Beyond the Void
BYVoid
Continuation.js - A solution to asynchronous JavaScript programming

Continuation.js is a compiler for Continuation-Passing Style transformation, which simplifies asynchronous JavaScript programming. It translates slightly flavored JavaScript syntax into standard JavaScript, so it can be also called a “translator”. Continuation.js introduces a virtual function cont, which allow you to write continuation-passing style code (or asynchronous callback style code) far easier. cont is not a actual function, but a mark with the same syntax to function calls in JavaScript. By using Continuation.js you can write asynchronous control flows like flat threaded code, and it compiles it into continuation-passing style code.

Slides of Continuation.js

Continuation.js

Overview

Typically, writing with asynchronous control flow is a pain, because you will easily write nested callbacks like below:

function textProcessing(callback) {
  fs.readFile('somefile.txt', 'utf-8', function (err, contents) {
    if (err) return callback(err);
    //process contents
    contents = contents.toUpperCase();
    fs.readFile('somefile2.txt', 'utf-8', function (err, contents2) {
      if (err) return callback(err);
      contents += contents2;
      fs.writeFile('somefile_concat_uppercase.txt', contents, function (err) {
        if (err) return callback(err);
        callback(null, contents);
      });
    });
  });
}
textProcessing(function (err, contents) {
  if (err)
    console.error(err);
});

This kind of coding style is called “callback hells” or “callback pyramids”. While using Continuation.js, you directly write:

function textProcessing(ret) {
  fs.readFile('somefile.txt', 'utf-8', cont(err, contents));
  if (err) return ret(err);
  contents = contents.toUpperCase();
  fs.readFile('somefile2.txt', 'utf-8', cont(err, contents2));
  if (err) return ret(err);
  contents += contents2;
  fs.writeFile('somefile_concat_uppercase.txt', contents, cont(err));
  if (err) return ret(err);
  ret(null, contents);
}
textProcessing(cont(err, contents));
if (err)
  console.error(err);

The code above is flatted by using the virtual function cont. Control flow must “wait” for the return of asynchronous function call fs.readFile. Parameters in the argument list of cont will be set after it returns. “Return” here is a little confusing because in an asynchronous function “return” means callback function called, not “return” in the literal sense. An asynchronous function usually returns immediately (by encountering return statement or the end of the function scope) while the callback function could be called later. You can be understood as all the statements after cont until the end of the function are the callback function of the asynchronous function call. The code feels like threaded code, but it is still asynchronous while executing.

Even more simply, the asynchronous error can be held with JavaScript try...catch syntax and another virtual function call obtain provided by Continuation.js. obtain(a) is equivalent to cont(err, a), and err would be thrown if it is not undefined.

function textProcessing(ret) {
  try {
    fs.readFile('somefile.txt', 'utf-8', obtain(contents));
    contents = contents.toUpperCase();
    fs.readFile('somefile2.txt', 'utf-8', obtain(contents2));
    contents += contents2;
    fs.writeFile('somefile_concat_uppercase.txt', contents, obtain());
    ret(null, contents);
  } catch(err) {
    ret(err);
  }
}
try {
  textProcessing(obtain(contents));
} catch(err) {
  console.error(err);
}

Features

  • A JIT (Just-in-time compilation) and AOT (Ahead-of-time) compiler
  • Strictly no runtime dependence
  • No additional non-native JavaScript syntax and very few reserved keywords (only cont, obtain and parallel)
  • Flexible coding style and readable compiled code
  • Compatible with CoffeeScript and LiveScript (and other compile-to-js language)
  • Supports both Node.js and browser-side JavaScript
  • Supports execution in parallel and lightweight thread

Documentation

cont

cont is a mark of asynchronous calls. It must be used in the place of callback function parameter of a function call. The parameters in cont will be set to the values of arguments of the callback function. If parameter of cont is a variable (not an expression), it will be defined automatically before the function call.

Example:

setTimeout(cont(), 1000);

fs.lstat('/path/file', cont(err, stats));

var obj;
fs.readdir('/path', cont(err, obj.files));

Compiled code:

var err, stats, obj;
setTimeout(function () {
  fs.lstat('/path', function () {
    err = arguments[0];
    stats = arguments[1];
    fs.readdir('/path', function () {
      err = arguments[0];
      obj.files = arguments[1];
    });
  });
}, 1000);

obtain

obtain is a syntax sugar of cont and throw. You can use try to catch asynchronous errors with obtain. One assumption is that the error object is the first parameter of the callback function (this is a convention of Node.js API).

Example:

function f1() {
  fs.readdir('/path', obtain(files));
}

function f2() {
  fs.readdir('/path', cont(err, files));
  if (err)
    throw err;
}

Compiled code:

function f1() {
  var _$err, files;
  fs.readdir('/path', function () {
    _$err = arguments[0];
    files = arguments[1];
    if (_$err)
      throw _$err;
  });
}
function f2() {
  var err, files;
  fs.readdir('/path', function () {
    err = arguments[0];
    files = arguments[1];
    if (err) {
      throw err;
    }
  });
}

parallel

parallel allows you to execute asynchronous functions “in parallel”. parallel is also a virtual function and its parameters should be all function calls with cont or obtain. All function calls in parallel will be executed in parallel and control flow continues executing the next statement after all the parallel function calls “return” (precisely speaking callback functions called).

Note that both Node.js and browser execute JavaScript code in single thread, so this is not a multithreaded implementation. The function calls feel like entrances of threads so they can be called “lightweight threads”. Only I/O and computing are executed in parallel while computing are actually executed in sequence. That is, when one “thread” is “blocked” by I/O, the other one runs. The “threads” can automatically utilize the gaps of I/O and computing. So you should let functions that both have I/O and computing run in parallel, rather than all functions with heavy computing and little I/O.

Example:

var contents = {};
parallel(
  fs.readFile('/path1', obtain(contents.file1)),
  fs.readFile('/path2', obtain(contents.file2)),
  fs.readFile('/path3', obtain(contents.file3))
);
console.log(contents);

Explicit mode

Modules can be written in Continuation.js and they will be recursively compiled automatically when using require. Add 'use continuation' into your source file, and use continuation script.js --explicit to run it and only files contains 'use continuation' will be compiled. This option can reduce loading time if many modules are recursively required.

Compilation cache

By using contination script.js --cache [cacheDir], all modules compiled by Contination.js will be cached into cacheDir. If the timestamp of your source file is newer than the cached version, it will be compiled again. This option rely on the system clock and it can also reduce loading time. It is recommended to use --explicit and --cache together.

By default cacheDir is /tmp/continuation.

Use Continuation.js in CoffeeScript (and other compile-to-js language)

Continuation.js is compatible with most kinds of compile-to-js language because it introduces no non-primitive syntax. The only 3 keywords cont, obtain and parallel are all virtual functions so you can use function call directly in your language.

Example (CoffeeScript):

dns = require('dns')
domains = ['www.google.com', 'nodejs.org', 'byvoid.com']
for domain in domains
  dns.resolve domain, obtain(addresses)
  console.log addresses

Until now CoffeeScript and LiveScript are supported by default.

Use Continuation.js programmatically

You are able to use Continuation.js programmatically. Continuation.js module has one interface function compile(code). code must be a string.

Example:

var continuation = require('continuation');

function fibonacci() {
  var a = 0, current = 1;
  while (true) {
    var b = a;
    a = current;
    current = a + b;
    setTimeout(cont(), 1000);
    console.log(current);
  }
};

var code = fibonacci.toString();
var compiledCode = continuation.compile(code);
console.log(compiledCode);
eval(compiledCode);
fibonacci();

You can run the code above using node command. That means you don’t have to install Continuation.js in global environment. The code above converts a function into a string, and then it is compiled by Continuation.js. After that you can run it via eval function.

Installation

Install Continuation.js with npm:

npm install -g continuation

Usage

Usage: continuation [options] <file.js/file.coffee/file.ls> [arguments]

Options:

  -h, --help               output usage information
  -V, --version            output the version number
  -p, --print              compile script file and print it
  -o, --output <filename>  compile script file and save as <filename>
  -e, --explicit           compile only if "use continuation" is explicitly declared
  -c, --cache [directory]  run and cache compiled sources to [directory], by default [directory] is /tmp/continuation
  -v, --verbose            print verbosal information to stderr

Run code written with Continuation.js (i.e. script.js):

contination script.js

Print compiled code on console:

contination script.js -p

Compile code and save as another file:

contination script.js -o compiled.js

Run code in “explicit mode” (only compile sources with 'use continuation'. see documentation):

contination script.js -e

Cache compiled code (including recursively required modules) to accelerate loading (see documentation):

contination script.js -c

Examples

Loops and sleep

Calculating Fibonacci sequence and printing one number by every one second:

var fib = function () {
  var a = 0, current = 1;
  while (true) {
    var b = a;
    a = current;
    current = a + b;
    setTimeout(cont(), 1000);
    console.log(current);
  }
};
fib();

Run asynchronous functions in sequence

Read 5 files in sequence:

var fs = require('fs');

for (var i = 0; i < 4; i++) {
  fs.readFile('text' + i + '.js', 'utf-8', obtain(text));
  console.log(text);
}

console.log('Done');

Run asynchronous functions in parallel

Do things (with heave I/O) in parallel:

var fs = require('fs');
var dns = require('dns');
var http = require('http');

var complexWork = function (next) {
  setTimeout(cont(), 500);
  http.get('https://byvoid.com', cont(res));
  next(null, res.headers);
};

parallel(
  fs.readdir('/', obtain(files)),
  dns.resolve('npmjs.org', obtain(addresses)),
  complexWork(obtain(result))
);

console.log(files, addresses, result);

Recursions

Calculating disk usage:

var fs = require('fs');

function calcDirSize(path, callback) {
  var dirSize = 0, dirBlockSize = 0;
  fs.readdir(path, obtain(files));
  for (var i = 0; i < files.length; i++) {
    var filename = path + '/' + files[i];
    fs.lstat(filename, obtain(stats));
    if (stats.isDirectory()) {
      calcDirSize(filename, obtain(subDirSize, subDirBlockSize));
      dirSize += subDirSize;
      dirBlockSize += subDirBlockSize;
    } else {
      dirSize += stats.size;
      dirBlockSize += 512 * stats.blocks;
    }
  }
  callback(null, dirSize, dirBlockSize);
}

var path = process.argv[2];
if (!path) path = '.';

calcDirSize(path, obtain(totalSize, totalBlockSize));

console.log('Size:', Math.round(totalSize / 1024), 'KB');
console.log('Actual Size on Disk:', Math.round(totalBlockSize / 1024), 'KB');

More examples are available in ’examples’ directory and ’test’ directory.

Relevant projects

Continuation.js is not the only solution of CPS transformation, there are some alternatives. Below is a comparison among projects related to synchronous to asynchronous transformation.

Project Continuation.js streamline.js TameJS Wind.js jwacs NarrativeJS StratifiedJS
Node.js support Yes Yes Yes Yes No No Yes
Browser side support Yes Yes No Yes Yes Yes Yes
Additional runtime dependence No Yes Yes Yes No Yes Yes
Additional syntax No No Yes No Yes Yes Yes
Compatibility with compile-to-js language Yes Yes (CoffeeScript) Yes (CoffeeScript) Yes (manually) No No No
Parallel support Yes Yes Yes Yes Yes Unknown Yes
Readable generated code Yes No Hardly Almost Unknown No Unknown
Asynchronous results passing syntax Parameters Return value Parameters Return value Return value Return value Return value
Documentation Yes Yes Yes No Yes Yes Yes
Implemented in JavaScript JavaScript JavaScript JavaScript Lisp Java JavaScript

Last modified on 2017-05-22