Breaking Racket

If you've looked at some of the concurrency and networking procedures available in Racket, you might've noticed a bunch that follow the naming pattern <name>/enable-break and it might not have been immediately obvious why or when you would use these variants of these procedures over the regular ones. Let's look at a snippet from the documentation of tcp-accept/enable-break:

If breaking is disabled when tcp-accept/enable-break is called, then either ports are returned or the exn:break exception is raised, but not both.

So, the procedure guarantees that either it will accept a connection and return a pair of ports, or it will raise a break exception, if breaking is disabled when it is called. Sounds straightforward enough, but it still might not be clear where this would come in handy. The part of the quote about "if breaking is disabled" is an important hint. Consider this simplified piece of code based on my previous post re. the protohackers challenge:

(define listener (tcp-listen 8000))
(with-handlers ([exn:break? void])
  (let loop ()
    (parameterize-break #f
      (define-values (in out)
        (tcp-accept listener))                     ;; •1
      (define conn-thd
        (thread (make-connection-handler in out))) ;; •2
      (thread (make-supervisor conn-thd in out)))  ;; •3
    (loop)))
(tcp-close listener)

We set up a listener, then enter a loop to start accepting new connections and spawn threads to handle every connection. For every connection-handling thread, we spawn a supervising thread to clean up resources once the handling thread is done.

We want to ensure that a handling thread gets spawned for every accepted connection, so we want to guarantee that no breaks can be received between •1 and •2. Likewise, we want to guarantee that every handler thread has an associated supervision thread, so no breaks should be allowed between •2 and •3 either. So, we've wrapped •1, •2, and •3 in the parameterize-break form to disable breaks. However, accepting a new connection will block the thread until a client actually tries connecting to the server. This means that we can only stop this server if we happen to send it a break at exactly the time after •3 returns, or by killing the process.

By replacing the call to tcp-accept with tcp-accept/enable-break, we can preserve the same guarantees as before while also allowing breaks to be raised before new connections are accepted.