Church booleans in Befunge-98

I like this Code Golf question on “Church booleans.” A “Church boolean” is either of the following two functions (in C++17 syntax):

auto T = [](auto t, auto f){ return t; };
auto F = [](auto t, auto f){ return f; };

So you can build up Boolean operations such as

auto NOT = [](auto x){ return x(F, T); };

static_assert(NOT(T)(0,1) == F(0,1));
static_assert(NOT(F)(0,1) == T(0,1));

and

auto AND = [](auto x, auto y){ return x(y, F); };

static_assert(AND(T,T)(0,1) == T(0,1));
static_assert(AND(T,F)(0,1) == F(0,1));
static_assert(AND(F,T)(0,1) == F(0,1));
static_assert(AND(F,F)(0,1) == F(0,1));

See my C++17 solutions here and here (curried).

Incidentally, I was surprised to find that a non-constexpr variable of lambda type can still be called in constexpr context. I’m not sure exactly what’s going on there.


I wondered if the Code Golf puzzle could be solved in my favorite esoteric programming language, Befunge.

I was already confident that there would be no sane way to define a “higher-level function” in Befunge-93. You can reuse code by directing the program’s control flow back over itself, but control flow in Befunge-93 is still entirely static; there’s no way to “name” a function so that it can be “called” from anywhere.

But Befunge-98 is overcomplicated enough that there ought to be a way to do it!

First I looked at the “fingerprint” feature, which lets you use special subroutine-like semantics for instructions A through Z. But there’s no way to define those semantics within Befunge-98. (There is the Dynafing spec, which is close! But by the time I remembered that Dynafing existed, I was on to my next idea.)


I looked for some way to do an “absolute jump” in Befunge-98. If I could do absolute jumps, then I could “name” a function by its two-dimensional address. Just as in real life, “calling” a function would mean pushing a return address and jumping, and “returning” would be implemented as jumping back. Unfortunately, Befunge-98 doesn’t have any “absolute load IP” instruction.

But it does have an “absolute load delta” instruction! The x instruction loads (dx,dy) from the stack; so for example 01x means “go south from the x” and 20x means “go east from the x at double speed.” So if I could find a way to store the current IP to the stack, then I could do some math and then x to accomplish an absolute jump. (Of course anywhere I jumped had better have a > “landing pad,” to curtail my IP’s insane velocity.)

How do I get a copy of the current IP in Befunge-98? It doesn’t seem to be easy. But I found that when you use the { “Begin Block” instruction, in addition to creating a new stack on the stack stack, it also resets the storage offset. It sets the storage offset to the value of the current IP! And then, even better, it pushes the old storage offset onto the second-on-stack-stack.

So { sets the storage offset to the current IP; then a second { pushes that storage offset onto the SOSS; and then I can use u “Stack under Stack” to transfer those coordinates up into my stack. Finally, a couple of 2}s put me back in my original execution context, with a copy of the storage offset.

0{{2u2}2}
    Push your IP's current address.
    0{{2u2}2}  is the same as  y x
    where (x-1, y) is the address of the first `{`

Two-dimensional addresses take up two stack entries each, rendering Befunge’s bare-bones stack-reordering command \ pretty helpless. But with { on the brain, we can write up some helpers:

abcdef n > :2+{\1u\1u\03-u}
    Bury the top-of-stack entry under n other entries.
    12345 1 :2+{\1u\1u\03-u}  is the same as  12354
    12345 3 :2+{\1u\1u\03-u}  is the same as  15234

abcdef n > 1+:{3u\01-u\01-u\}
    Dig up a stack entry from under n other entries.
    12345 1 1+:{3u\01-u\01-u\}  is the same as  12354
    12345 3 1+:{3u\01-u\01-u\}  is the same as  13452

And then we can put together our “find my absolute address” snippet to produce a “jump to this absolute address” snippet.

n >0{{2u2}2}$-063*-\x
    Jump to (0, n), where we hope to find a `>` landing pad.

That snippet assumes that the >0{… begins in column 0 so that the x instruction itself appears in column

  1. If it begins in some other column, you’ll have to adjust the 63*- accordingly. And if you want to actually detect which column it begins in, by saving the x-coordinate returned from 0{{2u2}2}, then the code gets much more complicated; it’ll have to use the dig/bury helpers above.

Incidentally, this helper is also useful:

> N1+u:N2+{1u\1u\03-uN1+}0N1+-u
    Copy the entry from under N other entries on the SOSS, to the top of the TOSS.
    (Remember that the SOSS is 2 entries bigger than you might think, because of the saved storage offset.)
    12345 0{ 678 31+u:32+{1u\1u\03-u31+}031+-u  is the same as  12345 0{ 6784
    67890 0{ 123 41+u:42+{1u\1u\03-u41+}041+-u  is the same as  67890 0{ 1238

So, we can make our Church booleans! Our input expression will be a stack of function names (that is, line numbers; that is, y-coordinates), which we feed to our fundamental building block EXEC, defined as

> 0{{2u2}2}$-0a9+-\x

EXEC will pop a function name from the stack and jump there. Each of our Church-boolean conjunctions will pop its operands and then EXEC the computed result. So TRUE and FALSE might look like

> $ 0{{2u2}2}$-0ab+-\x  FALSE
> \$ 0{{2u2}2}$-0ac+-\x  TRUE

OR might look like

> 3\ 0{{2u2}2}$-0ac+-\x    OR (assuming TRUE is on line 3)

For example, if the stack (from top to bottom) holds OR TRUE FALSE 17 42, we’ll kick things off by jumping to OR, which conceptually pops TRUE FALSE and execs TRUE (but in fact tucks TRUE and then jumps to TRUE, which itself pops TRUE FALSE and execs TRUE). Anyway, TRUE then pops 17 42 and jumps to 17. Line 17 might be a function that prints “the result is true” and halts.

This gives us something kind of like Polish notation… but not really, because of how it never distinguishes “operators” from “values.” For example, if the stack holds OR TRUE NOT TRUE 17 42, we’ll exec OR, which pops TRUE NOT — and then since NOT isn’t a Church boolean, the whole thing derails. We end up exec’ing TRUE, which pops TRUE 17 and execs TRUE, which pops 42 0 and execs 42. I’m not sure what to call this kind of execution model, nor whether it’s good for anything.

Of course in order to be able to “name” TRUE, the programmer-of-OR has to know what line number TRUE is on. And if the programmer-of-TRUE knows what line number TRUE is on, then TRUE doesn’t need the 0{{2u2}2} dance to figure out its own y-coordinate.


Conclusion: I made “Church booleans” in Befunge-98, and found a way to jump to any absolute address from within a Befunge-98 program (even if it ended up not really being needed in this case), and ways to dig and bury entries on the Befunge-98 stack.

Posted 2019-08-24