Herder
This library allows you to create machines that process a list of instructions using an optional list of processors (called "actors").
Actors receive the current result set. Actors can be invoked serially or in parallel.
Machines are started using the start method. Pass an Array to start, which array the actors will act upon.
The machines are evented, such that a data event is emitted on each iteration through the instruction list, to which is made available the current result set. When the machine is finished it emits end.
Additionally, the machine can be transformed into a state machine, where state change events are programmable.
Because machines maintain their last state, they can function like Promises, in that one may create a machine and pass it around while it is running. Once it has run, it's results remain available. If it is restarted, your handlers will receive the new data. A machine is a realtime interface to some aspect of your application state.
Herder is also a straightforward nested callback taming tool.
Flattening function calls and executing serially:
herder
.serial([1,2,3,4,5])
.actor(
function(it, idx, next) {
setTimeout(function() {
next(it * 2);
}, (Math.random() * 1000));
},
function(it, idx, next) {
setTimeout(function() {
next(it & 1);
}, (Math.random() * 1000));
}
)
.on("data", function() {
console.log("serial data last: " + this.results.last());
})
.on("result", function() {
console.log("FULL SERIAL RESULT OBJECT");
console.log(this.results.stack());
})
.start();
// serial data last: 2
// serial data last: 4
// serial data last: 6
// serial data last: 8
// serial data last: 10
// serial data last: 1
// serial data last: 0
// serial data last: 1
// serial data last: 0
// serial data last: 1
// FULL SERIAL RESULT OBJECT
// [2, 4, 6, 8, 10, 1, 0, 1, 0, 1]
Flattening function calls and executing in parallel:
herder
.parallel([1,2,3,4,5])
.actor(
function(it, idx, next) {
setTimeout(function() {
next(it * 2);
}, (Math.random() * 1000));
},
function(it, idx, next) {
setTimeout(function() {
next(it & 1);
}, (Math.random() * 1000));
}
)
.on("data", function(idx) {
console.log("parallel data idx: " + idx + " last: " + this.results.last());
})
.on("result", function() {
console.log("FULL PARALLEL RESULT OBJECT");
console.log(this.results.stack());
})
.start();
// parallel data idx: 3 last: 8
// parallel data idx: 4 last: 10
// parallel data idx: 0 last: 2
// parallel data idx: 2 last: 6
// parallel data idx: 1 last: 4
// parallel data idx: 2 last: 1
// parallel data idx: 0 last: 1
// parallel data idx: 1 last: 0
// parallel data idx: 3 last: 0
// parallel data idx: 4 last: 1
// FULL PARALLEL RESULT OBJECT
// [2, 4, 6, 8, 10, 1, 0, 1, 0, 1]
A map machine is easy to create:
var mapper = herder
.parallel([1,3,5,7,9])
.on("result", function() {
console.log("MAP -> " + this.results.stack());
})
.actor(function(it, idx, next) {
next(it * 2);
})
.start();
// MAP -> 2,6,10,14,18
Your new mapper machine can be re-used:
mapper.start([1,2,3,4,5]);
// MAP -> 1,2,3,4,5
mapper.start([10,20,30,40,50]);
// MAP -> 10,20,30,40,50
Machines can be re-used, redefined, or take a new list:
mapper
.actor(function(it, idx, next) {
next(it.toUpperCase());
})
.start(["foo","bar"]);
// MAP -> FOO,BAR
Machines can throw errors, and (optionally) can be stopped:
mapper
.on("error", function() {
console.log("MAP ERRORED...");
console.log(this.results.error());
this.stop();
})
.on("stop", function() {
console.log("MAP STOPPED...");
console.log(this.results.stats());
})
.actor(function(it, idx, next) {
next(this.results.error("Something bad happened"));
})
.start([1,2])
Machine events have access to the result object, so purely evented machines are possible, without actors. Simply update results on data events:
var eventedMap = herder
.parallel()
.on("data", function(idx) {
this.results.actual(10 * this.results.last());
})
.on("result", function() {
console.log("PURELY EVENTED MAP -> " + this.results.actual());
});
eventedMap.start([1,2,3,4,5]);
// PURELY EVENTED MAP -> 10,20,30,40,50
eventedMap.start([6,7,8,9,10]);
// PURELY EVENTED MAP -> 60,70,80,90,100
You can create a reduce method:
var reducer = herder
.serial()
.actor(function(it, idx, next) {
next(this.results.last() ? this.results.last() + it : it);
})
.on("result", function() {
console.log("REDUCED");
console.log(this.results.last());
});
reducer
.start([10,10,10]);
// REDUCED -> 30
reducer
.start(["a","b","c"]);
// REDUCED -> abc
Or a filter (see ResultObject):
herder
.parallel()
.actor(function(it, idx, next) {
it & 1 || this.results.actual(it);
next();
})
.on("result", function() {
console.log("FILTERED EVEN -> " + this.results.actual());
})
.start([1,2,3,4,5,6,7,8,9])
// FILTERED EVEN -> 2,4,6,8
.start([234,1,777,4,6,8,2001])
// FILTERED EVEN -> 234,4,6,8
Machines can be given a context to work with. Here we implement a find (or, some) machine that matches its instruction list against a given context value:
var somePet = herder
.parallel()
.actor(function(it, idx, next) {
if(this.context() !== it) {
return next();
}
this.stop();
return this.emit("result", idx, it);
})
.context("cat")
.on("result", function(idx, item) {
console.log(item + " EXISTS AT -> " + idx);
})
.start(["fish","dog","cat","turtle"])
// cat EXISTS AT -> 2
Note how a Machine can be stopped at any time. In the find method above, we are stopping the moment a match is found (at least one target item exists). Here we use this method to implement every:
var every = herder
.parallel()
.actor(function(it, idx, next) {
it !== this.context() && this.stop();
next();
});
every
.context(2)
.on("stop", function(idx) {
console.log("EVERY: false");
})
.on("result", function(bool) {
console.log("EVERY: true");
})
.start([2,2,2,2])
// EVERY: true
every
.start([2,2,3,2,2])
// EVERY: false
Note that a stopped machine can always be restarted with start.
The instruction list can be pushed onto during iteration, such that an initial list can be dynamically augmented. Here we keep adding to the original buffer passed to start until that buffer reaches 20 elements:
herder
.parallel()
.actor(function(it, idx, next) {
if(this.buffer.length() < 20) {
this.buffer.push(Math.floor(Math.random() * 100));
}
next(it * 2);
})
.on("result", function() {
console.log("HOT PUSH: " + this.results.stack());
})
.start([1,2,3,4,5]);
// The original array doubled, concat doubled randomly pushed values:
// HOT PUSH: 2,4,6,8,10,128,14,184,66,40,124,114,32,126,98,30,100,150,156,154 <- length: 20
A total running time (milliseconds) can be set using timeout. Here we set it for 200 milliseconds. As our actors will (likely) consume more than 200 milliseconds of time, this machine should time out:
herder
.serial()
.timeout(200)
.on("timeout", function(idx) {
console.log("TIMED OUT AT INDEX: " + idx, "WITH RESULTS: " + this.results.stack());
})
.on("result", function() {
console.log("This only executes if timeout is NOT flagged.");
})
.actor(function(it, idx, next) {
setTimeout(next, Math.floor(Math.random() * 100), it);
})
.start([11,22,33,44,55,66,77,88,99,1010]);
Try changing the above timeout argument to 10000.
Errors can be flagged:
herder
.parallel()
.actor(function(it, idx, res, next) {
next(res.error("Boo"));
})
.on("error", function(res, idx) {
console.log("!!!!!!!!!ERRORED!!!!!!!!!");
console.log(res.error());
})
.start()
Note that you are passing an error state on to the next iteration, at which point it can be handled. The goal of error is to indicate to other actors that an error has occurred upstream, which may or may not matter, depending on the machine's purpose. In other words, flagging an error does not automatically stop the machine -- use stop explicitly if that is needed.
If the error indicates an exception you should throw instead.
Asynchronicity is expected at the functional level -- your actors (functions) are making asynchronous calls. However, sometimes you will want a long operation (such as iterating a very long list) to be non blocking, especially in a NodeJS environment. You can force the machine to yield to the javascript execution context's event loop on each iteration with async:
.async()
The cases where this would be necessary are very rare: you probably want function-level async.
State Machine
To create a state machine, pass a definition object to addState (see StateMachine):
herder
.serial()
.addState({
initial: 'none',
events: [
{ name: 'openTag', from: ['none','inner','open','closed'], to: 'open' },
{ name: 'closeTag', from: ['inner','closed','open'], to: 'closed'},
{ name: 'innerHTML', from: ['open','inner','closed'], to: 'inner'}
]})
.actor(
function(it, idx, res, next) {
if(it.match(/^<\/[^>]+>$/)) {
this.state.closeTag();
} else if(it.match(/^<[^>]+>$/)) {
this.state.openTag();
} else {
this.state.innerHTML();
}
next(this.state.current);
}
)
.on("end", function(res) {
console.log("HTML STATE END");
})
.on("openTag", function() {
console.log("OPEN_TAG EVENT....");
})
.on("enteropen", function() {
console.log("ENTER OPEN STATE...");
})
.on("leaveopen", function() {
console.log("LEAVE OPEN STATE...");
})
.start(['<html>','<div>','hello','</div>','</html>']);
// ENTER OPEN STATE...
// OPEN_TAG EVENT....
// OPEN_TAG EVENT....
// LEAVE OPEN STATE...
// HTML STATE END
As you can see, multiple from states are possible, sent as an array.
In addition to setting an initial state, one can also set a terminal state:
herder
.serial()
.addState({
initial : 'none',
terminal : 'done',
events: [
{ name: 'myCompleteEvent', from: 'none', to: 'done' }
]})
.actor(
function() {
this.state.myCompleteEvent();
}
)
.on("myCompleteEvent", function(res) {
console.log("COMPLETE EVENT");
})
.on("enterdone", function() {
console.log("ENTER DONE STATE");
})
.start();
State machine control flow translates easily to evented models:
var login = herder
.serial()
.addState({
initial: 'none',
terminal: 'confirmed',
events: [
{ name: 'candidate', from: 'none', to: 'candidate'},
{ name: 'accepted', from: ['candidate', 'denied'], to: 'accepted'},
{ name: 'denied', from: ['candidate','accepted'], to: 'denied'},
{ name: 'confirmed', from: 'accepted', to: 'confirmed'}
]})
.on("candidate", function(ev, from, to, creds) {
if(creds.password === "safe!") {
return this.state.accepted(creds);
}
this.state.denied(creds);
})
.on("accepted", function(ev, from, to, creds) {
var serverLoad = 13;
if(serverLoad < 20) {
return this.state.confirmed(creds);
}
this.state.denied(creds);
})
.on("denied", function(ev, from, to, creds) {
console.log("DENIED");
console.log(creds);
this.stop();
})
.on("confirmed", function(ev, from, to, creds) {
console.log("CONFIRMED");
console.log(creds);
})
.on("finished", function(ev, from, to, creds) {
console.log("FINISHED......");
})
.start(function(idx, res, next) {
this.state.candidate({
username : "bobloblaw",
password : "safe!"
});
next();
})
State machines naturally throw on undefined transitions:
setTimeout(function() {
login.start(function() {
this.state.accepted();
// Error: event accepted inappropriate in current state confirmed
});
}, 100);