asynquence

asynquence

Promise-style async sequence flow control.

Explanation

asynquence ("async" + "sequence") is an abstraction on top of promises (promise chains) that lets you express flow control steps with callbacks, promises, or generators.

---

If you're interested in detailed discussion about asynquence, here's some reading to check out:

• "You Don't Know JS: Async & Performance", Appendix A and Appendix B.
• Two-part blog post series on David Walsh's site.

---

TL;DR: By Example

• Sequences & gates, at a glance
• Refactoring "callback hell"-style to asynquence
• Example/explanation of promise-style sequences
• More advanced example of "nested" composition of sequences
• Iterable sequences: sync loop and async loop and async batch iteration of list
• "State Machine" with generator coroutines
• "Ping Pong", CSP-flavored coroutine concurrency, via runner(..) contrib plugin. Also another example showing message passing.
• go-Style CSP emulating goroutines with generator wrappers (via runner(..) contrib plugin and CSP emulation adapter).
• Event-Reactive Sequences (example 1, example 2) (via react(..) plugin) inspired by RxJS's Reactive Observables
• API Usage Examples

Sequences

Say you want to perform two or more asynchronous tasks one after the other (like animation delays, XHR calls, file I/O, etc). You need to set up an ordered series of tasks and make sure the previous one finishes before the next one is processed. You need a sequence.

You create a sequence by calling ASQ(...). Each time you call ASQ(), you create a new, separate sequence.

To create a new step, simply call then(...) with a function. That function will be executed when that step is ready to be processed, and it will be passed as its first parameter the completion trigger. Subsequent parameters, if any, will be any messages passsed on from the immediately previous step.

The completion trigger that your step function(s) receive can be called directly to indicate success, or you can add the fail flag (see examples below) to indicate failure of that step. In either case, you can pass one or more messages onto the next step (or the next failure handler) by simply adding them as parameters to the call.

Example:

ASQ(21)
.then(function(done,msg){
    setTimeout(function(){
        done(msg * 2);
    },10);
})
.then(function(done,msg){
    done("Meaning of life: " + msg);
})
.then(function(done,msg){
   msg; // "Meaning of life: 42"
});

Note: then(..) can also receive other asynquence sequence instances directly, just as seq(..) can (see below). When you call then(Sq), the Sq sequence is tapped immediately, but the success/error message streams of Sq will be unaffected, meaning Sq can be continued separately.

If you register a step using then(...) on a sequence which is already currently complete, that step will be processed at the next opportunity. Otherwise, calls to then(...) will be queued up until that step is ready for processing.

You can register multiple steps, and multiple failure handlers. However, messages from a previous step (success or failure completion) will only be passed to the immediately next registered step (or the next failure handler). If you want to propagate along a message through multiple steps, you must do so yourself by making sure you re-pass the received message at each step completion.

To listen for any step failing, call or(...) (or alias onerror(..)) on your sequence to register a failure callback. You can call or() / onerror(..) as many times as you would like. If you call it on a sequence that has already been flagged as failed, the callback you specify will just be executed at the next opportunity.

ASQ(function(done){
    done.fail("Failed!");
})
// could use `or(..)` or `onerror(..)` here
.or(function(err){
    console.log(err); // Failed!
});

Gates

If you have two or more tasks to perform at the same time, but want to wait for them all to complete before moving on, you need a gate.

Calling gate(..) (or alias all(..) if you're from the Promises camp) with two or more functions creates a step that is a parallel gate across those functions, such that the single step in question isn't complete until all segments of the parallel gate are successfully complete.

For parallel gate steps, each segment of that gate will receive a copy of the message(s) passed from the previous step. Also, all messages from the segments of this gate will be passed along to the next step (or the next failure handler, in the case of a gate segment indicating a failure).

Example:

ASQ("message")
.all( // or `.gate(..)`
    function(done,msg){
        setTimeout(function(){
            done(msg + " 1");
        },200);
    },
    function(done,msg){
        setTimeout(function(){
            done(msg + " 2");
        },10);
    }
)
.val(function(msg1,msg2){
   msg1; // "message 1"
   msg2; // "message 2"
});

all(..) (or gate(..)) can also receive (instead of a function to act as a segment) just a regular asynquence sequence instance as a gate segment. When you call all(Sq), the Sq sequence is tapped immediately, but the success/error message streams of Sq will be unaffected, meaning Sq can be continued separately.

Handling Failures & Errors

Whenever a sequence goes into the error state, any error handlers on that sequence (or any sequence that it's been pipe()d to -- see Conveniences below) registered with or(..) will be fired. Even registering or(..) handlers after a sequence is already in the error state will also queue them to be fired (async, on the next event loop turn).

Errors can be programmatic failures (see above) or they can be uncaught JS errors such as ReferenceError or TypeError:

ASQ(function(done){
    foo();
})
.or(function(err){
    console.log(err); // ReferenceError: foo is not defined
});

In general, you should always register an error handler on a sequence, so as to catch any failures or errors gracefully. If there's no handlers registered when an error or failure is encountered, the default behavior of the sequence is to throw a global error (unfortunately not catchable with try..catch).

ASQ(function(done){
    foo();
});

// (global) Uncaught ReferenceError: foo is not defined

However, there will be plenty of cases where you construct a sequence and fully intend to register a handler at a later time, or wire it into another sequence (using pipe() or seq()-- see Conveniences below), and these sequences might be intended to latently hold an error without noisily reporting it until that later time.

In those cases, where you know what you're doing, you can opt-out of the globally thrown error condition just described by calling defer() on the sequence:

var failedSeq = ASQ(function(done){
    done.fail("Failed!");
})
// opt-out of global error reporting for
// this sequence!
.defer();

// later
ASQ(..)
.seq(failedSeq)
.or(function(err){
   console.log(err); // Failed!
});

Don't defer() a sequence's global error reporting unless you know what you're doing and that you'll definitely have its error stream wired into another sequence at some point. Otherwise, you'll miss errors that will be silently swallowed, and that makes everyone sad.

Conveniences

There are a few convenience methods on the API, as well:

• pipe(..) takes one or more completion triggers from other sequences, treating each one as a separate step in the sequence in question. These completion triggers will, in turn, be piped both the success and failure streams from the sequence.

  Sq.pipe(done) is sugar short-hand for Sq.then(done).or(done.fail).

• seq(..) takes one or more functions, treating each one as a separate step in the sequence in question. These functions are expected to return new sequences, from which, in turn, both the success and failure streams will be piped back to the sequence in question.

  seq(Fn) is sugar short-hand for then(function(done){ Fn.apply(null,[].slice.call(arguments,1)).pipe(done); }).

  This method will also accept asynquence sequence instances directly. seq(Sq) is (sort-of) sugar short-hand for then(function(done){ Sq.pipe(done); }). Note: the Sq sequence is tapped immediately, but the success/error message streams of Sq will be unaffected, meaning Sq can be continued separately.

  Additionally, this method can accept, either directly or through function-call, an Iterable Sequence. seq(iSq) is (sort-of) sugar short-hand for then(function(done){ iSq.then(done).or(done.fail); }).

• val(..) takes one or more functions, treating each one as a separate step in the sequence in question. These functions can each optionally return a value, each value of which will, in turn, be passed as the completion value for that sequence step.

  val(Fn) is sugar short-hand for then(function(done){ done(Fn.apply(null,[].slice.call(arguments,1))); }).

  This method will also accept non-function values as sequence value-messages. val(Va) is sugar short-hand for then(function(done){ done(Va); }).

• promise(..) takes one or more standard Promises/A+ compliant promises, and subsumes them into the sequence. See Promises/A+ Compliance below for more information.

  promise(Pr) is sugar short-hand for then(function(done){ Pr.then(done,done.fail); }).

  This method will also accept function(s) which return promises. promise(Fn) is sugar short-hand for then(function(done){ Fn.apply(null,[].slice.call(arguments,1)).then(done,done.fail); }).

• fork() creates a new sequence that forks off of the main sequence. Success or Error message(s) stream along to the forked sequence as expected, but the main sequence continues as its own sequence beyond the fork point, and neither sequence will have any further effect on the other.

  This API method is primarily useful to create multiple "listeners" at the same point of a sequence. For example: sq = ASQ()...; sq2 = sq.fork().then(..); sq3 = sq.fork().then(..); sq.then(..). In that snippet, there'd be 3 then(..) listeners that would be equally and simultaneously triggered when the main sq sequence reached that point.

  Note: Unlike most other API methods, fork() returns a new sequence instance, so chaining after fork() would not be chaining off of the main sequence but off of the forked sequence.

  sq.fork() is (sort-of) sugar short-hand for ASQ().seq(sq).

• duplicate() creates a separate copy of the current sequence (as it is at that moment). The duplicated sequence is "paused", meaning it won't automatically run, even if the original sequence is already running.

  To unpause the paused sequence-copy, call unpause() on it. The other option is to call the helper ASQ.unpause(..) and pass in a sequence. If the sequence is paused, it will be unpaused (and if not, just passes through safely).

  Note: Technically, unpause() schedules the sequence to be unpaused as the next "tick", so it doesn't really unpause immediately (synchronously). This is consistent with all other calls to the API (ASQ(), then(), gate(), etc), which all schedule procession of the sequence on the next "tick".

  The instance form of unpause(..) (not ASQ.unpause(..)) will accept any arguments sent to it and pass them along as messages to the first step of the sequence, each time it's invoked. This allows you to setup different templated (duplicated) sequences with distinct initial message states, if necessary.

  unpause() is only present on a sequence API in this initial paused state after it was duplicated from another sequence. It is removed as soon as that next "tick" actually unpauses the sequence. It is safe to call multiple times until that next "tick", though that's not recommended. The ASQ.unpause(..) helper is always present, and it first checks for an unpause() on the specified sequence instance before calling it, so that's safer.

• errfcb is a flag on the triggers that are passed into then(..) steps and gate(..) segments. If you're using methods which expect an "error-first" style (aka, "node-style") callback, {trigger}.errfcb provides a properly formatted callback for the occasion.

  If the "error-first" callback is then invoked with the first ("error") parameter set, the main sequence is flagged for error as usual. Otherwise, the main sequence proceeds as success. Messages sent to the callback are passed through to the main sequence as success/error as expected.

  ASQ(function(done){ somethingAsync(done.errfcb); }) is sugar short-hand for ASQ(function(done){ somethingAsync(function(err){ if (err) done.fail(err); else done.apply(null,[].slice.call(arguments,1))}); }).

You can also abort() a sequence at any time, which will prevent any further actions from occurring on that sequence (all callbacks will be ignored). The call to abort() can happen on the sequence API itself, or using the abort flag on a completion trigger in any step (see example below).

Static Methods

ASQ.failed(..) produces a sequence which is already in the failed state. If you pass messages along to failed(..), they will be the error messages for the sequence.

ASQ.messages(..) wraps a set of values as a ASQ-branded array, making it easier to pass multiple messages at once, and also to make it easier to distinguish a normal array (a value) from a value-messages container array, using ASQ.isMessageWrapper(..).

If you want to test if any arbitrary object is an asynquence sequence instance, use ASQ.isSequence(..).

ASQ.iterable(..) is added by the iterable-sequence contrib plugin. See Iterable Sequences below for more information.

ASQ.unpause(..) is a helper for dealing with "paused" (aka, just duplicated) sequences (see duplicate() above).

ASQ.noConflict() rolls back the global ASQ identifier and returns the current API instance to you. This can be used to keep your global namespace clean, or it can be used to have multiple simultaneous libraries (including separate versions/copies of asynquence!) in the same program without conflicts over the ASQ global identifier.

ASQ.clone() creates a fresh, clean copy of asynquence. This is primarily useful if you want to have different asynquence copies which are each extended with different plugins (see below).

Note: In node.js, if you load contrib bundle(s) from the standard top-level package location (./node_modules/asynquence-contrib/a-bundle-file.js), it will automatically look for and load (if found) the peer asynquence top-level package (./node_modules/asynquence/) and return it. So as a shortcut, you could simply do: var ASQ = require("asynquence-contrib") instead of loading both packages separately.

However, if you load contrib bundle(s) that cannot find a peer asynquence top-level package to load and use, a dependency-injection function is instead returned, which expects to be called with either an asynquence instance, or a relative path specifying where to load it.

In node, we can use the npm package freshy to let us reload the asynquence package to get a fresh copy of it, for each bundle to attach to:

var ASQ1 = require("./path/to/bundle1.js");

require("freshy").unload("asynquence");

var ASQ2 = require("./path/to/bundle2.js");

In the browser, you need to do something like this:

<script src="asq.js"></script>

<script>ASQ1 = ASQ.clone(); ASQ2 = ASQ.clone();</script>

<script>ASQ = ASQ1;</script>
<script src="./path/to/bundle1.js"></script>

<script>ASQ = ASQ2;</script>
<script src="./path/to/bundle2.js"></script>

Plugin Extensions

ASQ.extend( {name}, {build} ) allows you to specify an API extension, giving it a name and a build function callback that should return the implementation of your API extension. The build callback is provided two parameters, the sequence api instance, and an internals(..) method, which lets you get or set values of various internal properties (generally, don't use this if you can avoid it).

Example:

// "foobar" plugin, which injects message "foobar!"
// into the sequence stream
ASQ.extend("foobar",function __build__(api,internals){
    return function __foobar__() {
        api.val(function __val__(){
            return "foobar!";
        });

        return api;
    };
});

ASQ()
.foobar() // our custom plugin!
.val(function(msg){
    console.log(msg); // foobar!
});

The /contrib/ directory includes a variety of optional contrib plugins as helpers for async flow-controls. See these plugins for more complex examples of how to extend the asynquence API.

For browser usage, simply include the asq.js library file and then the contrib.js file. For node.js, these contrib plugins are available as a separate npm package: asynquence-contrib.

There are also other bundle options included with the npm package, such as contrib-es6.src.js and contrib-common.js. See Building Contrib Bundle for more information.

Iterable Sequences

One of the contrib plugins provided is iterable-sequence. Unlike other plugins, which add methods onto the sequence instance API, this plugin adds a new method directly onto the main module API: ASQ.iterable(..). Calling ASQ.iterable(..) creates a special iterable sequence, as compared to calling ASQ(..) to create a normal asynquence sequence.

An iterable sequence works similarly to normal asynquence sequences, but a bit different. then(..) still registers steps on the sequence, but it's basically just an alias of val(..), because the most important difference is that steps of an iterable sequence are not passed completion triggers.

Instead, an iterable sequence instance API has a next(..) method on it, which will allow the sequence to be externally iterated, one step at a time. Whatever is passed to next(..) is sent as step messages to the current step in the sequence. next(..) always returns an iterator result object like:

{
    value: ...          // return messages
    done: true|false    // sequence iteration complete?
}

Note: The return value will strictly either have a value property or a done property, but not both at the same time.

value is any return message(s) from the next(..) invocation (undefined otherwise). done is true if the previously iterated step was (so far) the last registered step in the iterable sequence, or false if there's still more sequence steps queued up.

Just like with normal asynquence sequences, register one or more error listeners on the iterable sequence by calling or(..). If a step results in some error (either accidentally or manually via throw ..), the iterable sequence is flagged in the error state, and any error messages are passed to the registered or(..) listeners.

Also, just like next(..) externally controls the normal iteration flow of the sequence, throw(..) externally "throws" an error into the iterable sequence, triggering the or(..) flow as above. Iterable sequences can be abort()d just as normal asynquence sequences. You can also call return(..) (just like on normal iterators), which abort()s the sequence and returns an iterator result with the value passed in, if any, and done: true.

Iterable sequences are a special subset of sequences, and as such, some of the normal asynquence API variations do not exist, such as gate(..), seq(..), and promise(..).

function step(num) {
    return "Step " + num;
}

var sq = ASQ.iterable()
    .then(step)
    .then(step)
    .then(step);

for (var i=0, ret;
    (ret = sq.next(i+1)) && !ret.done;
    i++
) {
    console.log(ret.value);
}
// Step 1
// Step 2
// Step 3

This example shows sync iteration with a for loop, but of course, next(..) can be called in various async fashions to iterate the sequence over time.

Just like regular sequences, iterable sequences have a duplicate() method (see ASQ's instance API above) which makes a copy of the sequence at that moment. However, iterable sequences are already "paused" at each step anyway, so unlike regular sequences, there's no unpause() (nor is there any reason to use the ASQ.unpause(..) helper!), because it's unnecessary. You just call next() on an iterable sequence (even if it's a copy of another) when you want to advance it one step.

Multiple parameters

API methods take one or more functions as their parameters. gate(..) treats multiple functions as segments in the same gate. The other API methods (then(..), or(..), pipe(..), seq(..), and val(..)) treat multiple parameters as just separate subsequent steps in the respective sequence. These methods don't accept arrays of functions (that you might build up programatically), but since they take multiple parameters, you can use .apply(..) to spread those out.

Promises/A+ Compliance

The goal of asynquence is that you should be able to use it as your primary async flow-control library, without the need for other Promises implementations.

---

If you're looking for actual Promises/A+ compliance, I've just released Native Promise Only, a tiny and fast polyfill of purely just the native ES6 Promise() mechanism.

---

asynquence is intentionally designed to hide/abstract the idea and use of Promises, such that you can do quick and easy async flow-control programming without some of the hassles/tedium of creating Promises directly.

As such, the asynquence API itself is not Promises/A+ compliant, nor should it be, because the "promises" used are hidden underneath asynquence's API. Note: the hidden promises behave predictably like standard Promises where they need to, so asynquence as an abstraction offers the same trust guarantees.

If you are also using other Promises implementations alongside asynquence, you can quite easily receive and consume a regular Promise value (or thenable) from some other method into the signal/control flow for an asynquence sequence.

For example, if using jQuery, the Q promises library, and asynquence:

// Using *Q*, make a standard Promise out
// of jQuery's Ajax (non-standard) "promise"
var p = Q( $.ajax(..) );

// Now, asynquence flow-control including a
// standard Promise
ASQ()
.then(function(done){
    setTimeout(done,100);
})
// subsume a standard Promise into the sequence
.promise(p)
.val(function(ajaxResp){
    console.log(ajaxResp);
});

Despite API similarities (like the presence of then(..) on the API), an asynquence instance is not itself designed to be used as a Promise value linked/passed to another standard Promise or other utilities that expect real promises.

Trying to do so will likely cause unexpected behavior, because Promises/A+ insists on problematic (read: "dangerous") duck-typing for objects that have a then() method, as asynquence instances do.

However, if you really need a standard native Promise from your sequence, you can use the toPromise contrib plugin, which vends/forks an actual native Promise off an asynquence sequence instance.

Browser, node.js (CommonJS), AMD: ready!

The asynquence library is packaged with a light variation of the UMD (universal module definition) pattern, which means the same file is suitable for inclusion either as a normal browser .js file, as a node.js module, or as an AMD module. Can't get any simpler than that, can it?

For browser usage, simply include the asq.js library file. For node.js usage, install the asynquence package via npm, then require(..) the module:

var ASQ = require("asynquence");

Note: The ASQ.noConflict() method really only makes sense when used in a normal browser global namespace environment. It should not be used when the node.js or AMD style modules are your method of inclusion.

Usage Examples

Using the following example setup:

function fn1(done) {
    alert("Step 1");
    setTimeout(done,1000);
}

function fn2(done) {
    alert("Step 2");
    setTimeout(done,1000);
}

function yay() {
    alert("Done!");
}

Execute fn1, then fn2, then finally yay:

ASQ(fn1)
.then(fn2)
.then(yay);

Pass messages from step to step:

ASQ(function(done){
    setTimeout(function(){
        done("hello");
    },1000);
})
.then(function(done,msg1){
    setTimeout(function(){
        done(msg1,"world");
    },1000);
})
.then(function(_,msg1,msg2){ // basically ignoring this step's completion trigger (`_`)
    alert("Greeting: " + msg1 + " " + msg2);
    // 'Greeting: hello world'
});

Handle step failure:

ASQ(function(done){
    setTimeout(function(){
        done("hello");
    },1000);
})
.then(function(done,msg1){
    setTimeout(function(){
        // note the `fail` flag here!!
        done.fail(msg1,"world");
    },1000);
})
.then(function(){
    // sequence fails, won't ever get called
})
.or(function(msg1,msg2){
    alert("Failure: " + msg1 + " " + msg2);
    // 'Failure: hello world'
});

Create a step that's a parallel gate:

ASQ()
// normal async step
.then(function(done){
    setTimeout(function(){
        done("hello");
    },1000);
})
// parallel gate step (segments run in parallel)
.gate(
    function(done,greeting){ // gate segment
        setTimeout(function(){
            // 2 gate messages!
            done(greeting,"world");
        },500);
    },
    function(done,greeting){ // gate segment
        setTimeout(function(){
            // only 1 gate message!
            done(greeting + " mikey");
        },100);
        // this segment finishes first, but message
        // still kept "in order"
    }
)
.then(function(_,msg1,msg2){
    // msg1 is an array of the 2 gate messages
    // from the first segment
    // msg2 is the single message (not an array)
    // from the second segment

    alert("Greeting: " + msg1[0] + " " + msg1[1]);
    // 'Greeting: hello world'
    alert("Greeting: " + msg2);
    // 'Greeting: hello mikey'
});

Use pipe(..), seq(..), and val(..) helpers:

var seq = ASQ()
.then(function(done){
    ASQ()
    .then(function(done){
        setTimeout(function(){
            done("Hello World");
        },100);
    })
    .pipe(done); // pipe sequence output to `done` completion trigger
})
.val(function(msg){ // NOTE: no completion trigger passed in!
    return msg.toUpperCase(); // map return value as step output
})
.seq(function(msg){ // NOTE: no completion trigger passed in!
    var seq = ASQ();

    seq
    .then(function(done){
        setTimeout(function(){
            done(msg.split(" ")[0]);
        },100);
    });

    return seq; // pipe this sub-sequence back into the main sequence
})
.then(function(_,msg){
    alert(msg); // "HELLO"
});

Abort a sequence in progress:

var seq = ASQ()
.then(fn1)
.then(fn2)
.then(yay);

setTimeout(function(){
    // will stop the sequence before running
    // steps `fn2` and `yay`
    seq.abort();
},100);

// same as above
ASQ()
.then(fn1)
.then(function(done){
    setTimeout(function(){
        // `abort` flag will stop the sequence
        // before running steps `fn2` and `yay`
        done.abort();
    },100);
})
.then(fn2)
.then(yay);

Builds

The core library file can be built (minified) with an included utility:

./build-core.js

However, the recommended way to invoke this utility is via npm:

npm run-script build-core

License

The code and all the documentation are released under the MIT license.

http://getify.mit-license.org/