@EdgyNerd APL. For sure. Only exception is J if it happens to have a needed built-in, like for primes or diagonals.

@Adám Sure, no problem

@dzaima Is there any way in dzaima/APL to catch "unknown token" errors?

@dzaima Why does ^ give NYIError: no built-in ^ defined in exec while other symbols give unknown token?

@Adám not currently, though I have no idea why I made them quietly just get skipped

@Adám I just have a list of symbols that might or might not be candidates for built-ins (because code-page), but the code execution part doesn't recognize them (i.e. ⌻⌼⍃⍄∍⋾∞ & many more do that too)

@dzaima Why does it have duplicates?

@Adám it doesn't?

the \\ is an escaped backslash

@dzaima Try it online!

@Adám the "+-/⍳⍬⍴∘⎕⊂÷⍺⍵≢¨" at the end is still a comment

@dzaima Ah, underhanded

@dzaima Perfect symbol for non-enclosing reduction: ⟉ :-)

@Adám pushed making unknown tokens errors (though to note is that {$} alone is gonna error, unlike dyalog)

@dzaima That's fine. I'd use ⍎ to test with anyway.

I wrote a program to test APLs for homoglyphs:

 dialect Test tests;Run;t;url;code;⎕IO;Hex
 ⎕IO←0
 Hex←(⎕D,⎕A)⌷⍨⊂
 url←'raw.githubusercontent.com/abrudz/TIO/master/Run.aplo'
 code←(⎕SE.SALT.New'HttpCommand'('GET'url)).Run.Data
 2 ⎕FIX''⎕R''⍠'ResultText' 'Nested'⊢code
 :For t :In tests
     :Trap 0
         :If ~':'∊('apl-',dialect)Run⍨t
             ⎕←⊃{⍵,': ',((⎕UCS 16∘⊥),' U+',Hex)⍎⍺}/'\(⎕UCS 16⊥(.*)\)' '⍝ (.*)'⎕S'\1'⊢t
         :EndIf
     :EndTrap
 :EndFor

Where test is vector of character vectors:

⍎⎕←'1',(⎕UCS 16⊥2  2  0 10),'1' ⍝ Epsilon
⍎⎕←'1',(⎕UCS 16⊥2  2  0  8),'1' ⍝ Epsilon
⍎⎕←'1',(⎕UCS 16⊥2  2  1  2),'1' ⍝ Minus
⍎⎕←'1',(⎕UCS 16⊥0  0  2 13),'1' ⍝ Minus
⍎⎕←'1',(⎕UCS 16⊥2  2  2  3),'1' ⍝ Modulus
⍎⎕←'1',(⎕UCS 16⊥0  0  7 12),'1' ⍝ Modulus
⍎⎕←'1',(⎕UCS 16⊥2  2 12  6),'1' ⍝ Star
⍎⎕←'1',(⎕UCS 16⊥0  0  2 10),'1' ⍝ Star
⍎⎕←    (⎕UCS 16⊥2  2  3 12),'1' ⍝ Tilde
⍎⎕←    (⎕UCS 16⊥0  0  7 14),'1' ⍝ Tilde
⍎⎕←'1',(⎕UCS 16⊥2  3  7  1),'1' ⍝ Nor
⍎⎕←'1',(⎕UCS 16⊥2  2 11 13),'1' ⍝ Nor
⍎⎕←'1',(⎕UCS 16⊥2  3  7  2),'1' ⍝ Nand

@Adám well, the only one dzaima/APL has is ⍩ & ᐵ which isn't on the test :)

@dzaima Sure, and there are a lot more that could be added, but they are either dialect specific or no dialect support any alternatives, so they don't really contribute to a comparison across dialects.

@dzaima the inverse of an inverse of a function with a defined obverse borks tio.run/##SyzI0U2pSszMTfz/P@1R2wQNw0cdM7Q1H/…

@Adám also for apl-extended

@KritixiLithos just (-⍣¯1⍣¯1) 10 alone doesn't work. I just haven't defined inverse for ⍣

@KritixiLithos Yeah, that I'm aware of due to how it is implemented. Too complicated to fix. Let's hope for a native ⍫ in 19.0

@KritixiLithos fixed

@Adám pretty sure this is wrong

Me too. ⍨

tried implementing inverting of ⍨ and got the same bug \o/

my problem was that I implemented inv⍵[A (f⍨) B] as B (inv⍵[f]) A, but it should be B (inv⍺[f]) A (inv⍵[x] being like ⍣¯1 and inv⍺[x] being f⍨⍣¯1⍨)

hard to test for me as almost no built-ins in my apl support inv⍺ and it can't be defined by ⍫ ⍨

oh, my derived monadic ops don't even have the function calls for inv⍺ yet :|

@dzaima :D

still have to understand and define inv⍺[A(f⍣C)B] though

@dzaima dyadic ops don't have it either. yay for past me laziness

@dzaima is (X (f⍣C) A) ≡ B even solvable for X in basic operations?

@dzaima I can tell you that Dyalog's internal system for inversions is complex indeed, and there are even cases that will crash the interpreter:

⎕←⍕⍣¯1⊢''

@Adám
VALUE ERROR

Or not ;-) Well, try it online!

@dzaima I could defined it for when C∊¯1…1 but I believe all other cases would require more info from f than inv⍺[f] and inv⍵[f]. is it worth it?

tio.run/##nVRBb9MwFL7nV/…

@Adám ∧

@KritixiLithos Huh, you fixed it?

@dzaima how does that inv notation work?

@Adám apparently, I modified obv and the obv∧... checks

@KritixiLithos inv⍵[A f B] asks to solve B ≡ A f X for X; inv⍺[A f B] asks to solve A ≡ X f B

how does dyadic inverse work anyways? -⍣¯1 appears to be -

@KritixiLithos OK, pushed. However, I fear something has happened to Dennis.

nvm figured it out, 10∘-⍣¯1⊢2 ≡ 10-⍣¯1⊢2

@dzaima i.e. inv⍵[A + B] (or just inv⍵[+]) is B - A (or just -⍨) because B = A + (B - A)

@KritixiLithos X(f⍣¯1)Z asks for a Y such that Z≡X f Y

@KritixiLithos Yeah, and @ behaves the same.

am i correct that Dyalog is using a numerical solver for 9((1+×)⍣5⍨⍣¯1)10?

@dzaima is that not X(f⍣¯C)B?

@KritixiLithos that ⍨ there breaks that.

which ⍨?

@dzaima I don't know of a way to detect that without attaching a debugger, other than measuring runtime. This one sure is very fast, like 100 ns or so.

@KritixiLithos the only one?

where is ⍨ in ` (X (f⍣C) A) ≡ B `?

@KritixiLithos oh, i didn't look at what message you were replying

@KritixiLithos we want to solve for X

ah, my bad

⋄f←((1+×)⍣5⍨)⋄⎕←9 f 9 f⍣¯1⊢10

@KritixiLithos
0.8999371946

I'm confused, shouldn't this give 10?

tl;dr f⍣A⍨⍣¯1 is broken

afaict the right action is either to do numerical approximation or domain error (or solve the function symbolically though that's probably way too much work)

@KritixiLithos Yeah, this sure looks suspect:

⋄ f←1+×
⋄ ⎕←9 f  ⍨⍣¯1⊢10
⋄ ⎕←9 f⍣1⍨⍣¯1⊢10

@Adám
1
0.8

j can't even do 9 (1+*)^:_1]10

⍨

@KritixiLithos dzaima/APL can't too, but I'm pretty sure inv⍵[A f g] is implementable

half-relatedly, why is f←⍣¯1 not allowed in TIOs code box but is in the REPL?

@dzaima ikr. Somehow, ⎕FIX doesn't like tacit operators. Even ¨ fails. Hm, I thought I had logged that one. I'll do so now then.

@dzaima Ah, it has been fixed in 18.0.

well i mean it's not wrong

implemented stuff. that's enough of ⍣¯1 for today.

@dzaima That's one epic commit message: f⍣¯1[⍨]⍣¯1, ⍟⍨⍣¯1, -⍨⍣¯1, +⍨⍣¯1, *[⍨]⍣¯1, ⍨⍣¯1, (A f g)[⍨]⍣¯1

@Adám actually i found a silly mistake, 1⊣⍺⍺ ns borks for non-obverse functions

maybe check that it is an object, then 1 if .NrmFn exists 0 if it fails

@KritixiLithos Can you PR?