on filling arrays very quickly news.ycombinator.com/item?id=22104576

(re: fast 'where', and figuring out what k lacks here in its primitives to make this fast - perhaps 'where' is meant to be the way k does fast fills)

@JohnEarnest reticulum in special k is so cool. the code output reminds me of ispc ispc.github.io

i assume with seeded over/scan taking the form seed dyad/:list form, the monadic form (for situations like {~10=#x} (1,)/:1) will also arrive, but is currently broken

(aside: should the condition be flipped? ie keep going until?)

i guess this way the condition can use the same truthiness as triadic $ ie empty list is false

what stops the monadic converge/do/predicate forms living on the no-colon version?

the fn will need to be explicitly monadic so that it's disambiguated from dyads so that it can share with seeded each/over, anyway

Can functions have more than 3 arguments in K?

Ok, apparently they can. I just accidentally separated them with commas instead of semicolons

@ksi nice improvement

so, you're building an index matrix like (0 1 1 1;1 0 2 2;2 2 0 3;3 3 3 0) instead of (0 1 2 3;1 2 3 0;2 3 0 1;3 0 1 2) and that improves locality when joining with ,/?

I don't understand why in k, you can compose function from left to right like this:

f: *400> which take first argument, then test with 400>.

@HaoDeng are you sure it doesn't evaluate 400> before * ?

@ngn You are right :)

They are actually exchangeable ( commute).

@StevanApter aha - got it

@ngn probably, but the speed difference is much smaller in ngn/k, maybe it's more useful when joining in parallel? i am not sure whether k9 does it in parallel.

@ksi it doesn't do it in parallel (yet?) - i can see that in the cpu usage widget on my taskbar

@ksi i suspect k9's ,/ is optimized. there's a big difference when i replace ,/' with ,'/+

,/ could compute the length of the result before it starts filling it in. in ngn/k i don't do that yet.

@ngn i see, otherwise you need keep copying the intermediate result as it grows

@ksi yep

btw, i was thinking of introducing some kind of buffer-like structure for the general case, when the length is not known in advance

and the items of the result are appended one by one

@ngn reinventing vector/ArrayList? (in dzaima/APL i got a 2x speedup in permutations from switching from doubling size on full to precalculating size)

to avoid unnecessary copying, the buffer could be realized as a list of lists - each being twice as large as the previous. append() appends to the last one if there's enough room, otherwise allocates a new one. this way, once the operation is done, it's easier to combine the chunks because the precise length is known.

@dzaima actually trying to avoid what they do

@ngn hm, that could be pretty good for an append-only array. i did a similar thing once for the layers of a binary heap

@dzaima yeah, i mean it for append-only, used internally for things like each with a user-defined function

bad example, sorry

things like converge-iterate

@ksi permutations and combinations are much faster in ngn/k at the moment because it can use chars as ints :)

@ngn What's the algorithm of the P function in e.k? especially the magic binary number

@HaoDeng it's the standard sieve of erathostenes with tweaks - no even numbers, and initially it's populated with a pattern that already has punched holes for small primes

oh, and it's recursive - it generates primes up to sqrt(n) and then uses them to sieve the rest

Thanks. like the wheel factor but for first 15 numbers.

@HaoDeng the evens are missing, so 30 numbers - that's 2*3*5

i could have done 2*3*5*7 but it was getting rather long

@ngn oops, i should learn how to spell "eratosthenes"

How do we write to stdout? Is there anything like Q's 0N! which print any x on stdout, and return x?

@HaoDeng in k9? " \" (space backslash)

Thank you. Yes.

Do you have a document somewhere?

@HaoDeng in the repl "\" (backslash without a leading space) prints a one-page summary

@HaoDeng there's also "k9 simples" - unofficial documentation