@rak1507 There was an issue with the interpreter when using a specific construct, and it could only happen with ⎕IO←0 ;-)

I require tips for writing complex language interpreters in APL

@Razetime You want to write a parser in APL?

yes, a parser

@Razetime Marshall's mini-series may be of interest, but also these items from dfns.dws and eval, min, and max.

Alternatively, you could write your language as a DSL that transpiles to APL.

@Adám what does this entail? I'm trying to make one for a relatively simple language

Knight

@Razetime If your language can have a syntax that is close to that of APL, then you can give each syntactic element a name, and (if needed) use string replacement to transpile your language to the DSL.

hm alright

:58345154
 answer←'uh oh'
 :If x<0
    :Trap 0
       answer←÷x
    :EndTrap
 :EndIf
 ⎕←answer

@MortenKromberg Sure, but the "else" block might be more than just a value, and you might not want to or be able to compute it unless the condition is false/the error is hit.

Understood, my answer is essentially "don't try to do that". As I understand it, you are using error trapping as PART of your validation and want to combine it with regular control structures. I do not think it is generally desirable to try to create some kind of Frankenstein monster syntax to make that easy. :OrIf is confusing enough already, imagine if it were blended with error handling...

Even :Finally doesn't solve this problem, does it?

That would depend on what you are trying to do, but not in general.

the comp.lang.apl newgroup is being spammed

Does that do what you are trying, even if you don't like the technique?

:58352894 DoThis←{
    b←...
    b:DoThat ⍵
    0::DoThat ⍵
    ...
}

Ok, how do I paste in a multi-line response and have in response and have it ping properly?

@PaulMansour You don't. ¯\_(⍨)_/¯

I always resort to using a non-code leader line with the response tag, and then adding the code block after.

ok, good to know.

You've probably told me that before, but I don't comment enough to remember...

I don't think I have.

@Adám you have

Oh. ⍨

Added to the wiki.

The above style is one I use a lot, though not necessarily for mixing guards and error guards. It requires naming functions, which is hard, but I think worth the effort for significant code.

@PaulMansour Well, kind of. You're missing the error-guarded block, which will end up separating the two paths to DoThat from each other. That's not so good.

Actually, that's not true what I just wrote.

That block is the lower ...

@PaulMansour I don't think the naming is an issue, but I still don't like having to refer to the name twice, and ToThat might need a dummy argument.

DoThis←{0::DoThat ⍵ ⋄ b:∘ ⋄ ...} :p

@dzaima Yeah, I actually asked this because I spotted such code.

@Adám ah. For dfns it's imo better than anything else i've seen, even if it's not nice

@Adám True. Neither of those issues bother me I guess.

See it online!

Hello, I am trying to solve P1/2015 of problems.tryapl.org and I have solution that works on tryapl.org, but that gives a syntax error on problems.tryapl.org when I click on the "Test" button. It is: {d[⍋d← (a∊⎕A) / (a ← 1 ⎕C ⍺)] ≡ c[⍋ c ← (w∊⎕A) / (w ← 1 ⎕C ⍵)]}. What could be the reason that it works on one website but not on the other?

it may be the use of ⎕C

Ok, thanks! I will then search for alternatives to converting case.

@p.vitzliputzli 1 (819⌶) might work.

819⌶ was the old version of the ⎕C function

yep, it works.

@Adám might be a good idea to offload problems.tryapl.org's evaluation things to the newer tryAPL instead of tio

Yes, it worked! Thanks a lot!

no problem.

Feel free to come back if you have trouble with any other questions.

Thanks, actually I am wondering if there is something like the dyadic compose (&) of J available in Dyalog APL that would allow me to do something like x u&v y ↔ (v x) u (v y). In this way I could get rid of duplicating the sorting of the characters in the left and right argument.

@p.vitzliputzli Yes, that is ⍥ but again, only works in the newest version (including TryAPL), but not on problems.dyalog.com

@Adám Wonderful, thanks!

@p.vitzliputzli You can easily implement a model for it: Ö←{(⍵⍵ ⍺)⍺⍺(⍵⍵ ⍵)}

@Razetime Problem is that the testing framework needs unrestricted APL code, and TryAPL uses a strict whitelist.

@Adám Thanks!

@Adám ah that's too bad

@Razetime Maybe I could inject covers for ⍨ and ⍥ and ⍤ and ⎕C and ≠ as those are easy. ⍸ and ⊂ a bit harder, but should be possible too. Then the only problem becomes their inverses, but that's a low cost, I think.

7 CMCs: Dyalog 18.0's ⍨ ⍥ ⍤ ⎕C ≠ ⍸ ⊂ in vanilla 17.1. Read about the enhancements here.

Announcement: Dyalog '21 will be a free online event.

@Adám does this mean there will be 3 grand prize winners + 3 professional winners at Dyalog 22?

@rak1507 I guess.

Yeah, I just got an email about postponing the offline event again

which is actually not bad for me because I'm not vaccinated yet (+ some other reasons)

@Bubbler Right, my wife isn't getting her second shot until a week into September.

@johnnynb Hey there, and welcome to the APL Orchard. Let us know if we can be of assistance with anything.

To check that there are no consecutive digits in a sequence I've come up with ~∨/ 1 ¯1 ∊(¯1↓i)-(1↓i). Is there a way to write in a more compact way the (¯1↓i)-(1↓i) part? It is almost like a fork

you can try 2-/?

@PuercoPop see: aplwiki.com/wiki/Windowed_Reduce

@Razetime Thanks. I didn't know about windowed reduced

what's the simplest way to get the diagonals of a matrix?

@Razetime aplcart.info?q=diagonals

       ,⌿(⊢⌽⍨¯1+⍳∘≢)⎕←3 4⍴⍳12
1  2  3  4
5  6  7  8
9 10 11 12
┌──────┬──────┬─────┬──────┐
│1 6 11│2 7 12│3 8 9│4 5 10│
└──────┴──────┴─────┴──────┘

that is cool

@Razetime I've bookmarked javascript:alert((location+"").replace("https://aplwiki.com/","[apl.")+"]("+location+")") so I can click it from an APL Wiki page and get markdown ready for posting in chat.

nice bookmarklet

I have a few nice bookmarklets.

@Adám this is cool but i was thinking (9) (5 10) (1 6 11) (2 7 12) (3 8) (4)

@Razetime dzaima/APL

something shear the matrix so each row is offset by 1 and then grab columns might work?

idk

antidiagonals builtin?!

@hyper-neutrino yeah I guess that should be it

@Razetime J has that too.

@hyper-neutrino That's what APLcart's entry does, but how do you deal with the raggedness?

yeah... i'm not quite sure how this'd work lol.

shear with padding element a←1+⌈/,matrix and then remove a from each

@Razetime Works for numeric matrices only.

that's what i was thinking. but then you'd need some padding element you can be sure doesn't exist

yeah, that's my use case

Total hack:

      {c/⍨×≢¨c←#~¨⍨⍪⌿(⊢⌽⍨¯1+⍳∘≢)⍵,⍨#⍨¨⍵}3 4⍴⍳12
┌─┬────┬──────┬──────┬───┬─┐
│9│5 10│1 6 11│2 7 12│3 8│4│
└─┴────┴──────┴──────┴───┴─┘

if it's just numeric matrices i guess you could pad with max + 1 and then remove the maximal element from all?

don't see # everyday

@hyper-neutrino That's exactly what Razetime wrote.

oh, lol

maybe use ⎕DR or something to figure out type and change it up for each case

Crazy wasteful (though I'm not sure it is really any worse than the above), but works universally:

a more neat method would be to find a math expression for finding the indices of each diagonal?

      {c/⍨×≢¨c←(⍪⌿(⊢⌽⍨¯1+⍳∘≢)⍵,⍨(⊂⍵)⍨¨⍵)~¨⊂⊂⍵}3 4⍴⍳12
┌─┬────┬──────┬──────┬───┬─┐
│9│5 10│1 6 11│2 7 12│3 8│4│
└─┴────┴──────┴──────┴───┴─┘

@Razetime Yes, but you still have to filter away the out-of-bounds locations.

{⍵⊢∘⊂⌸⍨⍥,+/↑⍳⍴⍵}

Nice.

woah

Should be added to APLcart.

how did you figure that out

Now I remember Marshall doing this once.

It really makes sense.

Simply measure the distance from the top left corner.

All elements sharing distance belong to the same diagonal.

@Razetime from you saying to find a maths expression for the indices

and remembering the pattern +/↑⍳⍴⍵

I guess I'll have to remember that now

wait, wouldn't that (dist from topleft) give antidiagonals?

@rak1507 Same as +/¨⍳⍴⍵

@Adám faster

@hyper-neutrino It does, but ⌽ fixes that.

@hyper-neutrino yes, it's not quite the same as the original question

but easily adaptable

ah. okay

@hyper-neutrino should be enough to ⊖ and then run the function

yeah this solution is clever

@rak1507 (fwiw, shorter in dzaima/APL)

@dzaima cool!

@dzaima Hold on, that only computes the indices, right? Doesn't return the elements.

No, wait, ⊢⌸ is dyadic. Hm.

OK, added to APLcart. Time to make dinner for the family.

is there a reason for ⊂⍤⊢ over ⊢∘⊂ ?

@rak1507 Readability. "Enclose on the right argument" vs "…deferred in preference for the enclosed…"

fair enough, enjoy your dinner

(And since it already uses ⍥, why not f⍤g…)

Thanks!