slide

Controlling Flow: callbacks are easy

What's actually hard?

• Doing a bunch of things in a specific order.
• Knowing when stuff is done.
• Handling failures.
• Breaking up functionality into parts (avoid nested inline callbacks)

Common Mistakes

• Abandoning convention and consistency.
• Putting all callbacks inline.
• Using libraries without grokking them.
• Trying to make async code look sync.

Define Conventions

• Two kinds of functions: actors take action, callbacks get results.
• Essentially the continuation pattern. Resulting code looks similar to fibers, but is much simpler to implement.
• Node works this way in the lowlevel APIs already, and it's very flexible.

Callbacks

• Simple responders
• Must always be prepared to handle errors, that's why it's the first argument.
• Often inline anonymous, but not always.
• Can trap and call other callbacks with modified data, or pass errors upwards.

Actors

• Last argument is a callback.
• If any error occurs, and can't be handled, pass it to the callback and return.
• Must not throw. Return value ignored.
• return x ==> return cb(null, x)
• throw er ==> return cb(er)

// return true if a path is either
// a symlink or a directory.
function isLinkOrDir (path, cb) {
  fs.lstat(path, function (er, s) {
    if (er) return cb(er)
    return cb(null, s.isDirectory() || s.isSymbolicLink())
  })
}

asyncMap

Usecases

• I have a list of 10 files, and need to read all of them, and then continue when they're all done.
• I have a dozen URLs, and need to fetch them all, and then continue when they're all done.
• I have 4 connected users, and need to send a message to all of them, and then continue when that's done.
• I have a list of n things, and I need to dosomething with all of them, in parallel, and get the results once they're all complete.

Solution

var asyncMap = require("slide").asyncMap
function writeFiles (files, what, cb) {
  asyncMap(files, function (f, cb) {
    fs.writeFile(f, what, cb)
  }, cb)
}
writeFiles([my, file, list], "foo", cb)

chain

Usecases

• I have to do a bunch of things, in order. Get db credentials out of a file, read the data from the db, write that data to another file.
• If anything fails, do not continue.
• I still have to provide an array of functions, which is a lot of boilerplate, and a pita if your functions take args like

function (cb) {
  blah(a, b, c, cb)
}

• Results are discarded, which is a bit lame.
• No way to branch.

Solution

• reduces boilerplate by converting an array of [fn, args] to an actor that takes no arguments (except cb)
• A bit like Function#bind, but tailored for our use-case.
• bindActor(obj, "method", a, b, c)
• bindActor(fn, a, b, c)
• bindActor(obj, fn, a, b, c)
• branching, skipping over falsey arguments

chain([
  doThing && [thing, a, b, c]
, isFoo && [doFoo, "foo"]
, subChain && [chain, [one, two]]
], cb)

• tracking results: results are stored in an optional array passed as argument, last result is always in results[results.length - 1].
• treat chain.first and chain.last as placeholders for the first/last result up until that point.

Non-trivial example

• Read number files in a directory
• Add the results together
• Ping a web service with the result
• Write the response to a file
• Delete the number files

var chain = require("slide").chain
function myProgram (cb) {
  var res = [], last = chain.last, first = chain.first
  chain([
    [fs, "readdir", "the-directory"]
  , [readFiles, "the-directory", last]
  , [sum, last]
  , [ping, "POST", "example.com", 80, "/foo", last]
  , [fs, "writeFile", "result.txt", last]
  , [rmFiles, "./the-directory", first]
  ], res, cb)
}

Conclusion: Convention Profits

• Consistent API from top to bottom.
• Sneak in at any point to inject functionality. Testable, reusable, ...
• When ruby and python users whine, you can smile condescendingly.