K is like APL with fewer built-ins. The syntax is ASCII-only, and most built-in functions are heavily overloaded single bytes of ASCII punctuation.
There are several implementations with subtly varying behavior. code.golf uses ngn/k, which is documented with the reference card found under "help" at the online interpreter. If you're new to K (and J, APL, etc...), this style of documentation might not be very helpful, but other learning material is available:
- Razetime has put together an extensive tutorial which introduces the language in multiple chapters, and then spends some time on solving larger tasks (including a Sudoku solver!).
- If you're looking for something in between the ref-card and a full blown tutorial, Stefan Kruger's kbook might be a good a fit.
- Another implementation of a similar K version is oK which comes with a more extensive manual
Example
Let's look at a program that outputs all of the input arguments with their length appended to them: `0:{x,$#x}'x
(Try it on United States: it outputs Nebraska8 Rhode Island12…)
`0:{x,$#x}'x
x Get the list of inputs.
{ }' Apply some lambda function to each input.
({} makes a lambda, and the adverb ' means each.)
x x Inside a monad lambda, "x" refers to the single argument.
# Monad # means "length"
$ Monad $ means "convert to string"
, Dyad , means "concatenate"
`0: Output the list of strings line by line.
0: is a dyadic writing-lines-to-a-file verb,
and the left argument ` (the empty symbol) means "write to STDOUT".
You can see that K is essentially read and written "right-to-left", much like J and APL. x,$#x means (x,($(#x))). There's no operator precedence.
Golf tips
\ preceded by a space is "trace" (an identity function for debugging that prints its argument to STDOUT). This can be shorter than using `0:. Trace however passes its argument through the K formatter before printing, so this may only be useful for numbers.$[x;T;F] can become x{T}/F when x is a non-negative integer, or (F;T)x if x is 0/1 and lazy evaluation is not needed.x in the global scope is a list of the command line args.- Folds can take an optional left argument to use as a seed value. This can often save a byte:
1-+/x → 1-/x, "1.",,/$x → "1.",/$x, ...
- Calling primitives with brackets can sometimes save a byte on parentheses:
((A)+B),C -> +[A;B],C (assume A, B and C are longer expressions).
- Strings are sequences of (signed) bytes and can be used in place of a list of small integers in many cases.
- Each type has its own null value which indicates a missing value, for example not found in the result of find, or out-of-bounds indexing:
| Type |
Null |
Meaning |
| int |
0N |
$-2^{63}$ |
| float |
0n |
NaN |
| char |
" " |
space |
| function |
:: |
identity function |
| symbols |
` |
empty symbol |
Base Conversion
These come up a lot, almost in every other problem. A few ideas can be found on the APL tips and J tips page on StackExchange, so here is a quick translation table:
|
K |
J |
APL |
| Convert from base |
/ |
#. |
⊥ |
| Convert to base |
\ |
#: or #.inv |
⊤ or ⊥⍣¯1 |
n\ combined with strings is very effective at compressing sequences of integers. Example:
28+,/4\"> >/" / days in a month (the string is `c$62 11 62 47)
0/x gets the last value of x. Compared to *|x, it can be combined with other adverbs, e.g. 0/' (last of each), and it returns 0 instead of 0N for an empty x.n/"..." is a good starting point for a string hash.
External resources