twopn/site-src/ccgoto.mdx
2026-02-19 19:29:00 -05:00

351 lines
11 KiB
Text
Raw Blame History

This file contains ambiguous Unicode characters

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

---
title: Emulating GOTO in Scheme with continuations
description: GOTO sucks and is evil and I hate it, but what if there were parentheses? `call/cc` is kinda like goto, so lets use it to make goto.
tags: post,short
date: 2026-02-18 14:28:46 -5
---
export function Basic() {
return <abbr title="Beginners All-purpose Symbolic Instruction Code">BASIC</abbr>;
}
In his 1968 letter, [<i>A case against the GO TO statement</i>](https://www.cs.utexas.edu/~EWD/transcriptions/EWD02xx/EWD215.html)
(known only by that name), Dijkstra said “[t]he go to statement as it stands is just too primitive, it is too much an invitation to make a mess of ones program.”
Unfortunately, scheme programmers arent given that invitation.
Thats no fair!
Fortunately, scheme has a procedure, `call/cc`, that we can use to emulate the style of control flow that `GOTO` provides.
We can use syntactic abstraction to invite scheme programmers to make a mess of their programs in a limited context.
## How `GOTO` works
Odds are, you know how `GOTO` works, but lets briefly review.
Perhaps youve seen a <Basic/> program that looks something like this:
```basic
10 PRINT "Hello, world!"
20 GOTO 10
```
This, as you may have guessed, outputs:
```text
Hello, world!
Hello, world!
Hello, world!
Hello, world!
...
```
…forever.
Normally, control proceeds from the lowest line number to the highest line number, but the `GOTO` statement “jumps” to the given line, no matter where it is.
(Forgive my imprecision, this is a basic tutorial, not a <Basic/> tutorial.)
Youre more likely to see `goto` in `C`:
```c
void do_something() {
char *important_stuff = (char*)malloc(/* ... */);
FILE *important_file = fopen(/* ... */);
// do stuff...
if (errno != 0) goto cleanup;
// do more stuff...
if (errno != 0) goto cleanup;
printf("Success!\n");
// control falls through even if everything goes well
cleanup:
free(important_stuff);
fclose(important_file);
}
```
Using `goto` here lets us avoid repeating the `cleanup` logic.
Not my thing, but this is what most `goto` fans like.
In `C`, `goto` uses `labels:` instead of line numbers, and it cant leave the function, but otherwise it is substantially similar to <Basic/>s `GOTO`.
Hopefully you understand `goto` now. It lets you jump around.
The second thing you need to understand before we can implement `goto` with `call/cc` is how `call/cc` works.
## How `call/cc` works
`call/cc` is short for `call-with-current-continuation`.
`call/cc` takes one argument, a procedure, and returns the result of applying that procedure with the current continuation as an argument.
What is “the current continuation?”
Lets start with an example.
```scheme
(define cont #f)
(begin
(+ 1 (call/cc
(lambda (k)
(set! cont k)
0)))
(display "The number is: ")
(write (cont 41))
(newline))
```
You might call this example contrived.
That is because I contrived it to be an example.
Lets run it anyway. It outputs:
```
The number is: The number is: The number is: ...
```
…forever‽
`cont` is actually something like
```scheme
(define cont
(lambda (x)
(+ 1 x)
(display "The number is: ")
(write (cont 41))
(newline)))
```
In this form, the unconditional recursion is obvious.
Continuations are a lot like procedures, but they dont necessarily come back to where you called them.
This example demonstrates that difference in behavior:
```scheme
(define (displayln obj)
(display obj)
(newline))
(define cont #f)
(displayln
(call/cc
(lambda (k)
(set! cont k)
"cont set")))
(begin
(displayln "procedure called")
(displayln "after procedure call")
(cont "continuation called")
(displayln "after continuation call"))
```
This outputs
```text
cont set
procedure called
after procedure call
continuation called
```
Notice how after calling a procedure, in this case `displayln`, the output continues but not after calling `cont`.
When we call `cont` with a new value, its like we ran the same code but chose another value—this
is the principle that underlies the
<a id="Sitaram-link" href="https://ds26gte.github.io/tyscheme/index-Z-H-16.html#TAG:__tex2page_sec_14.1"><code>amb</code>iguous choice</a>
operator.
The `k` that `call/cc` calls its argument with represents, roughly, the rest of the computation.
The “current continuation” is what will be executed next at the point that `call/cc` is called.
Incidentally, this helps me understand schemes multiple return values; `(values v1 v2 ...)` is just <code className="long">(call/cc (lambda (k) (k v1 v2 ...)))</code>.
I recommend reading about continuations in Dybvigs [<i>The Scheme Programming Language</i>](https://www.scheme.com/tspl4/further.html#g63)
if youre (justly) dissatisfied with my explanation or just want to learn more precisely how they work and their applications.
We now have a decent understand of how `call/cc` works, so lets finally use it to implement `goto` in scheme!
## `goto` in scheme
Here you go:
```scheme
(define-syntax with-goto
(syntax-rules ()
[(_ goto rest ...)
(let ()
(define goto #f)
(%labels rest ...)
(call/cc
(lambda (k)
(set! goto
(lambda (label) (k (label))))
rest ...)))]))
(define-syntax %labels
(syntax-rules ()
[(_) (begin)]
[(_ (_ ...) rest ...) (%labels rest ...)]
[(_ label rest ...)
(begin
(define (label) rest ...)
(%labels rest ...))]))
```
Lets run that with our favorite [<abbr title="Revised Revised Revised Revised Revised Revised Report on the algorithmic language Scheme">R⁶RS</abbr>](https://www.r6rs.org/) implementation (mine is [Chez Scheme](https://cisco.github.io/ChezScheme/)):
```scheme
(with-goto goto
loop (display "Hello, world!\n")
(goto loop))
```
```text
Hello, world!
Hello, world!
Hello, world!
Hello, world!
...
```
Heres an example that doesn't loop forever:
```scheme
(let ([x 1])
(with-goto go
(go loop)
double
(set! x (* 2 x))
loop
(display x) (newline)
(when (< x 1000)
(go double))
(display "done\n")))
```
It outputs:
```scheme
1
2
4
8
16
32
64
128
256
512
1024
done
```
Ill explain this macro one part at a time.
First,
```scheme
(define-syntax goto
(syntax-rules () [...]))
```
defines `goto` as a syntax transformer (more precise name for a macro) using the `syntax-rules` pattern-matching language.
The `()` after `syntax-rules` is the empty list of literals; we don't have any special words here, so it doesn't apply.
You can read more about how `syntax-rules` works in [TSPL](https://scheme.com/tspl4/syntax.html#./syntax:s14), but we'll only be using the most basic features here.
The important thing is to know that matched names are replaced in the output and that `x ...` matches/splices zero or more expressions.
Also, `syntax-rules` is hygienic, so dont stress about name collisions.
Then, we match `(_ goto rest ...)`.
Anything else is a syntax error.
The `_` is for `with-goto` (we do not want to repeat ourselves).
We output a big `let` expression.
Notice how the second example uses `go` instead of `goto`?
That's because the first element in `with-goto` is the name of the `goto` procedure.
We `define` it as false because we will set it later.
Next, we pass the body (`rest ...`) to `%labels`, which deserves its own heading.
## Extracting labels
`%labels` is a syntax transformer with three cases:
1. `(_)` Nothing is passed: `(begin)` (do nothing)
2. `(_ (_ ...) rest ...)` A list is passed: Ignore it and process `rest ...`. We treat expressions of the form `(x ...)` as statements, not labels.
3. `(_ label rest ...)` Finally, a label!
When we encounter a label, we define a thunk (procedure that takes no arguments) with the rest of the arguments as its body, like so:
```scheme
(define (label) rest ...)
```
Putting it all together,
```scheme
(%labels
a
(display 1)
b
(display 2)
c
(display 3))
```
(morally) expands to
```scheme
(begin
(define (a)
(display 1)
b
(display 2)
c
(display 3))
(define (b)
(display 2)
c
(display 3))
(define (c) (display 3)))
```
(The leftover labels have no effect)
This helper on its own is a really crappy way to define functions with shared tails, so lets bring it all together.
## Going to
We have our labels as functions, but what for?
If we call these procedures, they will return control to us, so they arent like `C` labels at all.
Well, remember how I said that continuations dont necessarily come back to where you called them?”
Were going to exploit that property to implement `goto`.
We wrap the body of `with-goto` in `(call/cc (lambda (k) ...))`.
Inside the body, if we call `k` like `(k (label))` we effectively replace the body with the result of calling `label`.
We accomplished a jump!
```scheme
(set! goto
(lambda (label)
(k (label))))
```
makes `goto` do exactly this (note that function arguments have to be evaluated before the procedure call takes place).
We use `(define goto #f)` combined with a `set!` because the labels we defined earlier need to be able to see the `goto` function.
This is what our first `with-goto` looks like when we expand it:
```scheme
(let ()
(define goto #f)
(define (loop) (display "Hello, world!\n") (goto loop))
(call/cc
(lambda (k)
(set! goto (lambda (label) (k (label))))
loop
(display "Hello, world!\n")
(goto loop))))
```
It is in fact expanded slightly differently and more efficiently, it does not use unbounded stack space <abbr title="As Far As I Know">AFAIK</abbr>, which makes sense because we arent actually increasing the depth of the callstack when we `goto`.
To demonstrate that fact and that labels are values, here is one last program.
Make an educated guess about what it does before running it, and see if you can make any general statements about its output (other than “the output never ends”).
```scheme
(with-goto go
a
(display "A")
b
(display "B")
(go (if (zero? (random 2)) a b)))
```
## Conclusion
This is useless.
There are a lot of cool things that you can implement with `call/cc`, but this is dumb!
There is a *lot* of nonsense that you can do with this implementation (try messing with nested `with-goto` or storing `goto` elsewhere).
Still, I hope you learned a bit about `call/cc` and what building abstractions with it can look like.
## Further reading
`call/cc` is awesome, but unfortunately [it sucks](https://okmij.org/ftp/continuations/against-callcc.html)!
This has been known for decades!
[Delimited continuations](https://en.wikipedia.org/wiki/Delimited_continuation) are way better!
Consider the [`⁻Ƒ⁻` operator](https://web.archive.org/web/20250112082613/https://legacy.cs.indiana.edu/~dyb/pubs/monadicDC.pdf)!
Thanks for the soapbox.
Heres a satisfying full-circle:
[Earlier](#Sitaram-link), I linked to an implementation of `amb` by Dorai Sitaram.
I recognized the name because Racket implements [the two operators](https://docs.racket-lang.org/reference/cont.html#%28form._%28%28lib._racket%2Fcontrol..rkt%29._~25%29%29)
he describes in [<i>Handling Control</i>](http://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.22.7256), a 1993 article in <i>Programming Language Design and Implementation</i>.