Home

Awesome

recalc

"[T]hus, the external world would only have a triggering role in the release of the internally-determined activity of the nervous system." --Humberto Maturana

"Should that function take the whole world as input and return a brand new world as output? Why even use functional programming, then?" --James Hague

recalc is a library for reasoning functionally about side effects. Its core ideas are: use events to pass control and a global store that is updated through events to pass data.

Current status of the project

The current version of recalc, v5.1.2, is considered to be stable and complete. Suggestions and patches are welcome. Besides bug fixes, there are no future changes planned.

recalc is part of the ustack, a set of libraries to build web applications which aims to be fully understandable by those who use it.

Installation

The dependencies of recalc are two:

recalc is written in Javascript. You can use it in the browser by sourcing the dependencies and the main file:

<script src="dale.js"></script>
<script src="teishi.js"></script>
<script src="recalc.js"></script>

Or you can use these links to the latest version - courtesy of jsDelivr.

<script src="https://cdn.jsdelivr.net/gh/fpereiro/dale@3199cebc19ec639abf242fd8788481b65c7dc3a3/dale.js"></script>
<script src="https://cdn.jsdelivr.net/gh/fpereiro/teishi@31a9cf552dbaee79fb1c2b7d12c6fad20f987983/teishi.js"></script>
<script src="https://cdn.jsdelivr.net/gh/fpereiro/recalc@38b1fc1c5397b27d9162147fddad103a43e9b826/recalc.js"></script>

And you also can use it in node.js. To install: npm install recalc

recalc should work in any version of node.js (tested in v0.8.0 and above). Browser compatibility has been tested in the following browsers:

The author wishes to thank Browserstack for providing tools to test cross-browser compatibility.

Philosophy

Recalc is a library for working with side effects in a functional way. Functional programming traditionally tries to marginalize side effects as much as possible, but I feel this is dishonest, because the side effects are still there and still needed. And as any form of denial, it ends up generating more trouble.

Programming in a functional style entails making programs out of functions. These functions call each other, in predetermined sequences, passing data between them exclusively in the form of arguments (inputs) and return values (outputs). Control and data are passed simultaneously: when a function returns, it both returns a value and also control to the function which called it in the first place. When all functions return, the program terminates.

These limitations (data & control being passed simultaneously, data being passed only as arguments and return values, and arguments not being modified by the function which receives them) are very empowering, because they make it much easier to reason about data and control flow, while still allowing for powerful programs.

Before I continue, let me say that I consider functional programming to be the most solid way of programming, for both small and large applications. Not only I consider it much better than imperative programming - I also think it is superior to object oriented programming. Functional is the paradigm I choose every time for all my projects.

However, there are certain circumstances where a pure functional approach won't work. Interestingly, virtually all the objections to functional programming which I consider valid emerge in these circumstances. It is also in these circumstances that many employ imperative or object oriented techniques.

In a decently pure functional program, there's only one side effect: when someone actually starts the program passing it some input. This both provides control and data to the main function of the program, which then calls other functions. When the whole thing is done, the main function dutifully returns its result and exits. In most cases, though, it is allowed for that function to commit another side effect to do something useful with its output, like writing that data to a file. So, besides the initialization and occasionally the exit, the rest of the program is purely functional.

The shape of this program, however, is pretty much identical to a traditional batch-and-queue program. The only interactions (or side effects, since they amount to the same thing) with the outside world are located only at the beginning and at the end of the process.

But an interactive program, like a game or an application, has to process interactions from the user and the server while it is running. The question is, how do we handle these interactions/side effects in a functional manner?

This library is an attempt to answer that question.

If we can't get rid of side effects, we can instead embrace them without forgetting our functional roots. We can consider that each side effect is an event. When a side effect comes in, we make our program call an event. And here's the kicker: there might be one or more responders for a given event. Those responders are merely functions that are executed as a consequence of a given event being called. In this way, you can work on the problem of passing control around when side effects come into the picture.

The other question is: how do we pass data around? Since there might be multiple responders to an event, operating sequentially or parallely, it would be messy to depend on return values. Instead, we can create a central object where we can store all the state of the application. A global object shared by all functions (I hope you're cringing now), which we call the store.

But wait! If these events and their handlers modify this global object, wouldn't be this merely be imperative programming? It would, except for one catch: data changes on the global state can also be expressed as events, to which functions can respond/react. So you don't just increment a variable. You increment it through an event that notifies everyone concerned. Events allow us to create first-class side effects, so that we can define them, pass them around and reason about them in a clear way.

From a functional perspective, this is tantamount to heresy. We're talking here about systematizing side effects, using a global state to pass data and ignoring the return values from event handlers. However, this is exactly how I think that functional programming should be extended to deal with programs which continually interact with the outside world. In other words, this is an attempt to find the general case of functional programming.

Programs as communication

The deep pattern behind recalc (and event systems in general) is communication. More precisely, the call and response pattern, which takes us to the origins of information.

By expressing computation as communication, some parts of the program perform calls and other parts of the program respond to those calls.

The two nouns with which we can structure this paradigm is: event and responder. The two corresponding verbs are call and match: events are called (almost always by responders), responders are matched by events. This constructive circularity can represent programs in an understandable and scalable way.

Usage

To use recalc, you first need to create a recalc object.

Let's assume that the recalc library has been loaded at the variable R (this should happen automatically in the browser - in node.js, you would need to write var R = require ('recalc')). To initialize a recalc object you would do this:

var r = R ();

r will be now our recalc object. The reason we use a constructor (instead of directly making R a recalc object) is that you may need to have multiple recalc objects at the same time; by executing the constructor more than once, you can get as many recalc objects as needed.

Every recalc object has six elements that you need to understand in order to use the library. The first three are objects, the other three are functions:

r.responders

The responders object is an object that contains all the responders. Whenever an event is called, recalc iterates all the elements within responders, and executes those that match the event being called.

To add and remove responders, rather than directly modifying the array, I recommend you use r.respond and r.forget (we'll see these functions in a moment).

r.store

Our beloved global state. It is initailized as an empty object. If you want to initialize it to an array, you can do so by initializing r like this:

var r = R ([]);

In this way, you can also initialize store to contain some data, right off the bat:

var r = R ({some: 'important', data: true});

By the way, that's the only argument you can pass to R (the constructor).

An important thing to notice is that the responders are stored outside of the global store.

r.log

An array that contains all the events called and all responders matched. While not essential for recalc's functioning, it is of great help for writing and debugging applications based on recalc. Each of the elements of this array are of the form {t: TIMESTAMP, id: STRING, from: STRING|UNDEFINED, verb: ..., path: ..., args: UNDEFINED|ARRAY}. We'll see more of r.log in the last section of the readme.

r.call

So far, responders, store and log are simple objects. Where's the action? Enter r.call. This function calls events.

The standard practice is to say that events are fired or triggered. We use the term call instead, in order to emphasize two things: 1) the complementary nature of calling and responding; 2) the fact that events represent communication.

Every event has two properties:

The combination of a verb plus a path seems versatile enough to serve as a general purpose way of building an event system. It works well for REST and it will probably work well for us too.

Both verb and path are required. If you don't pass them or pass invalid ones, r.call will return false. If its inputs are valid, the event will be called and the function will return a string with the id of the event just called (aren't side effects fun?).

For example, to call an event with verb 'fire' and a path ['hello', 1], you can write:

r.call ('fire', ['hello', 1]);

You can pass unlimited additional arguments when you call an event. These arguments will be received by any responders that are responding to that particular event. For example, if you want to pass 'foo' and 'bar', you can write:

r.call ('fire', ['hello', 1], 'foo', 'bar');

You can also use wildcards for both verbs and paths. For example:

r.call ('*', '*');
r.call ('fire', '*');
r.call ('fire', ['hello', '*']);

Wildcards are further explored in the next section - to really understand them, we need to understand how responders are matched against a new event.

r.respond

This function is the one in charge of placing responders into r.responders. Every time it is invoked with valid arguments, it will add a responder. It takes the following arguments:

As with r.call, if you pass invalid parameters, this function will return false and print an error message. Otherwise it will return the id of the responder.

It is important to notice that while call and respond are complementary operations, r.call and r.respond are not exactly complementary. r.call calls a single event, whereas r.respond creates a responder that can be matched multiple times over the course of a program. In other words, an invocation to r.call represents a single execution (with multiple possible ramifications) whereas r.respond represents multiple possible executions over the course of the program.

Let's see an example: to match the last event we showed on the last section, we could write a responder like this:

r.respond ('fire', ['hello', 1], function () {
   // Your logic here.
});

Remember that we said earlier that the responders object is an object that contains all the responders. Whenever an event is called, recalc iterates all the elements within responders, and executes those that match the event being called. The most important question regarding responders is: when does a responder match an event? And the answer is: when both the verb and path of the called event match those of the responder.

In the case of the verb, matching is straightforward: the verb of the path must be identical to that of the responder.

In the case of the path, matching is also straightforward (as of recalc v5): the paths of both event and responder must have the same length and each of their elements must match. For example:

r.respond ('fire', 'hello', function () {...});

The responder above will match the following event:

r.call ('fire', 'hello');

But it will not match the following events:

r.call ('fire', ['hello', 'there']);

r.call ('fire', 'nothello');

r.call ('fire', []);

Historical notes:

If the path of the responder is a regex (or given element of a path is a regex), said regex will be used to match against the corresponding member of the event's path. For example, this responder:

r.respond ('hello', /foo|bar/, function () {...});

Will be matched by the first and third events below, but not the second.

r.call ('hello', 'foo');
r.call ('hello', 'bach');
r.call ('hello', 'bar');

responder verbs can also be regexes. For example, this responder:

r.respond (/foo|bar/, 'bach', function () {...});

Will match the following two events.

r.call ('foo', 'bach');
r.call ('bar', 'bach');

Let's see now which options can be passed to a responder. options All of them are optional:

This last property reminds us that it is perfectly normal to have more than one responder matching a certain event. priority simply lets us make certain responders to be matched ahead of others. This allows for sophisticated (yet necessary) constructs like view redrawing, where some responders have to be called ahead of others to prevent unnecessary operations. As of recalc >= v4.1.0, responders of equal priority are run first by the order in they were created - older responders are matched before new ones (previous versions matched responders of equal priority in arbitrary order). To keep track of which responders were created first, an index parameter is added to each responder - the lower the index, the earlier the responder was created.

It is also perfectly possible to have zero responders matching a certain event.

Here's an example of how to invoke r.respond with options.

r.respond ('fire', 'test', {id: 'matchEverything', parent: 'someOtherresponderId', priority: 3, burn: false}, function () {...});

You can also place the verb and the path inside the options object. In this case, options becomes the first argument passed to the function, and is now mandatory (since r.respond must receive a verb and a path always).

r.respond ({verb: 'fire', path: 'test', id: 'matchEverything', parent: 'someOtherresponderId', priority: 3, burn: false}, function () {...});

r.respond advanced matching: wildcards & empty paths

Within event and responder paths, you can also use wildcards. For example, this responder will be matched by any event that has a path that's exactly one element long:

r.respond ('*', '*', function () {...});

You can also use wildcards for single elements in an array path. For example, this responder:

r.respond ('fire', ['hello', '*', 'handsome'], function () {...});

Will be matched by the following events:

r.call ('fire', ['hello', 'there', 'handsome']);

r.call ('fire', ['hello', 'out there', 'handsome']);

r.call ('*', ['hello', '*', '*']);

A path can also be an empty array. Conceptually, an empty path represents the root path. For example, this responder:

r.respond ('fire', []);

will only match the following event:

r.call ('fire', []);

But not the following events:

r.call ('fire', '*');

r.call ('fire', ['another', 'path']);

To make a responder match any event, you can pass a match attribute like the following:

r.respond ('*', [], function () {return true});

To make a responder match all event with a matching verb, you can pass a match attribute like the following:

r.respond ('someverb', [], function (ev, responder) {return r.compare (ev.verb, responder.verb)});

In both examples above, the path passed to r.respond makes no difference.

r.forget

This function allows us to remove responders. Its first argument is id, which should be a string or integer. This id is the one of the responder we are removing.

If you call r.forget ('foo'), this will also remove all the responders that have 'foo' as its parent, plus all their children. In effect, the parent property of the responder allows for tree-like deletion of responders.

If you call r.forget ('foo') and there's no responder named foo, this function will return false and print an error.

If you pass a function as the second argument to r.forget, this function will be executed for every responder deleted (that is, the responder with the id specified, plus all its children). This onforget function will receive as first argument the responder that is being deleted.

For example, r.forget ('foo', function (r) {console.log (r.id)}) will print foo, as well as the ids of any responders that have foo as parent.

If you pass a function as the second argument to r.forget, the responder being deleted will be removed from r.responders before this function is executed.

r.count

For the sake of completeness, we mention a seventh element, r.count, which is a mere counter variable to keep track of how many events and responders we added so far. This variable serves for generating incrementing ids for events and responders. The index variable is added by r.respond to each responder references r.count, solely for the purpose of disambiguating the matching order of responders that have the same priority (older responders are matched first).

responder functions

I lied - you also need to understand how responder functions work in order to use recalc. But this is the very last thing.

A responder function always receives an execution context (or merely context) as its first argument. The context is an object with six elements:

As a convention, I name context as x. Let's see an example:

r.respond ('fire', '*', function (x, extraArg1, extraArg2) {
   // Do something...
});

If we call the following event:

r.call ('fire', 'hello');

Then the responder above will receive an x of the form {verb: 'fire', path: ['hello'], cb: ..., responder: {...}, from: ...} and both extraArg1 and extraArg2 will be undefined.

To set extraArg1 and extraArg2 to 'foo' and 'bar' respectively, you could call the following event:

r.call ('fire', 'hello', 'foo', 'bar');

This means that any parameters passed to r.call besides verb and path will be passed to the responder function.

In this case, the x will be {verb: 'fire', path: ['hello'], args: ['foo', 'bar'], cb: ..., responder: {...}, from: ...}.

Having covered the arguments, we now have to see what you to have in mind when writing responder functions. There's two issues: control and data.

Regarding control, if your responder function is synchronous, there's nothing to worry about: once your function finishes executing, the responder functions of any other responders will be executed, one at a time. However, if your function is asynchronous, you need to signal this to recalc, so that the order in which responders are matched will be preserved.

Let's say that you have a responder function that executes an asynchronous operation. The proper way to write it is like this:

r.respond ('fire', 'hello', function (x) {
   asyncOperation ('foo', 'bar', function () {
      x.cb ();
   });
   return x.cb;
});

The above async responder function does two things that every async responder function should do:

  1. Call x.cb from within the asynchronous callback, to resume execution of further matching responders.
  2. Return x.cb, to notify recalc that this is an async function and it should not keep on executing other responders synchronously.

Interestingly enough, if you don't want to resume the execution of further responders, you can skip the call to x.cb from within the asynchronous callback. This may be handy in case of an error.

Note that if two responders are matched for a single call, and the responder with the highest priority is async, the second responder will be only called after the first responder finished its async operation.

Regarding the issue of calling events from within a responder (something which is both possible and necessary), it can be done freely as long as the events match responders with synchronous responder functions. If, however, you call an event that matches one or more asynchronous responder functions, you should do that as the last thing within the body of your responder function, in order to preserve the sequence. Not only that, you should also return x.cb in order to signal that this operation was async, and hence your responder function is async as well.

Regarding data: unless your responder function returns x.cb (which signals that it is async), any other return values will be discarded. If you want to pass data, you should place it in r.store, through a function that calls an event.

Because recalc wants to retain generality, we haven't bundled any functions for calling updates on r.store using the event system. This is done on gotoŠ², which is built on top of recalc. However, consider the following example, where r.store is an array:

r.respond ('setElement', '*', function (x, value) {
   r.store [x.path] = element;
});

r.call ('setElement', 2, 'something');

The two functions above will set the third element of r.store to the value 'something'. This function can be generalized to arrays and objects, and also to support deletions.

Tracking event & responder chains

Whenever an event is called with r.call, or whenever a responder is matched by an event call, an entry will be added to r.log with the following fields: {t: TIMESTAMP, id: STRING, from: STRING|UNDEFINED, verb: ..., path: ..., args: UNDEFINED|ARRAY}. id corresponds to the value returned by a successful invocation to r.call or r.respond (the id of either the event or the responder).

If you are inside a rfun and you want to call further events with r.call, you can pass the context (x) argument as the first argument to r.call. This will allow you to create a trackable chain of events.

This is quite hard to grasp conceptually, so let's see an example:

r.respond ('hello', 'there', {id: 'vamo'}, function (x) {
   r.call (x, 'goodbye', 'goodbye');
});

r.call ('hello', 'there');

We first define a rfun. Whenever it gets matched, we call r.call and we pass the context.

After that, we call an event. Let's say this event will return a value of id1 (just for exemplary purposes; actual ids generated by recalc are longer). When the rfun we defined above is matched, x.from will have a value of id1 (which is the id of the event that matched that rfun).

The invocation to r.call from within the rfun will return another id (let's call it id2). The interesting part will happen in r.log, which will look like this:

[
   {id: 'id1',  from: undefined,  verb: 'hello',   path: ['there'],   t: ..., args: undefined},
   {id: 'vamo', from: 'id1',      verb: 'hello',   path: ['there'],   t: ..., args: undefined},
   {id: 'id2',  from: 'vamo/id1', verb: 'goodbye', path: ['goodbye'], t: ..., args: undefined},
]

As you can see, the log for id2 will mark that this event came from vamo/id1, which denotes that it was called from the responder with id vamo, which in turn was matched by event id1. When an event is called by a responder which in turn was matched by a previous event, recalc puts the id of the responder followed by a slash and the id of the previous event. This is necessary when a responder is matched multiple times - by specifying the id of the event that matched the responder, the chain of execution can be disambiguated.

Any rfuns matched by id2 will receive id2 as its x.from, and if you pass the context to further event invocations, you'll have longer chains of dependencies. In this way, you will have the raw data to track complex (and often asynchronous) events, which can be useful in many applications.

Two more things to have in mind: if you don't pass x to an invocation to r.call within a rfun, recalc has no way of knowing where that event came from and then the chain will be broken. The second one is that initial invocations of r.call (that happen on their own, without being the result of previous events) will naturally receive no context, since there's no context yet for them.

Internally, recalc uses the function B.addLog to add elements to B.log. The function receives the log object that will be pushed to B.log and by default pushes it to B.log. This function can be overridden for different purposes:

r.prod and r.error

In addition to the six main elements of a recalc instance, there's two more that you can optionally use.

The first one is r.prod. If you set this flag to true, all the recalc functions will stop performing validations. This can be useful to speed up a thoroughly debugged application - you can roughly expect to save execution time by ~20% when invoking r.call and by ~90% when invoking r.respond. Please be aware that if you turn on r.prod and there's an error in your code, you'll experience either an exception or unexpected behavior, so be sure to turn this on only when you are very confident of your code.

The second one is r.error. This is a function that is executed whenever one of recalc's functions finds a validation error - if any recalc function receives invalid inputs, it will invoke r.error and pass to it one or more arguments describing the error. r.error is by default mapped to teishi.clog, which prints output to the console and returns false.

If you want to override r.error, make sure that the function returns false, otherwise It defaults to printing errors to the console, but you can override it with another function for custom error reporting.

Turning off a responder

If you set a responders disabled property to true (i.e.: r.responders.SOMEID.disabled = true), said responder won't be matched by any events until the property is set to false or eliminated. This allows for temporarily disabling a responder without having to store it in a variable, remove it and later add it again.

Internals

There's six other functions that support the usage functions. If you override them, you can change the innards of recalc. These are:

Source code

The complete source code is contained in recalc.js. It is about 200 lines long.

Below is the annotated source.

/*
recalc - v5.1.2

Written by Federico Pereiro (fpereiro@gmail.com) and released into the public domain.

Please refer to readme.md to read the annotated source.
*/

Setup

We wrap the entire file in a self-executing anonymous function. This practice is commonly named the javascript module pattern. The purpose of it is to wrap our code in a closure and hence avoid making the local variables we define here to be available outside of this module. A cursory test indicates that local variables exceed the scope of a script in the browser, but not in node.js. This means that this pattern is useful only on the browser.

(function () {

Since this file must run both in the browser and in node.js, we define a variable isNode to check where we are. The exports object only exists in node.js.

   var isNode = typeof exports === 'object';

We require dale and teishi. Note that, in the browser, dale and teishi will be loaded as global variables.

   var dale   = isNode ? require ('dale')   : window.dale;
   var teishi = isNode ? require ('teishi') : window.teishi;

We create an alias to teishi.type, the function for finding out the type of an element. We do the same for teishi.clog, a function for printing logs that also returns false.

   var type = teishi.type, log = teishi.clog;

Constructor

We define a main function that will return a recalc object. A function returning an object which contains the functionality of a library is commonly known as constructor. Recalc is built with a constructor pattern because it is reasonable to assume that you might want to use more than one recalc object at the same time; in this case, the constructor will be invoked once per each object that the user desires to create.

main takes a single argument, a store.

   var main = function (store) {

The store can be undefined, an array or an object. By passing a store to main, the user can already initialize the store to have certain values.

Note we pass true as the fourth argument to teishi.stop. We will do this for every invocation of teishi.stop and teishi.v, to tell teishi not to validate our validation rules. This will yield a (very small) performance improvement.

      if (teishi.stop ([
         ['store', store, ['array', 'object', 'undefined'], 'oneOf']
      ], undefined, true)) return;

We create r, a local object which will be the sole thing returned by the constructor. We'll put everything inside of r.

      var r = {

We set r.responders to be an empty object; as for r.store, we set it to the store passed to main, else we initialize it to an empty object. Notice that the responders are going to be stored outside the store.

         responders: {},
         store:     store || {},

We set r.log to an empty array.

         log:       [],

We set r.error to clog (teishi.clog). By using r.error below to report errors, we make it possible to report errors in a different way other than printing them to the console.

         error:     clog,

Finally we set r.count to an object with two keys e and r, each initialized to 1; we will use this variable to generate ids for events in r.call and in r.respond to register the order in which responders were added and optionally to generate responder ids.

         count:     {e: 1, r: 1}
      }

The three usage functions

We now define the first of the three usage functions, r.call, which serves the purpose of firing an event.

      r.call = function () {

r.call is a variadic function (this means it takes a variable number of arguments). While it must always receive a verb and a path, it can optionally receive a context (named x, here and henceforth). Since the verb cannot be an object, we consider x to be present if the first argument is an object. We then consider the following arguments to be the verb and the path.

         var x    = type (arguments [0]) === 'object' ? arguments [0] : undefined;
         var verb = arguments [x ? 1 : 0];
         var path = arguments [x ? 2 : 1];

A valid path can only be an array. However, it is quite convenient to write paths with a single element (['foo']) as the element itself ('foo'); for this reason, r.call automatically wraps any simple path (ie: neither an array nor an object) into an array, using teishi.simple to detect whether this is the case.

         if (teishi.simple (path)) path = [path];

If r.prod is falsy, we validate the arguments passed to r.call.

         if (! r.prod && teishi.stop ('r.call', [

If x is not undefined, it must be an object (remember that when detecting whether x was passed, we only considered that to happen only if the first argument was an object).

            ['context', x, ['object', 'undefined'], 'oneOf'],

If x is present, we check that x.from should be either a string or undefined.

            [x !== undefined, [function () {
               return ['x.from', x.from, ['string', 'undefined'], 'oneOf'];
            }]],

This concludes our validation of x.

We check that verb is a string.

            ['verb', verb, 'string']

If any part of the validation fails, we return false. Note we pass r.error as an apres argument so that this function will also receive the validation error, if any.

         ], function (error) {
            x ? r.error (x, 'r.call', error) : r.error ('r.call', error);
         }, true)) return false;

If r.prod is falsy, we check that path is valid through a helper function r.isPath that will return true or false. While we'll see the implementation of r.isPath later, this function makes sure that path is an array composed of strings and integers. In case of error, we print it through r.error and return false.

         if (! r.prod && ! r.isPath (path)) return x ? r.error (x, 'r.call', 'Invalid path. Arguments:', {verb: verb, path: path}) : r.error ('r.call', 'Invalid path. Arguments:', {verb: verb, path: path});

We define two local variables. First, oargs, which is a mere reference to arguments. Second, we create args, which will be either an array with all the "extra" arguments received by r.call (that is, arguments beyond x, verb and path). If no extra arguments are passed, this variable will be undefined.

args is done by iterating the arguments pseudo-array and using oargs as a reference to it.

         var oargs = arguments;
         var args  = arguments.length === (x ? 3 : 2) ? undefined : dale.go (dale.times (arguments.length - (x ? 3 : 2), x ? 3 : 2), function (k) {
            return oargs [k];
         });

Notice that if x was passed, extra arguments will be those starting with index 3 (fourth element and above), while if x wasn't passed, extra arguments will be those starting with index 2 (third element and above).

We define a local variable from and set it to either x.responder.id/x.from (if x.responder is present) or x.from, only if x is present. If present, this will represent the id of a previous responder (or a previous event) that is calling the current event. If x.responder is present, x.responder.id will be defined, and so will be x.from (since a responder can only be matched by an event, and events always have ids).

         var from = x ? (x.responder ? (x.responder.id + '/' + x.from) : x.from) : undefined;

We now set x to a brand new object with the following fields: from, with a freshly generated id prepended by an E (using and incrementing r.count.e); the verb and path received as arguments; and args, the array (or undefined) variable we just created above.

Notice that if x was passed to this invocation of r.call, we have already overwritten it and only kept a reference to the original x.from - no other field of the original context is used.

         x = {from: 'E' + r.count.e++, verb: verb, path: path, args: args};

We invoke a helper function, r.addLog, to add a log entry to r.log. We will cover this function later. This function receives the from field from the old context (if present, otherwise it will be undefined), the id of the current event being called, the verb and path received, and the args variable containing further arguments. Note we also pass a timestamp in the t field. Also note that we copy args in case they are modified later, so we have a snapshot of them at the moment we log this event.

         r.addLog ({t: teishi.time (), from: from, id: x.from, verb: verb, path: path, args: teishi.copy (args)});

We call an internal function, r.mill, passing it an array with all the arguments. If there are no extra arguments, we merely pass x wrapped in an array; otherwise, we pass an array with x plus all the extra arguments. Note that x already contains verb and path.

         r.mill (args === undefined ? [x] : [x].concat (args));

We return x.from, which contains an id to identify the current event invocation. This concludes the function.

         return x.from;
      }

We now define r.respond, a function for creating responders, which are functions that are executed when a matching event is called.

      r.respond = function () {

We set two local variables: options (as of yet uninitialized) and rfun (set to the last argument passed to r.respond. options will be an object with options passed to the responder, while rfun will be the responder function.

         var options, rfun = arguments [arguments.length - 1];

r.respond is a variadic function that must receive a minimum of two arguments. If it receives less than two arguments, we will print an error and return false.

         if (arguments.length < 2) return r.error ('r.respond', 'Too few arguments passed to r.respond');

If we receive two arguments only, we'll consider options to be the first one. Since we already set rfun to the last argument received, there's nothing else to do in this case.

         if (arguments.length === 2) options = arguments [0];

If there's more than two arguments, the function expects verb and path to be the first two arguments. If there's four arguments in total, the function expects options to be the third argument. In either case, rfun will be considered to be the last argument. Notice that we place both the verb and path inside options.

         else {
            options      = arguments.length === 3 ? {} : arguments [2];
            options.verb = arguments [0];
            options.path = arguments [1];
         }

In the same way as r.call, r.respond automatically wraps any simple path (ie: neither an array nor an object) into an array, using teishi.simple to detect whether this is the case.

         if (teishi.simple (options.path)) options.path = [options.path];

If r.prod is falsy, we validate the arguments.

         if (! r.prod && teishi.stop ('r.respond', [

options must be an object and only certain keys are allowed in it.

            ['options',   options, 'object'],
            ['keys of options', dale.keys (options), ['verb', 'path', 'id', 'parent', 'priority', 'burn', 'match'], 'eachOf', teishi.test.equal],

options.verb can be either a string or a regex.

            function () {return [
               ['options.verb',     options.verb,     ['string', 'regex'],                'oneOf'],

If defined, both options.id and options.parent must be strings or integers.

               ['options.id',       options.id,       ['string', 'integer', 'undefined'], 'oneOf'],
               ['options.parent',   options.parent,   ['string', 'integer', 'undefined'], 'oneOf'],

If defined, options.priority must be an integer, as options.burn must be a boolean.

               ['options.priority', options.priority, ['undefined', 'integer'],           'oneOf'],
               ['options.burn',     options.burn,     ['undefined', 'boolean'],           'oneOf'],

If defined, options.match must be a function.

               ['options.match',    options.match,    ['undefined', 'function'],          'oneOf']
            ]},

rfun must be a function.

            ['responder function', rfun, 'function']

If any part of the validation fails, we return false. Note we pass r.error as an apres argument so that this function will also receive the validation error, if any.

         ], function (error) {
            r.error ('r.respond', error);
         }, true)) return false;

If r.prod is falsy, we validate path with r.isPath, which we'll define later. In this case, we consider a path as valid if it's an array composed of strings, integers and regexes (which is the reason for passing true as the second argument to r.isPath). If there's an error, we print it through r.error and return false.

         if (! r.prod && ! r.isPath (options.path, true)) return r.error ('r.respond', 'Invalid path. Options:', options);

If options.id is not undefined, we first coerce it into a string (in case it is an integer). Second, if r.prod is falsy, we check that 1) the id has a shape different to Rd... (where d represents a digit) or Ed... since those id forms are used by recalc; and 2) that there's no responder within r.responders with that id. If it does, we print an error and return false.

         if (options.id !== undefined) {
            options.id += '';
            if (! r.prod && options.id.match (/^[R|E]\d+$/)) return r.error ('r.respond', 'If you pass an id for a responder, it cannot be an `R` or an `E` followed by digits, since that is the form that recalc gives to auto-generated ids.');
            if (! r.prod && r.responders [options.id])       return r.error ('r.respond', 'A responder with id', options.id, 'already exists.');
         }

If options.id is undefined, we generate an id of the form R followed by the current value of r.count.r (which is a number).

         else options.id = 'R' + r.count.r;

We place rfun within the options object.

         options.rfun = rfun;

We set options.index to the current value of r.count.r, and increment r.count.r.

         options.index = r.count.r++;

By this point, options contains all the information relevant for the responder - it is effectively the responder object. We now set it within r.responders.

         r.responders [options.id] = options;

We return the id of the responder and close the function.

         return options.id;
      }

We now define the third and last usage function of recalc, r.forget, which has the purpose of removing responders from r.responders.

This function takes two arguments, id and fun. The first one is the id of the responder that must be removed. fun is an optional function that will be executed when the responder is deleted.

      r.forget = function (id, fun) {

If present, fun must be a function; otherwise we print an error and return false. Note we skip the check if r.prod is falsy.

         if (! r.prod && fun !== undefined && type (fun) !== 'function') return r.error ('r.forget', 'Second argument to r.forget must be a function or undefined. Id is:', id);

If r.prod is falsy and the responder does not exist, we print an error and return false.

         if (! r.prod && ! r.responders [id])                            return r.error ('r.forget', 'responder', id, 'does not exist.');

We store the responder in a local variable responder. We then remove it from r.responders.

         var responder = r.responders [id];
         delete r.responders [id];

If fun is present, we invoke it passing responder as its first argument. This is why we stored responder locally, so that we could pass it to this function after the responder was removed from r.responders.

         if (fun) fun (responder);

We iterate all the other responders; if any of them has id as its parent, we recursively call r.forget on it, taking care to also pass fun. This allows for tree-like deletion of responders. Notice that if other responders have id as a parent, fun will be executed multiple times.

         dale.go (r.responders, function (v, k) {
            if (v.parent === id) r.forget (k, fun);
         });

There's nothing else to do, so we close the function.

      }

Internals

We now define r.isPath, a helper function that determines whether a path is valid. Besides path it takes an optional regex flag.

This function is used by the two main functions, once by r.call and once by r.respond.

      r.isPath = function (path, fun, regex) {

path can be either an array, an integer or a string. If it's an array, it must be comprised of integers and strings. If the regex flag is passed, path can also be a regex (and if it's an array, it can also include regexes).

The regex flag is used by r.respond, because responder paths can be (or include) regexes, unlike event paths.

We return the result of invoking teishi.v with these rules. We also pass an empty apres argument to teishi.v so that if there's an error, we don't print it.

         return teishi.v ([
            ['path', path, ['array', 'integer', 'string'].concat (regex ? 'regex' : []), 'oneOf'],
            ['path', path,          ['integer', 'string'].concat (regex ? 'regex' : []), 'eachOf']
         ], function () {}, true);
      }

We define the second helper function, r.compare. It takes two arguments, a verb or path element belonging to an event and a verb or path element belonging to a responder. This function is used later by r.match to compare verbs and paths.

      r.compare = function (eventItem, responderItem) {

If either of the arguments is a string containing a star (wildcard), the function returns true.

         if (eventItem === '*' || responderItem === '*') return true;

If the responder element (its verb or a part of its path), we check whether the corresponding event element matches it. Notice we coerce the event element to a string in case it's an integer. Depending on whether there's a match or not, the function will return true or false.

         if (type (responderItem) === 'regex') return (eventItem + '').match (responderItem) !== null;

If we're here, it means that both elements are either strings or integers. The only way there can be a match is if they're identical. We return the result of this comparison and close the function.

         return eventItem === responderItem;
      }

We define r.addLog, the third and last helper function, which adds log entries to r.log and is invoked only by r.call and r.mill. The function takes a single argument, the log object to be pushed.

      r.addLog = function (log) {
         r.log.push (log);
      }

We now define r.mill, the implementation function that represents the core engine of the library. This function is invoked by r.call. The function takes a single argument, args, which is an array with the context (x) plus other optional arguments.

      r.mill = function (args) {

We define an inner function that will execute each of the responders matching the event in turn. matching, its sole argument, must be an array containing responders.

         var inner = function (matching) {

If matching is empty, there's nothing else to do, so we return undefined.

            if (matching.length === 0) return;

We set up a cb (callback) within the context of args; the purpose of this function is to be executed by asynchronous responder functions, so as to preserve the order in which the responder functions are executed, even when some of them are asynchronous and some of them are not. Notice that the callback simply executes inner recursively.

            args [0].cb = function () {
               inner (matching);
            }

We pick the first responder within matching and remove it from the array.

            var responder = matching.shift ();

We set the responder of the context to the responder itself.

            args [0].responder = responder;

If the responder does not exist, we ignore this responder and call inner recursively. This check is useful if a certain responder function removes other responder functions that were matched by the same event.

            if (! r.responders [responder.id]) return inner (matching);

We invoke r.addLog to add a log entry to r.log for the matched responder. As before, this function receives the from field from the old context (if present, otherwise it will be undefined), the id of the responder being matched, the verb and path of the responder; if args were passed to the responder, we pass them as well, otherwise we pass undefined as the last argument to denote their absence. Note that we copy args so that if they are modified later, we'll have a snapshot of them at the moment of logging.

            r.addLog ({t: teishi.time (), from: args [0].from, id: responder.id, verb: responder.verb, path: responder.path, args: args.slice (1).length ? teishi.copy (args.slice (1)) : undefined});

If the responder has the burn attribute, we invoke r.forget to remove it.

            if (responder.burn) r.forget (responder.id);

inner does two more things. First, it invokes the responder function of the responder, passing args to it. Second, if the rfun did not return a function, then it invokes itself recursively.

Why does inner avoid calling itself to process the next responder if rfun returns a function? Because in case rfun is asynchronous, it must signal that to recalc, and to do so it returns a function. However, the asynchronous rfun is responsible for invoking inner to execute any other responders; by making cb available to the rfun, we allow asynchronous functions to resume execution of other matching responders for the same event.

            if (type (responder.rfun.apply (null, args)) !== 'function') inner (matching);
         }

The last thing left to do is to call inner with the matching responders. For this, we invoke r.match with the verb, path, any extra arguments passed to the event invocation, and the list of responders. We then invoke r.sort on the output of r.match, so that the responders are placed in the right order.

         inner (r.sort (r.match (args [0].verb, args [0].path, r.responders, args.slice (1))));
      }

This concludes r.mill. Let's now define r.match, which is the function that takes a verb, a path, optional args and a list of responders and returns only those responders that match an incoming event.

Instead of reading r.responders directly, it is possible to use r.match to scan a different set of responders - this is the reason for which responders is an argument instead of internally being set to r.responders.

      r.match = function (verb, path, args, responders) {

We initialize a matching array to store the responders that match the verb and path.

         var matching = [];

We iterate the responders.

         dale.go (responders, function (responder) {

If the responder is disabled, we ignore it.

            if (responder.disabled) return;

If the responder contains a match function, we use that function to determine whether the responder matches the event. We invoke that function passing an object with verb, path and args, plus the responder. Depending on whether that function returned true or not, we push the responder to matching; in any case, we're done processing this particular responder.

            if (responder.match) return responder.match ({verb: verb, path: path, args: args}, responder) === true ? matching.push (responder) : undefined;

We invoke r.compare to compare the verb of the responder with that of the event. If the comparison is not successful, the responder doesn't match.

            if (! r.compare (verb, responder.verb)) return;

If we're here, the verbs match!

If the paths don't have equal length, the responder cannot be a match, so we return undefined.

            if (path.length !== responder.path.length) return;

If both the responder and the event have paths of length 0, then the responder will be a match. We push responder to the list of matching responders.

            if (path.length === 0 && responder.path.length === 0) return matching.push (responder);

We compare both paths. If any of the comparisons returns false, the loop will be interrupted.

            if (dale.stop (path, false, function (v, k) {

We use r.compare to compare the kth elements of both paths. r.compare returns only true or false, and if it returns the latter it means that the comparison failed.

               return r.compare (path [k], responder.path [k]);

If the last iteration of the loop returned true, all iterated elements of both paths match with each other. If this is the case, we push the responder to matching. Notice that if any of the iterations returned false, this will not happen.

            })) matching.push (responder);

We close the iteration of the responders.

         });

We return the array of matching responders and close the function.

         return matching;
      }

We define r.sort, a function that takes a list of responders (outputted by r.match) and sorts it by priority.

      r.sort = function (matching) {

We apply sort on matching, using a custom sort function. Notice that we modify matching instead of returning a new sorted array.

         return matching.sort (function (a, b) {

We sort the responders based on their priority value. If no priority is defined, we'll consider the responder to have a priority of 0. The higher the priority, the closer a responder will be to the beginning of the array.

If both priorities are the same, we disambiguate by giving priority to the responder with the lower index. Since oldest responders have a lower index, given the same priority, the older responder will get matched first.

            var priority = (b.priority || 0) - (a.priority || 0);
            return priority !== 0 ? priority : a.index - b.index;
         });
      }

We are done defining recalc's methods. We return the r object and close main.

      return r;
   }

If we are in node, we set module.exports to main. If we're in the browser, we create a global variable named R and place main there.

   if (isNode) module.exports  = main;
   else        window.R        = main;

We close the module.

}) ();

License

recalc is written by Federico Pereiro (fpereiro@gmail.com) and released into the public domain.