :Namespace … and change the inner brace to a tradfn.

Razetime

7:24 AM

ok gianf tradfn it is

sO

Adám

Don't forget to manually localise used names then.

Razetime

I guess I'll make the tradfn print out the values as well

Adám

Any problem with returning a result?

Razetime

I just wanna make sure if it's printing everything correctly

Adám

OK, but if you return a result, it gets printed unless captured.

Razetime

7:40 AM

ok my jpg part is wrong

it works great for anything else

Adám

Do think you can fix the jpg part or do you need help?

Razetime

ok I think I fixed it

lemme send another link

Most of the problem was with the buffer having negative numbers

and I was searching for positive numbers

@Adám Here

Adám

7:56 AM

@Razetime That looks much better. No for the final version, all the used variable names should b localised, and the tradfn should take an argument instead of prompting, and return a result instead of printing.

Razetime

wait how do tradfns take arguments

Adám

Btw, pro-tip: ⎕NTIE 0 will assign the first available tie number.

Razetime

Returning results is easy enough

@Adám Ah, I took the snippet from the docs

@Adám how do i get the number for ⎕NREAD?

Adám

@Razetime Check out APL Wiki.

@Razetime That's imgTie

Razetime

gr8

Adám

8:04 AM

@Razetime What does pos←⊃⍸(255 216 255)⍷buffer do? I don't see pos used later.

@Razetime Is there a reason you have ⎕FR←1287 and ⎕PP←34? afaict, you're only dealing with integers here.

Razetime

@Adám paranoia

Adám

:-)

Razetime

@Adám um i need to look at the vb file

Adám

@Razetime I have a question about the logic here:

	 dataPos←⊃⍸buffer[markerPos+1]∊192,193,194,195,197,198,199,201,202,203,205,206,207
 ⍝ If the data exists, add in the info.
	 :If ×dataPos

Razetime

that is wrong I think

need to find a proper section

Adám

8:09 AM

Could it happen that the first element of buffer[markerPos+1] is indeed a member of that long vector? If so, then ⊃⍸ will give 0 which probably isn't right when doing ×dataPos

Razetime

yep that's not right

should be 2⊃ probably

Adám

The third element?

Razetime

sorry, 1⊃

Adám

Don't you mean to ask if any such position exists?

1⊃ will fail with a length error if the list has 0 or 1 elements.

Razetime

i think datapos is supposed to find the FF, XX block for the image data

and pos is supposed to verify if jpeg

and those two got mixed up

cause if there's no FF, D8, FF is should say unknown

so it should be :If ≢pos

Adám

8:16 AM

@Razetime Only 1 is truthy, so :If 0<≢pos or :If ×≢pos

Razetime

ok

ok now let me finish the apl cultivation tradfn clas

so, I'm supposed to return all attributes as a heterogeneous array, correct?

∇(imgattrs)←readImageInfo

Adám

Yeah, that makes sense. Maybe a for element vector or a three element vector where one element is a two-element diminsions' vector. Btw, you can use a name list for the result in the header.

Razetime

then this: ∇(d_imageType d_width d_height d_depth)←readImageInfo

∇(d_imageType d_width d_height d_depth)←readImageInfo filename

Adám

That looks sensible.

Razetime

LOL

ok, width and height of jpg's are turning out wrong

Adám

8:34 AM

Testing is a good thing.

Razetime

sure is

Razetime

9:08 AM

it is very satifying to see the lines of a trdfn appear one by one after you paste it

@Adám So I'm doing something wrong here:

⍝ get first jpeg marker which matches required spec
		markerPos←2+⊃⍸(2{(⍺=255)∧⍵∊192,193,194,195,197,198,199,201,202,203,205,206,207}/buffer)
		⍝ dataPos←1+⊃markerPos[⍸{buffer[⍵+1]∊192,193,194,195,197,198,199,201,202,203,205,206,207}¨markerPos]
		d_imageType←'JPG'
	⍝ get the two required bytes, and get dims
		d_height←buffer[mult/markerPos+5 4]
		d_width←buffer[mult/markerPos+7 6]
	⍝ Color depth
		d_depth←buffer[markerPos+8]×8

Adám

I have to bring my son to school. Will have a look in about half an hour.

Elias Mårtenson

9:46 AM

I have noted a discrepancy between Dyalog and GNU APL. For the following expression: ⊃⍬ Dyalog returns 0, while GNU APL returns ⍬. Which one is more correct?

To me, the GNU APL result seems more logcal.

Dyalog APL

<klg> remember ⎕ML

<klg> ⋄↑⍬

dzaima

dyalog and GNU act the same afaict, it's just the ⎕ML difference again (↑ and ⊃ being swapped between GNU APL and Dyalog)

Dyalog APL

0 ⍴ 0

Adám

@EliasMårtenson As klg says, GNU's ⊃Y is Dyalog's ↑Y and vice versa, so they agree.

Elias Mårtenson

Oh right

Adám

9:49 AM

Now, a valid question is which of the two glyph schemes makes the most sense.

@Razetime Other than some redundant parentheses and braces, what seems to be the problem?

Razetime

10:57 AM

it doesnt seem to give the right dimensions

Adám

That might be an issue ;-) Have you tried tracing through it to see if it goes along the expected pathways?

Razetime

11:13 AM

tracing?

Razetime

11:24 AM

@Adám IT WORKS

Adám

@Razetime Great! Did you try with all supported file types?

@Razetime Ctrl+Enter instead of Enter allows you to step through the code line by line.

Razetime

PNG, JPG, GIF working

need to try BMP

couldn't find one of those yet

gonna download one

Adám

Microsoft Paint can create BMPs.

Razetime

I am on mac.

maybe preview can

Adám

Oh :-)

Razetime

11:30 AM

yup preview can

it's just hidden away

Adám

So, BMP works too?

Razetime

height is wrong

not sure why, cause it's just direct indexing

Adám

Maybe an off-by-one error?

Razetime

		' get the width
		m_Width = Mult(bBuf(18), bBuf(19))

		' get the height
		m_Height = Mult(bBuf(22), bBuf(23))

 ⍝ Get dimensions
     d_width←mult/buffer[18 19]
     d_height←mult/buffer[22 23]

both use 0-indexing

Adám

Hm, that does look the same. And you're using the right name? d_ vs m_?

Razetime

11:36 AM

yeah I named eveything d_

Adám

Did you localise all your variables and clean up your workspace?

Razetime

let me step through

@Adám ok, so what exactly does this mean

⎕SHADOW?

Adám

@Razetime No, just ;name1;name2 after the header (either on the same line or one or more separate lines).

Razetime

oh I'll do that then

I cleared workspace and it still gives a wrong answer

Adám

Time to step through then. But sounds like you're almost there.

Razetime

11:43 AM

ok, looks like bmp images have a property of negiative height?!

file /Users/⎕⎕⎕⎕/Pictures/mostafa-meraji-RXooPt4tz9U-unsplash.bmp
/Users/⎕⎕⎕⎕/Pictures/mostafa-meraji-RXooPt4tz9U-unsplash.bmp: PC bitmap, Windows 3.x format, 640 x -427 x 24

so, issues.apache.org/jira/browse/…).

Adám

@Razetime Aha, but that doesn't matter for our purposes of finding the dimensions, so just take the absolute value of that.

Razetime

yeah, I'm not getting negative height

I'm getting a huge value instead

readImageInfo '/Users/⎕⎕⎕⎕/Pictures/mostafa-meraji-RXooPt4tz9U-unsplash.bmp'
 BMP  640 65109 24

Adám

Oh, ⋄ (2*16)-65109

Dyalog APL

@Adám 427

Razetime

65536|<height expression> then

Adám

11:49 AM

What does your height-expression give?

Razetime

oh, 65536|⍨

height expression gives 65109

mult←{⍺+⍵×256}

mult is used in every image type

Adám

@Razetime Sure, but do you know that all BMPs will be in reverse, or do you have to handle both cases somehow?

Razetime

need to handle both, probably

I need a positive test case

looks like mac only makes bottom up ones

Adám

I'll try making a few.

Razetime

docs.microsoft.com/en-us/windows/win32/api/wingdi/…

javaproblem201

11:56 AM

hello

if i have a java program whats the best room to ask it

Adám

@Razetime Aha, so it isn't a gigantic value you're getting but rather a negative one, but you end up with an unsigned int. So, if the value is ≥2*15, do the mod-thing.

javaproblem201

these stack exchange chats seem rather inactive.

Razetime

@javaproblem201 hello there

javaproblem201

hello!

Adám

Java

Dedicated to the discussion of the Java programming language a...

3

javaproblem201

11:57 AM

thank you!

uh i need 20 rep

Adám

But of course, the proper thing is to ask a full question on SO.

Razetime

@Adám d_height←{⍵>2*15:⍵:⍵|65536}mult/buffer[22 23]

Adám

@Razetime I think ≥

Razetime

don't mind me using : twice

Adám

I know, you meant diamond.

Razetime

12:00 PM

ok great

looks like it works

Adám

Do one more test on each file type, and then I'd be delighted to see the code.

Razetime

ok, tests work

drive.google.com/file/d/19H_HmEmLsGitzHqXE9G1vymJ-Ig-bQpg/…

Adám

@Razetime Looks good. You still have a few names to localise (hint: APL's editor will colour global differently from locals). You don't have to localise names that occur in the header. It is also good practice to untie (hint: ⎕NUNTIE) files when you're done with them.

Razetime

yes I'll do that

Adám

Other than that. Congratulations on your very first (?) real-world APL program!

Razetime

12:15 PM

haha thanks

@Adám colour meaning syntax highlighting?

Adám

@Razetime Yes, yours is a better term.

Razetime

i don't see it

nevermind it's very faint

Adám

You can customise your syntax highlighting to make it clearer.

Razetime

underline makes everything better

@Adám drive.google.com/file/d/1J4yfuKXGBmLbMVqupPH7bNBqVtIApXoC/…

Adám

@Razetime Perfect. Thank you muchly, sir!

Razetime

12:28 PM

you are welcome

whats it for

Adám

@Razetime Not sure exactly. Our CTO (Morten Kromberg) asked me to find someone here that could do it. I think it is for a Bridge app of some sort.

Razetime

have fun

why here though

should be simple enough for anyone to do

Adám

12:46 PM

@Razetime d_depth←1+buffer[10]∧7 should probably be a bitwise AND, not a logical AND.

Razetime

1:06 PM

@Adám oh, I'll add that

Adám

@Razetime Also, ⎕IO is a variable that needs localising.

Razetime

@Adám bitwise and snippet {⍉2⊥∧/2⊥⍣¯1⍉⍺,[0.5]⍵} gives me a rank error

Elias Mårtenson

I just implemented disclose with axis. The specification of it in the ISO spec is broken. Not only does it have actual errors, it doesn't actually cover multi-dimensional inner cells. I think I've implemented it correctly, and for my test cases it returns the same as Dyalog. However, if anyone would like to test it, that would be really useful.

Razetime

what exactly is going wrong with the bitwise and

RANK ERROR: Invalid axis
readImageInfo[52] d_depth←1+buffer[10]{⍉2⊥(-∧/0>b)⍪∧/2⊥⍣¯1⊢b←⍉⍺,[0.5]⍵}7

Adám

2:32 PM

@Razetime It is ⎕IO-sensitive, so you need ¯0.5 instead of 0.5.

RGS

@Adám ⎕IO-0.5 as some would say

Adám

@EliasMårtenson Disclose? It is a good idea to avoid that word… (I'm a ware the standard uses that term.)

@EliasMårtenson So you're saying that {(((⍳⍴⍴Z1)~A),A)⍉Z1←↑⍵} (with the Dyalog def. of ↑) is wrong?

dzaima

@Adám ಠ_ಠ why would mix be called disclose

(really disclose makes sense only for ⊂⍣¯1 but whatever)

Adám

@dzaima It says right there in the APL Wiki article:

> a primitive to extract the contents of a nested scalar

CMC: A tacit, strictly monadic no-op function. I.e. like ⊢ but always erroring if given a left argument.

dzaima

@Adám 3 if ⊢ itself is allowed to be in the solution

Adám

2:47 PM

yes yes, that.

dzaima

⊣ but only dyadic seems a bit harder

Adám

Was next CMC.

(or ⊢)

dzaima

@dzaima 3 in dzaima/APL, still thinking about dyalog

@dzaima currently have 7 in dyalog

@Adám have 5 for ⊢ but only dyadic

@dzaima also 3 for dyadic-only ⊢. So all 3 are 3-byters in dzaima/APL :D

Elias Mårtenson

3:07 PM

@Adám No, I'm saying that the ISO spec is wrong.

And since I mostly try to follow the spec, I'm a bit stumped. Now, Dyalog seems to have made a reasonable interpretation of the intent of the spec, so I was wondering if I did it right.

dzaima

3:23 PM

@EliasMårtenson my random tests of comparing result shapes with dyalog didn't show any differences

Adám

@dzaima Yup, me too.

@EliasMårtenson Uh, but that is what the ISO spec says, no?

Elias Mårtenson

Check page 174 in the spec: drive.google.com/file/d/1FPqYokZ5SvY_wZ_tegK8oGseUEcKgw1E/…

Let me know if it makes sense :-)

Adám

3:50 PM

@EliasMårtenson Not sure what you're getting at. It has that exact formula.

Razetime

@Adám drive.google.com/file/d/1Et2B2DwX4wnkSMEAXdbRM9JJlTVk5zf-/…

Elias Mårtenson

@Adám In section 10.2.25, under evaluation sequence, it refers to A without explaining what it is.

If I assume A is actually K, it doesn't explain what happens when the size of K is less than the rank of first-thingy of B.

It also doesn't explain what happens if K is, say 5 vs. ,5

Adám

Right K and A are clearly confused here.

@EliasMårtenson It may be that that behaviour is up to the implementation.

@EliasMårtenson That's probably addressed separately under general singleton leniency.

Elias Mårtenson

@Adám Possibly, but in other places the spec is quite specific on these things.

Adám

E.g. the order in which issues are detected (which influences which error is signalled) is up to the implementation. (Which is why error-based programming can be unreliable.)

Elias Mårtenson

4:00 PM

In the spec, there is an empty line after the specification of Z1. I'm wondering if there is missing text there that is supposed to say: define A to be ,K

@Adám If you look at, say, the description in 10.2.10, you can see how much more specific that description is.

It explains very clearly the types of errors, and in what order those are checked.

Adám

Yeah, you appear to be right. I can take it up internally at Dyalog if you want, and maybe we can all approach ISO to ask for an amendment.

Elias Mårtenson

@Adám I don't know if there is a way to report errors in the standards.

Perhaps they have an errata that I don't have a copy of.

Elias Mårtenson

4:27 PM

What is an efficient implementation of dyadic ~? Is it possible to do better than O(n^2)?

rak1507

create a set out of the left argument, o(n)

Elias Mårtenson

That assumes insertion into the set is O(1). It's typically O(log n), but sure, close enough.

I guess I'll do it that way. Thanks.

rak1507

I thought it was O(1) but idk much about time complexity

Marshall

4:54 PM

@rak1507 Depends on how many lies about hash tables you're willing to believe. Practically speaking hash table insertion is more like √n.

@EliasMårtenson In Dyalog 18.0, all set functions are implemented using Membership and then filtering. See the performance notes, section "Dyadic Set Functions".

Elias Mårtenson

5:10 PM

@Marshall Thanks.

In Dyalog, when I use ]BOX ON, the boxing is very limited. Can I change it so it's more like what GNU APL gives me?

Adám

@EliasMårtenson ]box -s=max

Elias Mårtenson

@Adám Thanks

Adám

Or if you only want it for a single thing: ]display expression

Elias Mårtenson

@Adám Thanks. I do want it for everything.

rak1507

The fact ⊃⍋ is special cased is nice

Why is 1⊥ so much faster than +⌿, couldn't +⌿ be adapted to be as fast?

Adám

5:17 PM

I personally find it distracting and too verbose, so I keep the min style and use ⎕SE.Dyalog.Utils.repObj if I want to know exactly what something is.

@rak1507 No, because with +/ and +⌿ we promise the order in which numbers are summed, precluding vector instructions.

rak1507

Does the order in which they are summed really matter? Do people use +⌿ specifically to sum in a particular order?

Adám

@rak1507 Probably not, but we can't have people's accounts suddenly change amounts. (Yes, I know, one shouldn't use floats for money…)

Elias Mårtenson

There. Dyadic ~ implemented.

I get so exhausted every time I realise I forgot to implement a core APL function. My list of unimplemented functions is still quite long.

Adám

@EliasMårtenson How, like the standard, or like the more sensible version based on dyadic ⍳?

Elias Mårtenson

I didn't even read the standard for that one. I just did it the way Dyalog does it.

Adám

5:26 PM

;-(

Elias Mårtenson

@Adám Oh, what should be changed?

Dyalog allows multi-dimensional values on the right side of ~. So I did the same.

dzaima

@EliasMårtenson dyadic ⍳ allows high rank on both sides. (imo this is largely pointless, but that's my thoughts about much of high rank stuff so idk)

Adám

@EliasMårtenson Yeah, but it is simply ~∘, whereas aplcart.info?q=high-rank without is much more powerful.

Elias Mårtenson

My version requires rank 0 or 1 on the left, any rank on the right. It just uses ravel order anyway.

I don't understand ∘ completely. Isn't ~∘ just equal to ~?

dzaima

@EliasMårtenson Adám is saying that the current definition of ~ is pointless, as if it required a vector right argument, you could reimplement the current version with just ~∘, (equal to a dfn of {⍺~,⍵})

Adám

5:31 PM

@EliasMårtenson Yes, that's the point. The high-rank, leading-axis oriented (i.e. sees the set as the collection of major cells) is more powerful and ^

dzaima

@EliasMårtenson f∘g is just "f, but right argument is preprocessed with g" (and the , after ∘ in these messages is a part of the code block)

Elias Mårtenson

Got it.

@dzaima Yeah, I realise that now :-)

dzaima

@dzaima (personally, i'm fine with ~s current definition as the cell-based high rank stuff is useless to me, and the ravelling fits my ⌿ ravelling args)

Adám

While f⍥g is "f but both arguments are preprocessed with g"

Elias Mårtenson

I've written 595 tests, mostly for different APL functions. 8000 lines of code in the tests compared to 10000 lines for the actual software. Did I mention I really hate writing tests?

Adám

5:36 PM

In which language are the tests written?

dzaima

@EliasMårtenson so 13 lines per test? You could certainly use a bit less lines per test

Elias Mårtenson

@Adám Well, the test case is APL, but the verification of the test is in Kotlin.

They look kinda like this: github.com/lokedhs/array/blob/master/array/src/commonTest/…

Adám

At least all your tests are fun ;-)

Elias Mårtenson

@Adám Haha. :-)

rak1507

Could you store the test cases in a more compact format like json and then go through each of them?

Elias Mårtenson

5:39 PM

The tests are not just verifying the output of the interpreter. It also verifies that the underlying data is of the correct form.

@rak1507 Well, possibly. I've been thinking about ways to do it. It's not the amount of code that is the problem though. It's thinking about what to test, and then verifying it manually in GNU APL and Dyalog.

I have macros to generate a lot of the boilerplate anyway.

rak1507

Fair enough

dzaima

@EliasMårtenson even checking for deferred execution should be possible within the interpreter. If your ≡ & ⍴ & outputting work correctly (for which i'd understand writing native tests), the rest should be doable within

Elias Mårtenson

I just find it horribly boring, and I was just venting here. :-)

@dzaima Yes. It's a good point.

dzaima

my BQN test stuff - test is the actual tester, rest are files with tests (except types which tests for behavior equality over different datatypes of equal things)

Elias Mårtenson

@dzaima That's pretty neat.

So the first # is the separator between code and the expected value?

I was looking at this one: github.com/dzaima/BQN/blob/master/test/shape#L38

dzaima

5:48 PM

@EliasMårtenson yeah. (# is the comment character in BQN, so i allow comments after the expected value)

Elias Mårtenson

@dzaima I like that one.

dzaima

@dzaima (to note is that doesn't actually test many builtins, but rather language features. i rely on marshalls ones for builtin stuff)

Elias Mårtenson

@dzaima The reason I'm building mine in Kotlin is that it integrates with the Kotlin test framework, so when I get failues it integrates nicely in the development tool.

I get nice stack traces and can very quickly find the offending code.

dzaima

@EliasMårtenson that specific tests message is about overtaking in reshaping (like the behavior of 12↑,¨⍳10), and what that specifically is might change but the test will always work

dzaima

6:22 PM

@dzaima testing performance is incredibly annoying. The timings of an expression often stabilize in a session, only to stabilize to a different value (with a 5% difference) in another, and i'm definitely dealing with something that needs to be more precise than 5% :|

Java

Transcript for