Last week, Rich Hickey announced Clojure core.async in a blog post. As mentioned in the post, the new core.async library has a lot in common with Go. In this post, I’ll compare the fundamental building blocks of concurrency in core.async and Go with code examples.

Note: Clojure core.async provides two sets of operations on channels. The blocking operations are for use with native threads and the non-blocking operations are for use with go blocks. In this post, I’ll be focusing on the non-blocking operations used with go blocks but I’ll briefly mention the blocking versions.

Update: It is important to note that I’m using Thread/sleep in the clojure examples for clarity. This will block the entire thread and eventually starve the thread pool used for go blocks. Don’t use it in real code, use a timeout instead (thanks MBlume and pron).

Setup

To install Go on OSX, just use homebrew.

$ brew install go

For clojure, you’ll want to install leiningen via homebrew.

$ brew install leiningen

After generating a leiningen project, you’ll need to add core.async as a dependency. Unfortunately it’s not yet available on maven central, so you’ll need to clone it and install it in your local maven repository first.

$ git clone [email protected]:clojure/core.async.git
$ cd core.async
$ maven install

Now, we can add core.async as a dependency in our project.clj file.

(defproject async_example "0.1.0-SNAPSHOT"
  :description "Async example"
  :url "http://example.com"
  :license {:name "MIT License" :url "http://opensource.org/licenses/MIT"}
  :main async-example.core
  :dependencies [[org.clojure/clojure "1.5.1"]
                 [org.clojure/core.async "0.1.0-SNAPSHOT"]])

Update: To avoid having to install core.async locally, you can add the following line to your project.clj (thanks weavejester):

:repositories {"sonatype-oss-public"
"https://oss.sonatype.org/content/groups/public/"}

We’re all set to start comparing Go and core.async.

Goroutines and Go Blocks

Both core.async and Go provide a facility for spawning “lightweight threads”. In core.async, this is handled via go blocks. In Go, we use goroutines.

Let’s write an example spawning 10 lightweight threads that will sleep for a random amount of time and then print a number (0-9).

package main

import (
  "fmt"
  "math/rand"
  "time"
)

func main() {
  for i := 0; i < 10; i++ {
    go func(i int) {
      sleep := time.Duration(rand.Intn(1000))
      time.Sleep(sleep * time.Millisecond)
      fmt.Println(i)
    }(i)
  }

  time.Sleep(2000 * time.Millisecond)
}

As you can see, we use the go keyword to spawn goroutines and each waits a bit and prints its designated number.

(ns async-example.core
  (:require [clojure.core.async :refer :all])
  (:gen-class))

(defn -main [& args]
  (doseq [i (range 10)]
    (go
      (Thread/sleep (rand-int 1000))
      (println i)))

  (Thread/sleep 2000))

The clojure code looks quite similar (besides being a lisp) to the Go code. The main difference is we use the (go ...) macro to spawn a go block.

Channels

While goroutines and go blocks are slightly interesting in isolation, they become much more powerful when combined with channels. Channels can be thought of as blocking queues that goroutines or go blocks can push messages onto and pull messages off of. In Go, we use ch <- and <-ch to push and pull from a channel respectively. In clojure, we use >! and <!.

To construct channels in Go we use make(chan <type>), in clojure we use (chan).

It is important to remember that, by default, when a value is pushed onto a channel it blocks until it is pulled off. Likewise, when a value is pulled from a channel it blocks until there is something to pull.

Below is an example of 10 goroutines/go blocks pushing values onto a channel and a main goroutine/go block pulling values off the channel and printing them.

package main

import (
  "fmt"
  "math/rand"
  "time"
)

func main() {
  c := make(chan int)

  for i := 0; i < 10; i++ {
    go func(i int) {
      sleep := time.Duration(rand.Intn(1000))
      time.Sleep(sleep * time.Millisecond)
      c <- i
    }(i)
  }

  for i := 0; i < 10; i++ {
    fmt.Println(<-c)
  }
}

(ns async-example.core
  (:require [clojure.core.async :refer :all])
  (:gen-class))

(defn -main [& args]
  (let [c (chan)]
    (doseq [i (range 10)]
      (go
        (Thread/sleep (rand-int 1000))
        (>! c i)))

    (<!!
      (go
        (doseq [_ (range 10)]
          (println (<! c)))))))

There are a few differences here to point out. First, you’ll notice that we didn’t spawn a goroutine for the main loop that reads the values in the go example. This is because the main program itself is running in a goroutine. In clojure, because core.async is a library, we must put the pulling component in a go block as well.

Second, you’ll notice that the last go block in the clojure example is surrounded by (<!! ...). This is an equivalent function to <! except that it is used with native threads instead of go blocks. In core.async, go blocks return a channel that have the last value of the go block pushed onto it when execution is complete. By wrapping the final go block in a call to <!!, we block the main thread of the program until all the pulling is complete.

Select and Alts!

The last piece of the puzzle is the ability to pull a value off many channels. Go provides select and core.async provides alts!. Each will take a collection of channels and execute some code based on the first channel with activity.

We can use select or alts! to add timeouts to our actions. Suppose we have a goroutine/go block that will put a value onto a channel sometime between now and a second from now, but we want to stop the operation if it takes longer than half a second. The following code would accomplish this task.

package main

import (
  "fmt"
  "math/rand"
  "time"
)

func main() {
  rand.Seed(time.Now().UTC().UnixNano())

  c := make(chan string)

  go func() {
    sleep := time.Duration(rand.Intn(1000))
    time.Sleep(sleep * time.Millisecond)
    c <- "success!"
  }()

  select {
  case <-c:
    fmt.Println("Got a value!")
  case <-time.After(500 * time.Millisecond):
    fmt.Println("Timeout!")
  }
}

It’s important to understand that the function time.After returns a channel onto which a value will be pushed after the specified timeout. Note that I’m seeding the rand package so that we get different results every time the program is run.

(ns async-example.core
  (:require [clojure.core.async :refer :all])
  (:gen-class))

(defn -main [& args]
  (let [c (chan)]
    (go
      (Thread/sleep (rand-int 1000))
      (>! c "success!"))

    (<!!
      (go
        (let [[result source] (alts! [c (timeout 500)])]
          (if (= source c)
            (println "Got a value!")
            (println "Timeout!")))))))

Similar to time.After, the timeout function returns a channel that will have a value pushed onto it after the timeout. The call to alts! returns a vector of the value from the channel and the channel that returned the value (called source in the above example).

Wrap Up

After spending a few days with clojure’s core.async, I’m very excited about the possibilities. Previously, I was using Go because I enjoyed its approach to concurrency. Now, this same functionality has been added to clojure via a library. To me, this is a huge win. It means I can program using the concurrency style from Go without fighting its type system and verbosity. To make things even better, you retain all the benefits of lisp and the java ecosystem.

You can learn more about core.async from the excellent code walkthrough.

Setup#

Goroutines and Go Blocks#

Channels#

Select and Alts!#

Wrap Up#

Setup

Goroutines and Go Blocks

Channels

Select and Alts!

Wrap Up