Funciton
Funciton (pronounced: /ˈfʌŋkɪtɒn/) is a twodimensional, minimalistic, declarative programming language invented by Timwi in 2011.
╓───╖ ║ ! ║ ╙─┬─╜ ┌───╖ ╔═══╗ ┌─────┴─────┤ > ╟──╢ 2 ║ │ ╘═╤═╝ ╚═══╝ ╔════╗ ┌───╖ │ │ ║ −1 ╟──┤ + ╟─┴─┐ │ ╚════╝ ╘═╤═╝ │ │ ┌─┴─╖ │ ╔═══╗ │ │ ! ║ │ ║ 1 ║ │ ╘═╤═╝ │ ╚═╤═╝ │ │ ┌─┴─╖ ┌─┴─╖ │ │ │ × ╟──┤ ? ╟──┘ │ ╘═╤═╝ ╘═╤═╝ └─────┘ │
Example: The factorial function.
(None of the functions seen here (addition, multiplication, greaterthan) are builtin.
All of them have their own implementation in Funciton.)
Contents
Features
 Programs use the Unicode boxdrawing characters to create a graphical representation of data flow.
 Funciton is not electronics. Programs look a bit like circuit diagrams, but they are not circuit diagrams.
 Funciton is invariant under 90° rotations. This means you can take any program or any individual function declaration and turn it by 90° without altering its semantics or the semantics of anything that calls it.
 Funciton currently has only one datatype, the arbitrarysize integer. However, Funciton defines an encoding that expresses arbitrary strings as a single integer.
 Funciton has only five builtin instructions (NAND, lessthan, shiftleft, function invocation and lambda expressions); all other functionality (arithmetic, string handling, lazyevaluated sequences, etc.) is implemented in terms of these.
 Funciton has a fully working interpreter and a compiler to IL; see External resources at the bottom.
Syntactic elements
Function declaration header 
╓───╖ ╒═══╕ ──╢ + ╟── ──┤ + ├── ╙───╜ ╘═══╛ 
Function call 
┌───╖ ╓───┐ ╔═══╕ ╒═══╗ ──┤ + ╟── ──╢ + ├── ──╢ + ├── ──┤ + ╟── ╘═╤═╝ ╚═╤═╛ ╙─┬─┘ └─┬─╜ │ │ │ │ 
NAND or splitter 
│ │ │ ──┴── ├── ──┬── ──┤ │ │ │ 
lessthan and shiftleft 
│ ──┼── │ 
Function return value 
│ ──── │ │ 
Integer literal 
╔═══╗ ║ 1 ╟── ╚═══╝ 
Read string from STDIN 
╔═══╗ ║ ╟── ╚═══╝ 
Lambda expression 
│ │ │ │ ╒═╧═╗ ╓─┴─╖ ╔═╧═╕ ╔═╧═╗ ──┤ ╟── ──╢ ╟── ──╢ ├── ──╢ ╟── ╘═╤═╝ ╚═╤═╝ ╚═╤═╛ ╙─┬─╜ │ │ │ │ 
Lambda invocation 
│ │ │ │ ┌─┴─╖ ┌─┴─┐ ╓─┴─┐ ╒═╧═╕ ──┤ ╟── ──┤ ├── ──╢ ├── ──┤ ├── └─┬─╜ ╘═╤═╛ ╙─┬─┘ └─┬─┘ │ │ │ │ 
Comment 
╔════════════╗ ║ Hi, Mom! ║ ╚════════════╝ 
Introduction
Simply speaking, Funciton programs consist of lines and boxes. Each line carries data from one box to the next.
A Funciton program must have exactly one loose end; a line that just stops somewhere. This is the program output.
Boxes with a doublestruck line and at least one outgoing line are literals. Such a box may contain an integer literal. A very simple program might look like this:
╔════╗ ║ 47 ║ ╚═╤══╝ │
This program prints /
, which is Unicode character number 47.
Boxes with a doublestruck line can be empty, too. In this case, such a box represents the data passed to the program via STDIN. The data is encoded according to the string encoding which is described later in string handling. (Multiple such boxes in a program do not read multiple different inputs from STDIN. The interpreter reads STDIN only once, and then all such boxes return the same data.)
╔═══╗ ║ ║ ╚═╤═╝ │
This program prints its input, so it is a cat program.
NAND and the splitter
A Tjunction in which the two oppositedirection connectors are inputs and the perpendicular one is an output calculates the NAND of two numbers. In C notation, NAND(a,b) = ~(a & b). Consider the following programs:
╔═══╗ ╔═══╗ ╔═══╗ ╔═══╗ ╔════╗ ╔════╗ ║ 5 ║ ║ 5 ║ ║ 5 ║ ║ 0 ║ ║ 2 ║ ║ 1 ║ ╚═╤═╝ ╚═╤═╝ ╚═╤═╝ ╚═╤═╝ ╚══╤═╝ ╚══╤═╝ │ │ │ │ │ │ └───┬───┘ └───┬───┘ └───┬───┘ │ │ │
The above three programs calculate −6, −1 and 1, respectively. (The actual output will be an unprintable character.)
Remember that numbers in Funciton are arbitrarysize integers. Negative numbers are logically modeled by having infinitely many 1bits on the left (2’s complement). 1
has all bits set.
Also remember that Funciton programs are rotationinvariant, so all of the following programs do exactly the same thing:
╔════╗ ╔════╗ │ ╔════╗ ╔════╗ ║ 2 ║ ║ 1 ║ ┌───┴───┐ ║ 1 ╟──┐ ┌──╢ 2 ║ ╚══╤═╝ ╚══╤═╝ │ │ ╚════╝ │ │ ╚════╝ │ │ ╔══╧═╗ ╔══╧═╗ ├── ──┤ └───┬───┘ ║ 1 ║ ║ 2 ║ ╔════╗ │ │ ╔════╗ │ ╚════╝ ╚════╝ ║ 2 ╟──┘ └──╢ 1 ║ ╚════╝ ╚════╝ ╔══════════════════════════════════════════════════╗ ║ All four of the above programs are equivalent. ║ ╟──────────────────────────────────────────────────╢ ║ Also, a doublelined box with no outgoing ║ ║ lines (like this one) is a comment. ║ ╚══════════════════════════════════════════════════╝
A Tjunction in which the two oppositedirection connectors are outputs and the perpendicular one is an input is a splitter. This is not an operation, but simply a syntactic element that splits a line into two. It allows you to access the same value more than once. For example, consider the following program:
╔═══╗ ╔═══╗ ╔═══╗ ║ 1 ║ ╔════════╗ ║ 2 ║ ╔════════╗ ║ 3 ║ ╚═╤═╝ ║ ↓ NAND ║ ╚═╤═╝ ║ ↓ NAND ║ ╚═╤═╝ │ ╚════════╝ │ ╚════════╝ │ └─────┬────────────┴─────┬────────────┘ │ ╔════════════╗ │ │ ║ splitter ↑ ║ │ │ ╚════════════╝ │ │ ╔════════╗ │ │ ║ ↓ NAND ║ │ │ ╚════════╝ │ └───────┬──────────┘ │
This program computes the value (1 NAND 2) NAND (2 NAND 3)
and thus outputs character number 2 (which, again, is an unprintable character).
Whether a Tjunction is a NAND operation or simply a splitter is implicit in which incoming lines are inputs and which are outputs:
╔═════════════════╗ ╔═════════════════════╗ ╔══════════════════════════╗ ║ This is a NAND. ║ ║ This is a splitter. ║ ║ This program is invalid. ║ ╚═════════════════╝ ╚═════════════════════╝ ╚══════════════════════════╝ ╔═══╗ ╔═══╗ ╔═══╗ ──────┬────── ║ 3 ║ ║ 3 ╟────┬────╢ 2 ║ │ ╚═╤═╝ ╚═══╝ │ ╚═══╝ ╔═╧═╗ │ ╔═══╗ │ ║ 3 ║ ├────╢ 2 ║ │ ╚═══╝ │ ╚═══╝
(But notice that the second program is also invalid because it has more than one output.)
Declaring a function
A function declaration consists of a declaration header and one or more loose ends. The header is a box, containing the function name, that has two doublelined and two singlelined edges on opposite sides. A loose end defines a function output.
We can now construct a function for each of the familiar logic gates:
┌──────────────────────┐ ╓─────╖ ╓─────╖ ╓────╖ │ ╓─────╖ │ ║ not ║ ┌──╢ and ╟──┐ ┌──╢ or ╟──┐ ├──╢ xor ╟─────┐ │ ╙──┬──╜ │ ╙─────╜ │ ┌┴┐ ╙────╜ ┌┴┐ │ ╙─────╜ ┌┐ │ ┌┐ │ │ └─────┬─────┘ └┬┘ └┬┘ └───┬──────┤├──┴──┬─┤├─┘ ┌┴┐ ┌┴┐ └────┬─────┘ │ └┘ │ └┘ └┬┘ └┬┘ │ └────┬────────┘ │ │ │
Notice that NOT consists of a splitter and a NAND.
The lines emerging from the declaration header define inputs to the function. Loose ends define outputs. A function can have multiple outputs. The directions in which the inputs “leave” the declaration header and the directions in which the outputs “point” determine how the function can be called; more about this later.
Lessthan and shiftleft
Lessthan and shiftleft are both calculated by a single operator, the cross:
╔═══╗ ║ a ║ ╚═╤═╝ ╔═══╗ │ ║ b ╟──┼──── (a < b) ╚═══╝ │ │ │ (a SHL b)
The lessthan operation returns −1
for true and 0
for false. This enables boolean logic using the NAND operation; if true were represented by 1
, then NOT 1
would be −2
.
The shiftleft operation can also do shiftright simply by giving it a negative operand.
In order to use this to evaluate only lessthan or only shiftleft, we need a little trick.
The Starkov construct
The Starkov construct, named for Cambridge computer scientist and entrepreneur Roman Starkov, who discovered this construct months into Funciton’s existence, is a trick to “nullify” a wire. It consists of a NAND operator that is connected to itself. The following declarations for a lessthan and greaterthan function demonstrate this:
╔╤════════════╗ ╔╤═══════════════╗ ║│ less than ║ ║│ greater than ║ ╚╧════════════╝ ╚╧═══════════════╝ ╓───╖ ┌────────┐ ┌─╢ > ╟──┐ │ ╓───╖ │ │ ╙───╜ │ └─╢ < ╟──┼─────┐ └────────┼─────┐ ╙───╜ ├──┐ │ ├──┐ │ └──┘ └──┘
In these examples, the lessthan part of the cross operator is connected to the function output, while the shiftleft part is connected to a Starkov construct. Since this does not constitute a function output, it is never evaluated and thus effectively nullifies that part of the calculation.
Calling a function
It would be no use declaring functions if you couldn’t call them. A call box consists of two adjacent doublelined edges and two singlelined ones. This is best demonstrated by an example, so let’s define some more comparison operators:
╔╤══════════════╗ ╔╤═══════════════╗ ║│ less than ║ ║│ greater than ║ ║│ or equal to ║ ║│ or equal to ║ ╚╧══════════════╝ ╚╧═══════════════╝ ╓───╖ ╓───╖ ┌──╢ ≤ ╟──┐ ┌──╢ ≥ ╟──┐ │ ╙───╜ │ │ ╙───╜ │ │ │ │ │ │ ┌───╖ │ │ ┌───╖ │ └──┤ < ╟──┘ └──┤ > ╟──┘ ╘═╤═╝ ╘═╤═╝ ┌┴┐ ┌┴┐ └┬┘ └┬┘ │ │
These declarations may at first look wrong. After all, “lessthanorequalto” is not the same as “NOT lessthan”! To understand why this declaration is correct, you have to consider the direction in which data is flowing:
So basically, what goes in to the right of the call node comes out of the left of the declaration box (if the outputs are oriented the same way in the declaration and the call). The fact that the wires take a 180° turn in the code for less than or equal to effectively swaps the order of the operands.
The conditional operator
The conditional operator, written in Clike languages as a ? b : c
, which returns c
if a
is zero and b
otherwise, is equivalent to the expression ((a ≠ 0) NAND b) NAND ((a = 0) NAND c)
. Here we can really show off the splitter: the test for a ≠ 0
is evaluated only once even though it is used in two places.
╔═══╗ ┌───╖ ╓───╖ ║ 0 ╟──┤ ≠ ╟──╢ ? ╟─┐ ╔══════════════════════════════╗ ╚═══╝ ╘═╤═╝ ╙─┬─╜ │ ║ conditional operator ║ │ ├─┬─┤ ║ WITH A PROBLEM (read on) ║ │ │ ┌┴┐ ╟──────────────────────────────╢ │ │ └┬┘ ║ ?(a, b, c) = (a ≠ 0 & b)  ║ │ └─┬─┘ ║ (a = 0 & c) ║ └────────┘ ╚══════════════════════════════╝
Now, I hate to break it with you, but it’s not quite this simple. The above declaration has a big problem which we will come to later.
Addition
Incomplete first attempt
Addition will be our first recursive function. The simplest way to do addition is to add/subtract 1 at a time:
+(a, b) = a ? +(a−1, b+1) : b;
However, this would be incredibly slow (it would be exponentialtime). We can do better — we can do linear time by harnessing the power of the shiftleft. We will implement addition as a recursive function using the following formula:
+(a, b) = b ? +(a XOR b, (a AND b) SHL 1) : a
You can easily convince yourself that this works by considering the following:
 The XOR does the addition on each individual bit, ignoring the carry.
 The AND calculates the carry. SHL carries it to the next bit and then the recursive call adds it to the previous result.
Unfortunately, we run into three problems. The first problem is immediately apparent here:
╓───╖ ┌──╢ + ╟──┐ │ ╙───╜ │ ┌────┴────┬────┴────┐ │ ┌┐ │ ┌┐ │ ┌─┴──┤├──whoops!─┬─┤├─┴─┐ │ └┘ │ │ └┘ │ ╔════════════════════════════╗ │ ┌──────┴───┬┘ │ ║ addition — WITH ║ │ ┌┴┐ ┌┴┐ │ ║ THREE PROBLEMS (read on) ║ │ └┬┘ └┬┘ │ ╟────────────────────────────╢ │ ┌──┴─╖ ┌───╖ │ │ ║ +(a, b) = b ║ │ │ << ╟──┤ + ╟─┘ │ ║ ? +(a^b, (a&b) << 1) ║ │ ╘══╤═╝ ╘═╤═╝ │ ║ : a ║ │ ╔═╧═╗ │ │ ╚════════════════════════════╝ │ ║ 1 ║ │ │ │ ╚═══╝ ┌─┴─╖ │ └─────────┤ ? ╟─────────┘ ╘═╤═╝ │
Problem #1: crossing lines (solution: crossnop)
We need for those two lines to cross. But as we saw before, crossing two lines causes lessthan and shiftleft to be calculated; we don’t want that. We need a cross that is a noop. A crossnop.
Fortunately, there is a way to cross two lines. We can declare a function that takes two inputs and two outputs, and which... does nothing:
╔═══════════════════╗ ╒═══╕ ║ crossnop ║ │ · ├── ╟───────────────────╢ ╘═╤═╛ ║ (x, y) = (x, y) ║ │ ╚═══════════════════╝
Notice that this declaration says that:
 the function has two inputs which are adjacent
 the function has two outputs which are adjacent, and opposite the inputs
 each output is equal to the input opposite to it
which is exactly what we want.
If you are a fan of the wirecrossing problem, you can set yourself the challenge of writing programs without this function.
Problem #2: recursion (solution: shortcircuit evaluation)
Another problem with our addition function above is that it is recursive. A recursive function needs a terminating condition, otherwise it... won’t terminate. The terminating condition in this case is, of course, b
; but we somehow need to guarantee that the rest of the code (the part containing the recursive call) is not going to get evaluated when b
is zero.
To this end, the NAND operator has builtin shortcircuit semantics:
╔═══╗ ╔═══╗ ║ x ║ ║ 0 ║ ╔═══════════════════════════════╗ ╚═╤═╝ ╚═╤═╝ ║ shortcircuit evaluation: ║ ┌─┴─╖ │ ║ the NAND operator returns ║ │ F ║ │ ║ −1 without evaluating F(x). ║ ╘═╤═╝ │ ╚═══════════════════════════════╝ └───┬───┘ │
Whenever the first operand to a NAND operation evaluates to 0
, the second operand is never evaluated. Notice that if the NAND’s output is pointing down, the first operand is on the right; if it is pointing up, the first operand is on the left, and analogously for the other two possible rotations.
Now the problem with the conditional operator we defined before may become apparent. The diagram actually encodes the operation ((a = 0) NAND c) NAND (b NAND (a ≠ 0))
; the second part is the wrong way round because we need to evaluate a ≠ 0
first in order to know whether to evaluate b
. Therefore, we need to flip this around. At this point we run into the linecrossing problem again, so we use our newlydefined crossnop:
╔═══╗ ┌───╖ ╓───╖ ║ 0 ╟──┤ ≠ ╟────╢ ? ╟─┐ ╚═══╝ ╘═╤═╝ ╙─┬─╜ │ ╔══════════════════════════════╗ │ ┌─┴─╖ │ ║ conditional operator ║ │ ┌───┤ · ╟─┤ ╟──────────────────────────────╢ │ │ ╘═╤═╝ │ ║ ?(a, b, c) = (a ≠ 0 & b)  ║ │ └──┬──┤ ┌┴┐ ║ (a = 0 & c) ║ │ │ │ └┬┘ ╚══════════════════════════════╝ │ └─┬─┘ └──────────┘
Problem #3: It doesn’t work with negative numbers
Well, it works with some negative numbers. But there are combinations of inputs for which the recursive formula is endless (the terminating condition is never reached).
Specifically, it works fine only if:
 a ≥ 0 and b ≥ 0, or
 a + b < 0
But it doesn’t manage to do the jump from a negative input to a positive (or zero) output. a XOR b
will be negative and stay negative, and the (a AND b) SHL 1
will be positive and just keep on growing indefinitely.
The successor function
We need a separate function to do the job of jumping from −1
to 0
. We will use the successor function for this, which is defined as:
♯(a) = a + 1
But we can’t define it in terms of addition because we need it in order to do addition, so we need to do something else:
╓───╖ ║ ♯ ║ ╔════════════════════════════╗ ╙─┬─╜ ║ successor function ║ ┌──────────┴────────┐ ╟────────────────────────────╢ ╔════╗ ┌────╖ │ ╔═══╗ │ ║ ♯(a) = a = 1 ? 0 : ║ ║ 1 ╟──┤ << ╟──┴─┬──╢ 1 ║ │ ║ a & 1 ? ♯(a>>1) << 1 : ║ ╚════╝ ╘══╤═╝ │ ╚═══╝ │ ║ a  1 ║ ┌─┴─╖ ┌┴┐ ╔═══╗ ╔════╗ │ ╟────────────────────────────╢ │ ♯ ║ └┬┘ ║ 0 ║ ║ 1 ║ │ ║ Note: can’t actually use ║ ╘═╤═╝ │ ╚═╤═╝ ╚══╤═╝ │ ║ >> because >> depends on ║ ┌──┴─╖ ┌─┴─╖ ┌─┴─╖ ┌─┴─╖ │ ║ ~ (unary minus) which in ║ │ << ╟──┤ ? ╟──┤ ? ╟──┤ = ║ │ ║ turn depends on the ║ ╘══╤═╝ ╘═╤═╝ ╘═╤═╝ ╘═╤═╝ │ ║ successor function. ║ ╔═╧═╗ ┌┐ │ ┌┐ │ ├───┘ ╚════════════════════════════╝ ║ 1 ╟─┤├─┴─┤├──────────┘ ╚═══╝ └┘ └┘
Note that the formula uses both SHL and SHR, but we haven’t defined SHR yet; we need unary minus for that, which we’ll get to later. Therefore, we use SHL −1
.
Introducing private functions
Since the addition function we already defined is potentially an infinite loop, you don’t really want any code to call it other than code that knows what it’s doing. Therefore, we will declare it private. A private function is a function that can only be called by functions within the same source file; in other words, its scope is its containing file. Private functions are marked with an extra little tag in the declaration box as seen here:
╓┬───╖ ┌──╫┘+p ╟──┐ │ ╙────╜ │ ┌────┴─────┬────┴────┐ │ ┌───┐ ┌─┴─╖ ┌┐ │ ┌─┴─┤ ├──┤ · ╟──┬─┤├─┴─┐ │ └───┘ ╘═╤═╝ │ └┘ │ ╔═════════════════════════════╗ │ ┌───────┴───┬┘ │ ║ addition in the case of ║ │ ┌┴┐ ┌┴┐ │ ║ (a≥0 & b≥0)  (¬b)≥a ║ │ └┬┘ └┬┘ │ ╟─────────────────────────────╢ │ ┌──┴─╖ ┌────╖ │ │ ║ +p(a, b) = b ║ │ │ << ╟──┤ +p ╟─┘ │ ║ ? +p(a^b, (a&b) << 1) ║ │ ╘══╤═╝ ╘══╤═╝ │ ║ : a ║ │ ╔═╧═╗ │ │ ╚═════════════════════════════╝ │ ║ 1 ║ │ │ │ ╚═══╝ ┌─┴─╖ │ └──────────┤ ? ╟─────────┘ ╘═╤═╝ │
Finally, we define the real addition function in terms of this private one. We know that the above function works for negative numbers if the result is negative. Therefore, if we can tell (using ≥) that it is going to be nonnegative, just negate both operands; then the result will be negative, which we negate again to get the real result. Since binary negation is actually one away from unary minus, we need to use the successor function to correct the discrepancy.
┌──────────────────────────────┐ ╔══════════════════════════════════╗ │ ╓───╖ │ ║ addition ║ ├───────╢ + ╟───────┐ │ ╟──────────────────────────────────╢ ┌─┴─╖ ╙───╜ │ │ ║ +(a, b) = (a≥0 & b≥0)  (¬b)≥a ║ ┌──┤ · ╟─────────────────┴──┐ │ ║ ? +p(a,b) ║ │ ╘═╤═╝ ┌─────────┴──┐ │ ║ : ¬(♯(+p(¬a, ¬b))) ║ │ │ │ ┌────╖ ┌─┴─╖ │ ╚══════════════════════════════════╝ │ │ └──┤ +p ╟──┤ · ╟──┴──────────────────┐ ┌┴┐ ┌┴┐ ╘═╤══╝ ╘═╤═╝┌───╖ ╔═══╗ ┌───╖ │ └┬┘ └┬┘ │ └──┤ ≤ ╟──╢ 0 ╟──┤ ≥ ╟──┴─┐ │ │ │ ╘═╤═╝ ╚═══╝ ╘═╤═╝ │ │ ┌──┴─╖ ┌───╖ ┌┐ ┌─┴─╖ └──────┬──────┘ │ │ │ +p ╟──┤ ♯ ╟──┤├──┤ ? ╟─────────────────┤ │ │ ╘══╤═╝ ╘═══╝ └┘ ╘═╤═╝ │ │ │ │ │ │ │ └────┤ ┌─┴─╖ │ └───────────────────────────────────┤ < ╟───────────┘ ╘═══╝
String handling
As hinted at in the introduction, Funciton programs receive all text passed in through STDIN encoded as a single humongous integer. The format in which these strings are encoded could be referred to as UTF21: every 21 bits form a Unicode character. This way the complexity arising from surrogates in UTF16 is avoided, while still being more memoryefficient than UTF32.
The least significant 21 bits contain the first character in the string. Thus, if you were to look at the bit pattern of an integer, the string would appear to be stored backwards (unless you habitually write in Arabic, Hebrew, Syriac, or any other RTL script). The operation s SHR 21 removes the first character and thus appears to be shifting the characters in the string to the left.
After the last character in the string, the rest of the integer may optionally be either all zeroes or all ones. The latter allows strings ending in the NUL character (U+0000) to be encoded unambiguously. A compliant interpreter must use the latter if the text provided through STDIN does indeed end with the NUL character, and it must tolerate both formats when interpreting the number returned by the Funciton program.
This takes us directly to some simple string handling functions, string length and string concatenation:
╔════╗ ╔═══════════════════════════════════════════════╗ ║ 21 ║ ║ String length ║ ╚═╤══╝ ╓───╖ ╟───────────────────────────────────────────────╢ ┌───╖ ┌─┴──╖ ║ ℓ ║ ║ ℓ(s) = (s = 0  s = −1) ? 0 : ♯(ℓ(s >> 21)) ║ ┌───┤ ℓ ╟──┤ >> ║ ╙─┬─╜ ╚═══════════════════════════════════════════════╝ │ ╘═══╝ ╘═╤══╝ │ ┌─┴─╖ └───────┴──┐ ╓───╖ │ ♯ ║ ╔═══╗ ┌───╖ │ ┌──╢ ‼ ╟────────────────────────┐ ╘═╤═╝ ┌────╢ 0 ╟──┤ ≠ ╟─┴┐ │ ╙───╜ ┌───╖ ┌───╖ ╔════╗ │ │ ┌─┴─╖ ╚═══╝ ╘═╤═╝ │ ┌──┴─────────┤ ℓ ╟──┤ × ╟──╢ 21 ║ │ └───┤ ? ╟───────────┤ │ │ ╘═══╝ ╘═╤═╝ ╚════╝ │ ╘═╤═╝ ╔════╗ ┌─┴─╖ │ ┌──┴─────────┐ ┌────┴──┐ │ │ ║ −1 ╟──┤ ≠ ╟──┘ │ ┌───╖ ┌─┴─╖ ┌─┴──╖ │ │ ╚════╝ ╘═══╝ │ ┌─┤ + ╟──┤ · ╟──┤ << ║ │ │ └─┤ ╘═╤═╝ ╘═╤═╝ ╘═╤══╝ │ │ ╔════════════════════════════════════════════╗ │ ┌─┴─╖ ┌─┴─╖ ╔═╧═╗ │ ┌────╖ │ ║ String concatenation ║ └─┤ ? ╟──┤ > ║ ║ 1 ║ └─┤ << ╟──┘ ╟────────────────────────────────────────────╢ ╘═╤═╝ ╘═╤═╝ ╚═══╝ ╔═══╗ ╘══╤═╝ ║ ‼(a, b) = ║ ┌┴┐ └──────────╢ 0 ║ ┌┴┐ ║ let c = 21 × ℓ(a); ║ └┬┘ ╚═══╝ └┬┘ ║ (a < 0 ? a + (1 << c) : a)  (b << c) ║ └──────────────┬───────────┘ ╚════════════════════════════════════════════╝ │
With string concatenation implemented, 99 bottles of beer on the wall is now very easy.
List handling
The Funciton language does not have any builtin functionality for lists; but that is not necessary because we can implement lists entirely in Funciton itself. The encoding used to represent lists in a single humongous integer is not fundamental to the language; anyone could come up with a different encoding and implement all the list handling functions for that.
(However, the official Funciton interpreter has a feature that allows you to trace the execution of a specific function. This feature will detect values that encode lists and decode them for the purposes of debugging. The encoding I describe here is the encoding understood by this feature of the interpreter as well as the one used by all the list handling functions.)
Since strings are encoded as integers, a list of integers can automatically double as a list of strings. Furthermore, since lists themselves are encoded as single integers, one can also have lists of lists, arbitrarily nested.
List encoding grammar
Since integers are arbitrarysize, we cannot assume each element in a list fits into any particular number of bits. Therefore, we have to define a variablelength encoding for integers. The encoding I used chops each integer into chunks of 21 bits. (The number 21 is arbitrary, but since characters in a string are already 21 bits, I decided to reuse the number.)
 In the above graphic, the least significant bit is on the left.
 Each element in the list is split into chunks of 21 bits.
 The least significant bits of the final encoding contain the most significant chuck from the element.
 Each chunk is preceded by a continuation bit: 0 = more chunks to come, 1 = this is the last chunk.
 The last chunk is followed by a sign bit. If the sign bit is set, the entire element is stored negated.
Notice that this encoding has several consequences:
 The empty list is represented by the number 0.
 The number 1 encodes a list containing one element, which is 0.
 The number 2 is not a valid list.
 Negative numbers also never constitute a valid list.
 Many of the list functions will get stuck in infinite loops if you pass them an invalid list.
Examples
List containing two items, each ≤ 21 bits long (again, in this diagram, the least significant bit is on the left):
List containing one item between 22 and 42 bits long:
Lambda expressions
Since Dec 2013, Funciton supports lambda expressions (anonymous functions). There are two new types of box for this, summarized in the following:
╔══════════════════════════════╗ ╔══════════════════════════════╗ ║ Lambda expression ║ ║ Lambda invocation ║ ╟──────────────────────────────╢ ╟──────────────────────────────╢ ║ output 1 ║ ║ lambda ║ ║ ↓ ║ ║ ↓ ║ ║ ╔═╧═╕ ║ ║ ┌─┴─╖ ║ ║ output 2 → ─╢ ├─ → input ║ ║ input → ─┤ ╟─ → output 2 ║ ║ ╚═╤═╛ ║ ║ └─┬─╜ ║ ║ ↓ ║ ║ ↓ ║ ║ lambda ║ ║ output 1 ║ ╚══════════════════════════════╝ ╚══════════════════════════════╝
Every lambda has one input (parameter) and two outputs (return values). In cases where only one output is desired, the convention is that the lambda expression returns 0 for the second output and the invocation uses a Starkov construct to discard it.
Since a lambda can easily itself return a lambda, any number of parameters are possible via currying. Similarly, if more than two return values are desired, simply return the first value in output 1 and another lambda in output 2 which will successively produce the remaining values.
Since we already have lists, we can now implement interesting operations that take a list and a lambda. The map and filter functions (wellknown to functional programmers) can be implemented thusly:
┌──────────────┐ ┌──────────────┴─────┐ ╓─┴─╖ ╔════════════════════════════════╗ │ ┌───┬─┐ │ ║ ᴍ ║ ║ map a function over a list ║ │ ┌─┴─╖ └─┘ ┌───╖ ┌─┴─╖ ╙─┬─╜ ╟────────────────────────────────╢ └─┤ ╟──────┤ › ╟─┤ ᴍ ║ ┌──┴──┐ ║ ᴍ(l, f) = ║ └─┬─╜ ╘═╤═╝ ╘═╤═╝ ┌─┴─╖ ┌─┴─╖ ║ let (h, t) = ‹(l); ║ │ ╔═══╗ ┌─┴─╖ └───┤ · ╟─┤ ‹ ╟─┐ ║ l ? ›(ᴍ(t, f), f(h)) : 0 ║ │ ║ 0 ╟─┤ ? ╟───┐ ╘═╤═╝ ╘═══╝ │ ╚════════════════════════════════╝ │ ╚═══╝ ╘═╤═╝ └─────┘ │ │ └─── │ └────────────────────────────────┘ ┌────────────────────────────┐ │ ┌───╖ ┌───┐ │ │ ┌───────┤ ‹ ╟─┘ ┌─┴─╖ │ │ ┌───┴─────┐ ╘═╤═╝ │ Ƒ ╟──┤ ╔══════════════════════════════════╗ │ │ │ └─┐ ╘═╤═╝ │ ║ filter a list ║ │ │ ┌───╖ ┌─┴─╖ ┌─┴─╖ ┌─┴─╖ │ ╟──────────────────────────────────╢ │ └─┤ › ╟─┤ · ╟─┤ · ╟─┤ · ╟──┘ ║ Ƒ(l, f) = ║ │ ╘═╤═╝ ╘═╤═╝ ╘═╤═╝ ╘═╤═╝ ║ let (h, t) = ‹(l); ║ │ ┌─┴─╖ ┌─┴─╖ ┌─┴─╖ ├────┐ ║ let r = Ƒ(t, f); ║ └───┤ ? ╟─┤ ╟─┤ · ╟───┘ ╓─┴─╖ ║ l ? f(h) ? ›(r, h) : r : 0 ║ ╘═╤═╝ └─┬─╜ ╘═╤═╝ ║ Ƒ ║ ╚══════════════════════════════════╝ ╔═══╗ ┌─┴─╖ ├─┐ │ ╙─┬─╜ ║ 0 ╟─┤ ? ╟─┐ └─┘ ├──────────┘ ╚═══╝ ╘═╤═╝ └───────┘ └──
Observe that both examples use only the first lambda output as they use a Starkov construct to swallow the other one.
The following (partial) code would use the above filter function to extract only the odd numbers from a list.
(list) ↓ ┌─┴─╖ │ Ƒ ╟── ╘═╤═╝ ╔═══╗ ╔═╧═╕ ╔═══╗ ║ 1 ╟──┬──╢ ├──╢ 0 ║ ╚═══╝ ┌┴┐ ╚═╤═╛ ╚═══╝ └┬┘ │ └────┘
This example provides a lambda expression that bitwiseands its input with 1 (and sets the second return value to zero as is conventional for singleoutput lambdas), thus returning 0 for even numbers, causing those to be filtered out.
Since the only datatype in Funciton is the arbitrarysize integer, a compliant interpreter must allocate a nonzero integer to every lambda closure the program creates. The lambda expression box returns an integer that identifies the closure, and the lambda invocation box will use the number to identify the lambda closure to invoke. This approach has many advantages; in particular, you can automatically have lists of functions. The integers returned are required to be nonzero as a convenience so that the user code can still use the number 0 to mean null or false in cases where a lambda is optional.
Now that we have lambda expressions, we can do with them some of the things that are typical of functional programming languages. For example, if we take two lambdas that each produce only one output, we can implement function composition easily:
╓───╖ ┌────╢ ∘ ╟────┐ │ ╙───╜ │ │ ┌───┬─┐ │ ╔════════════════════════╗ │ ┌─┴─╖ └─┘ ┌─┴─╖ ┌─┐ ║ Function composition ║ └─┤ ╟─────┤ ╟─┴─┘ ╟────────────────────────╢ └─┬─╜ └─┬─╜ ║ ∘(f, g) = α·f(g(α)) ║ │ ╔═══╗ ╔═╧═╕ ╚════════════════════════╝ │ ║ 0 ╟─╢ ├─┐ │ ╚═══╝ ╚═╤═╛ │ │ └─┘ │ └─────────────┘
Lazyevaluated sequences
Another exciting usecase for lambda expressions is lazyevaluated sequences.
First, we decide to use the number 0
to represent an empty sequence. The lambda expression feature specifically allocates only positive integers to represent lambdas for exactly this kind of purpose. This way we can always use the ?
function to test if a sequence is empty.
Second, we make use of the ability for lambda expressions to return two outputs. The first output returns the first element (the head), while the second returns the rest of the sequence (the tail). If the rest of the sequence is empty, this is 0
; otherwise, it’s another lambda. This way, we can retrieve element after element simply by calling the lambdas until we get 0
.
To illustrate this, let’s write a function that takes an input value and returns a lazy sequence containing just that value and nothing else:
╓───╖ ║ ⌑ ║ ╙─┬─╜ ╔═══╗ ╔═╧═╕ ┌─┐ ║ 0 ╟─╢ ├──┴─┘ ╚═══╝ ╚═╤═╛ │
The function’s argument goes into the lambda expression’s first output, making it the head of the sequence. The second output is set to 0
, which indicates the end of the sequence. The completed lambda expression is returned as the function’s output. (The lambda expression doesn’t care what value you pass into it when you call it, so it uses a Starkov construct to nullify its input.)
We can go further than this. If we wanted a sequence containing lots of repetitions of a value, we could concatenate sequences like the above, but there’s a more elegant way. Consider this function:
╓───╖ ║ ⁞ ║ ╙─┬─╜ ╔═╧═╕ ┌─┐ ┌─╢ ├──┴─┘ │ ╚═╤═╛ └───┴──┐ │
This is almost the same function as above, only instead of the 0
(which would indicate the end of the sequence) it returns another copy of the same lambda. This means that every time you call the lambda, you get the value and another copy of the same lambda. You keep getting the same value and never reach the end of the sequence. We have created an infinite sequence that repeats a value indefinitely.
Since the sequences are lazyevaluated — meaning the next element in a sequence is not evaluated until you ask for it by invoking the lambda — they do not actually have to terminate. You only need a terminating sequence if your code actually traverses the whole sequence until it encounters the 0
that indicates its end.
To round this up, let’s take a look at the truncate function. You could use this to turn an infinite sequence into a finite one for output.
┌────────────┬───────────┐ ┌─┴─╖ │ ╓───╖ │ ┌─┤ ╟─────┐ └──╢ ȶ ╟──┐ │ │ └─┬─╜ ┌─┴─╖ ╙───╜ │ │ ╔═════════════════════════════╗ │ └─────┤ · ╟────┐ │ │ ║ truncate ║ │ ╘═╤═╝ │ │ │ ╟─────────────────────────────╢ │ ┌───┐ ┌─┴─╖ ╔═╧═╕ │ │ ║ Truncates a lazy sequence ║ │ ┌┴┐ │ │ ȶ ╟──╢ ├─┬─┐ │ │ ║ after at most n elements ║ │ └┬┘ │ ╘═╤═╝ ╚═╤═╛ └─┘ │ │ ╟─────────────────────────────╢ │ ┌─┴─╖ └───┘ ┌─┴─╖ │ │ ║ ȶ(q, n) = ║ │ │ ♯ ║ ┌────────┤ · ╟───┐ │ │ ║ let (e, r) = q(0); ║ │ ╘═╤═╝ │ ╘═╤═╝ ├─┘ │ ║ q ? n ? •[e, ȶ(r, n−1)] ║ │ ┌┴┐ │ ┌─┴─╖ │ │ ║ : 0 : 0 ║ │ └┬┘ │ ┌────┤ ? ╟───┘ │ ╚═════════════════════════════╝ │ └───┘ │ ╘═╤═╝ │ │ ╔═╧═╗ ┌─┴─╖ │ └─────────╢ 0 ╟──┤ ? ╟───────┘ ╚═══╝ ╘═╤═╝ │
If the input sequence is empty, or n = 0, return an empty sequence. In all other cases, return a lambda expression that will retrieve the head of the original sequence and then call truncate recursively with n decremented. Eventually this will reach 0 and the sequence terminates. Observe how this will evaluate each element of the original sequence only if and when the new sequence is evaluated. In the same manner, all the other familiar functional constructions, like map, filter, zip, etc. can be implemented, and unlike the map and filter function we defined for lists above, which always evaluate the entire result list, these would be lazy as well and would therefore work on infinite sequences as well.
More functions
The interpreter comes with an extensive library of useful functions, all listed here:
Fundamentals
·  crossnop (used to cross wires, otherwise a noop) 

<  less than 
>  greater than 
=  equal to 
≤  less than or equal to 
≥  greater than or equal to 
≠  not equal to 
?  conditional operator (if a, then b, else c) 
<<  shift left 
>>  shift right 
^  XOR (exclusive or) 
Arithmetic

Prime numbers

String handling
The numbers 0 and −1 both represent the empty string.
ℓ  string length 

‼  string concatenation 
ʃ  substring (string, index, length) 
ʘ  index of first occurrence of a substring, or −1 if not found (haystack, needle) 
…  repeat string a number of times (string, number) 
⇄  reverse string 
↯  convert string to a lazy sequence of its characters 
str→int  parse string containing decimal digits into an integer (0 if any other characters are encountered) 
int→str  convert integer to string 
List handling
The number 0 represents the empty list.


Lazyevaluated sequences
The number 0 represents the empty sequence. Everything else is a lambda expression that returns the head element and the tail. The tail is another lazy sequence, or 0 if the sequence ends here.


Example programs
 Hello, World!
 99 bottles of beer on the wall
 Digital root calculator
 Brainfuckiton (Brainfuck interpreter)
 Quine
External resources
 Funciton Interpreter and Compiler to .NET, written in C# (requires .NET 4.0 or Mono 2.8; compiler requires the Mono.Cecil library)