stupid bot was starred D:

@J.Sallé @dzaima another approach is to encode in base 100 and then encode each in base 10 10

the shortest i can get that way is 29 bytes, left as an exercise to the reader :)

I have 34 :|

@dzaima 27

i haven't given out apl bounties lately, this looks like a good opportunity

I'll reward the shortest Dyalog APL answer to the digit pairs challenge published within 3 days from now with +50 rep for every byte below 35 (max +500 rep).

@Adám please pin ^

that was quick :) 26 is also possible

is there a reason that an A f train doesn't exist? e.g. (2+) 3 → 5

@dzaima 2+⊢ is equivalent to {2}+⊢, so would 2+ be equivalent to {2}+?

@ngn hm, I always thought of arrays in trains as immediately assigning themselves as the left argument of the function to the right, not as fns returning themselves

@dzaima well, arrays in trains do affect the odd-even pattern, for instance -*1+⊢ works like -*({1}+⊢), not -(*(1∘+)⊢)

what do you think A f g h should do in your model?

@ngn ((A f ⍵) g (A h ⍵)), dyadically erroring (or ignoring ⍺)

@dzaima interesting... what about A f B g? I suppose that's A f (B g ⍵)?

@ngn that doesn't currently work (as the 2-train is only checked at the end), but yeah, seems reasonable enough to be useful

and now (1+2×)5 returns 11

@ngn really the even-odd pattern doesn't apply in my model anymore, but yeah, instead of immediately, I should've said after grouping the trains

@dzaima so, it's similar to how it works in k - there's no even-odd pattern there

(train)x is simply train x, no tricky rules

@ngn I should've probably learned at least a bit of K and J before making an APL derivative :p

@ngn does that mean that the classic +/÷≢ equivalent sums the reciprocal of the length?

@dzaima I think it's a useful experience to implement any of them. Each of the three has its own unique flavour.

@dzaima yes, except that the symbols are a little different

and no primitive is wasted for reciprocal: x%y is divide but %x is sqrt

@ngn My APL still has forks - (+/÷≢)1 5 6 still returns 4 just that I can do cool stuff like (1++/÷≢)1 5 6 for 5

@dzaima well, if I understood correctly, that particular fork is compatible with Dyalog and ngn/apl, but +/÷≢1+⊢ probably isn't

@ngn why does that not work in ngn/apl?

oh, you just don't support have broken A f g forks

@dzaima huh... I've broken something there

ok, now it works (but it takes some time for gitlab to flush its caches)

every time I look at my own code after a long pause, it looks horribly verbose :)

@ngn what should be incompatible there?

@dzaima didn't you say arrays in trains are bound to the function immediately to the right?

I imagine it's like +/÷≢(1∘+)⊢

@ngn I corrected myself later

ah, sorry

@ngn argh i want the reciprocal built-in but sqrt fits so nicely on ÷

@dzaima you have plenty of unicode chars :)

oh wait i was about to say that making √ only for monadic would feel pointless but then realized that there are roots other than square roots

@dzaima Yeah, I've long been wondering why APL hasn't had (square)root built-in. Well, now in Dyalog has *∘.5/*∘÷⍨ but those are pretty awkward for such a fundamental function. NARS has √.