Saturday, May 31, 2014

JavaScript Performance tip using a Closure

Here is a simple performance JavaScript tip using a JavaScript closure that is so quick to explain you have no reason not to keep reading.   Consider this: you want a function which will tell you the position on a rugby team for a certain number. Those of us who are familiar with the sport will know that in rugby, the various positions always have the same number.
  1. Loose head
  2. Hooker
  3. Tight head
  4. Lock
  5. Lock
  6. Flanker (blind side)
  7. Flanker (open side)
  8. No. 8
  9. Scrum half
  10. Out half
  11. Left wing
  12. Inside Centre
  13. Centre
  14. Right wing
  15. Full back
So for your first attempt you define a positions array in the global namespace and then a function to resolve the position for a given number.
var positions = ['No position', 'Loose head', 'Hooker', 'Tight head', 'Lock', 'Lock', 'Flanker', 'Flanker', 'No. 8',  'Scrum half', 'Out half', 'Left wing', 'Inside centre', 'Outside centre', 'Full back']

var getPosition = function(n) {
   return positions(n);
}
Everyone starts laughing at you at code review time. You get comments such as:
"Don't pollute the name space, re-write please."
"Did you even bother using JsLint?"
"Do you want fries with that?"

For your second attempt, you put the positions array into a function as a local variable.
var getPosition = function() {
    var positions = ['No position', 'Loose head', 'Hooker', 'Tight head', 'Lock', 'Lock', 'Flanker', 'No. 8', 'Flanker', 'Scrum half', 'Out half', 'Left wing', 'Inside centre', 'Outside centre', 'Right wing' 'Full back'];   // positions is a local variable
   return positions[n];
}
Back at the second code review:
"Well it's better, but you do realise that every time that function is called, the array is allocated?"
"But you said, don't pollute the global name space"
"Look we have standards here, don't pollute global name space and don't incur performance costs when you don't need do"
"Ok"

You leave the code review despondent and crestfallen. Back at your coffee stained desk, when your headphones are on low volume, you keep hearing the word closure, closure, closure from a few of the JavaScript nerds sitting near you.  That evening, you are having difficult getting to sleep.  Closure, closure, closure.  The word won't leave your head.  Closure, closure, closure.   Yeah you know that closures offer a way to encapsulate code in JavaScript, but this isn't about encapsulation. Wait a sec! Closures close over free variables and make them accessible to the lexical scope. Hmmm... This means you could do something like...
var getPosition = function () {
    var positions = ['No position', 'Loose head', 'Hooker', 'Tight head', 'Lock', 'Lock', 'Flanker', 'Flanker', 'No. 8', 'Scrum half', 'Out half', 'Left wing', 'Inside centre', 'Outside centre', 'Right wing', 'Full back'];   // positions is a local variable
    return function(n) {
        return positions[n];
    }
}()

console.log(getPosition(4));
What's going on here?
  • The inner function "closes over" the positions array
  • The outer function returns the inner function, which is then assigned it to the variable getPosition
  • The outer function is self invoking. This means the declaration, invocation and assignment will happen all at the same time.
  • Because it is the outer function which allocated the array and because the outer function is only invoked once, the array is only allocated once.
  • Because the inner function has access to what it "closes over" even after the outer function has executed, it means the inner function will have access to the positionsarray (that is only allocated once).
So you return to code review a little more sanguine. You get one comment:
"Third time lucky"

You might be thinking, it doesn't take that long to allocate an array in JavaScript is this not overkill? Well, it depends. Is your application needing to squeeze every nano second? Is the amount of data you need to initialise a lot more than 15 elements in an array? These are pertinent questions of course. But the technique is a handy one to know because encapsulation is clearly not the only advantage of closures.

Saturday, May 24, 2014

Grails tip: refactoring your URLs

On the current project I am working we use a lot of integration tests. For you non-Grails users out there, Integration tests test your Controller APIs, your Services and any persistence that might happen all very neatly. The only slice of the cake they don't test from a back end perspective are your Grails filters for which you'd need something like a functional test. In the Grails world, Controller API's are mapped to URL requests in the URLMappings.groovy file. This is just a simple Groovy to configure which HTTP request go to which Controller.

For example:

class UrlMappings {
    static mappings = {
        "/sports/rugby/ball" (controller: "rugbyBall", action = [POST: "createBall", DELETE: "removeBall", GET: "getBall"])
        ...

So in the above example, the HTTP request /sports/rugby/ball will go to the RugbyBallController and will go to the methods: createBall(), deleteBall(), getBall() depending on weather the request is a GET, POST or DELETE. Now suppose, you have your project all set up to server up the CRUD operations for the rugby ball and after a few hectic sprints some software entropy creeps and you need to refactor your Controller APIs but before you race ahead and do that your project manager looks you in the eye and says:

"You must support all existing APIs as clients are using them".

This is how generally refactoring works in the real world when things go into production. There is always a phase of supporting of old and new, deprecating the old and then when everyone is happy removing it. Anyway, you begin by updating your URLMappings.groovy
class UrlMappings {
    static mappings = {
        // Old APIs
        "/sports/rugby/ball" ( controller: "rugbyBall", action = [POST: "oldCreateBall", DELETE: "oldRemoveBall", GET: "oldGetBall"])
        ...

        // New APIs
        "/sports/rugby/v2/ball" ( controller: "rugbyBall", action = [POST: "createBall", DELETE: "removeBall", GET: "getBall"])
        ...

The URLMappings.groovy show the old and the new. The old APIs are going to controller methods that you have renamed. Clients using these APIs are not impacted because they only send HTTP requests, they do not which Controller are behind these endpoints. The old APIs already have really good integration tests and our project manager has mandated that the new APIs must have similar quality integration tests before they go anywhere near pre-production.
def "test adding a single item to your cart"() {
    setup: "Set up the Cart and Landing Controller"
       //...
    when:
       //...
       rugbyBallController.oldGetBall();
       rugbyBall = JSON.parse(rugbyBallController.response.contentAsString)
    then:
       rugbyBall.isOval();
Mr. Project manager says: "I want this all new tests added by Friday or you are not going for a pint after work. You need a quit way to get your integration tests done". Thinking about that cool larger and its quenching effect on the back your throat, you remember Groovy's excellent support for invoking methods dynamically where you can specify the name of the method as a variable.
   myObject."$myMethod"()  // myMethod is a Groovy String variable.
In the above code snippet, myMethod is a variable that corresponds to the name of method you want to invoke on myObject. "$myMethod" means, evaluate the variable myMethod (which of course will be the method name), the () of course just means invokes the method. Eureka moments happens when you remember that the old and new APIs will return the exact same JSON. All you need to do is run the same test twice, once for the old code and once for the new. Since you are using the spock framework for your integration tests that's easily achieved using a where block.
def "test adding a single item to your cart"(String method) {
    setup: "Set up the Cart and Landing Controller"
       //...
    when:
       //...
       rugbyBallController."$method"();
       rugbyBall = JSON.parse(rugbyBallController.response.contentAsString)
    then:
       rugbyBall.isOval();
    where:
       method = ["oldGetBall", "getBall"]
Happy days. Now go off and drink that lager.

Saturday, May 17, 2014

The magic of Groovy's with...

In a recent post, this blog examined the meaning of Closures' delegates and how they could be changed. This post will show how the ability to set the closure's delegate allows the language to become quite expressive where other languages remain cumbersome and verbose. For the purpose of example (and some fun) consider there is a groovy object drico, which represents the legendary Irish rugby player, Brian O'Driscoll

In case you don't know, Brian O'Driscoll is the best Rugby player to ever come from Ireland.  He could tackle, break, off load, ruck, jackal, kick you name it. An exceptional skilful athlete and widely regarded as one of the greatest players the game has ever seen.  In case, you don't know what Rugby is think American Football with no helmets, timeouts and a slightly longer shaped oval ball.   It is basically superior in every regard.   Anyway, so we have the drico object in our Groovy code suppose and we want to invoke a bunch of methods on it.  We could find ourselves doing:

RugbyPlayer drico = new RugbyPlayer();
...
drico.youWereAmazing();
drico.seriouslyYouWereAmazing();
drico.sayTotsAmazeBalls();
drico.howManyTriesDidYouScore();
drico.youreTheBest();
...
Wow, that's a lot methods being invoked on Drico.  There must be some Groovy trick we could do to lessen the text here? Remember, it is generally better to be coding than typing

Well we could put all the methods we want to invoke on Drico into a Closure and then set that Closure's delegate to Drico.
RugbyPlayer drico = new RugbyPlayer();
...
def dricoClosure = {
    youWereAmazing();
    seriouslyYouWereAmazing();
    sayTotsAmazeBalls();
    howManyTriesDidYouScore();
    youreTheBest();
    ...
}

dricoClosure.delegate = drico
dricoClosure();
Not bad. We have removed all the drico., and just gone straight to the method invocations but we have also added some code setting up the closure delegate. So one step forward, one step back. We can get rid of the one step back by using Groovy's with method. The with method is applicable to any Groovy object and accepts one closure. The delegate of the closure is the object with is being invoked on. This means we can do this:
RugbyPlayer drico = new RugbyPlayer();
drico.with {
    youWereAmazing();
    seriouslyYouWereAmazing();
    sayTotsAmazeBalls();
    howManyTriesDidYouScore();
    youreTheBest();
    ...
}
Using this approach means we have still removed all the drico. and gone straight to the method invocations. It also means we don't need to set the closure's delegate and we don't need to explicitly invoke the closure. The result is we have code with a much better signal to noise ratio and which is far more expressive. Until the next time take care of yourselves.

Saturday, May 3, 2014

Groovy Closures: this, owner, delegate let's make a DSL.

Groovy closures are super cool. To fully understand them, I think it's really important to understand the meaning of this, owner and delegate. In general:
  • this: refers to the instance of the class that the closure was defined in.
  • owner: is the same as this, unless the closure was defined inside another closure in which case the owner refers to the outer closure.
  • delegate: is the same as owner. But, it is the only one that can be programmatically changed, and it is the one that makes Groovy closures really powerful.
Confused? Let's look at some code.
class MyClass {
  def outerClosure = {
    println this.class.name    // outputs MyClass
    println owner.class.name    // outputs MyClass
    println delegate.class.name  //outputs MyClass
    def nestedClosure = {
      println this.class.name    // outputs MyClass
      println owner.class.name    // outputs MyClass$_closure1
      println delegate.class.name  // outputs MyClass$_closure1
    }
    nestedClosure()
  }
}

def closure = new MyClass().closure
closure()
With respect to above code:
  • The this value always refers to the instance of the enclosing class.
  • owner is always the same as this, except for nested closures.
  • delegate is the same as owner by default. It can be changed and we will see that in a sec.
So what is the point of this, owner, delegate? Well remember, that closures are not just anonymous functions. If they were we could just call them Lambdas and we wouldn't have to come up with another word, would we?

Where closures go beyond lambdas is that they bind or "close over" variables that are not explicitly defined in the closure's scope. Again, let's take a look at some code.

class MyClass {
  String myString = "myString1"
  def outerClosure = {
    println myString;     // outputs myString1
    def nestedClosure = {
       println myString;  // outputs myString1
    }
    nestedClosure()
  }
}

MyClass myClass = new MyClass()
def closure = new MyClass().outerClosure
closure()

println myClass.myString 
Ok, so both the closure and the nestedClosure have access to variables on the instance of the class they were defined in. That's obvious. But, how exactly do they resolve the myString reference? Well it's like this. If the variable was not defined explicitly in the closure, the this scope is then checked, then the owner scope and then the delegate scope. In this example, myString is not defined in either of the closures, so groovy checks their this references and sees the myString is defined there and uses that. Ok, let's take a look at an example where it can't find a variable in the closure and can't find it on the closure's this scope, but it can find's it in the closure's owner scope.
class MyClass {
  def outerClosure = {
    def myString = "outerClosure";     
    def nestedClosure = {
       println myString;  // outputs outerClosure
    }
    nestedClosure()
  }
}

MyClass myClass = new MyClass()
def closure = new MyClass().closure
closure()
In this case, Groovy can't find myString in the nestedClosure or in the this scope. It then checks the owner scope, which for the nestedClosure is the outerClosure. It finds myString there and uses that. Now, let's take a look at an example where Groovy can't find a variable in the closure, or on this or the owner scope but can find it in on closure's delegate scope. As discussed earlier the owner delegate scope is the same as the owner scope, unless it is explicitly changed. So, to make this a bit more interesting, let's change the delegate.
class MyOtherClass {
  String myString = "I am over in here in myOtherClass"
}

class MyClass {
  def closure = {
    println myString
  }
}


MyClass myClass = new MyClass()
def closure = new MyClass().closure
closure.delegate = new MyOtherClass()
closure()   // outputs: "I am over in here in myOtherClass"
The ability to have so much control over the lexical scope of closures in Groovy gives enormous power. Even when the delegate is set it can be change to something else, this means we can make the behavior of the closure super dynamic.
class MyOtherClass {
  String myString = "I am over in here in myOtherClass"
}

class MyOtherClass2 {
  String myString = "I am over in here in myOtherClass2"
}

class MyClass {
  def closure = {
    println myString
  }
}


MyClass myClass = new MyClass()
def closure = new MyClass().closure
closure.delegate = new MyOtherClass()  
closure()     // outputs: I am over in here in myOtherClass

closure = new MyClass().closure
closure.delegate = new MyOtherClass2() 
closure()     // outputs: I am over in here in myOtherClass2
Ok, so it should be a bit clearer now what this, owner and delegate actually correspond to. As stated, the closure itself will be checked first, followed by the closure's this scope, than the closure's owner, then its delegate. However, Groovy is so flexible this strategy can be changed. Every closure has a property called resolvedStrategy. This can be set to:
  • Closure.OWNER_FIRST
  • Closure.DELEGATE_FIRST
  • Closure.OWNER_ONLY
  • Closure.DELEGATE_ONLY
So where is a good example of a practical usage of the dynamic setting of the delegate property. Well you see in the GORM for Grails. Suppose we have the following domain class:
class Author {
   String name 

   static constraints = {
       name size: 10..15
   }
}
In the Author class we can see a constraint defined using what looks like a DSL whereas in the Java / Hibernate world we would not being able to write an expressive DSL and instead use an annotation (which is better than XML but still not as neat as a DSL). So, how come we can use a DSL in Groovy then? Well it is because of the capabilities delegate setting on closures adds to Groovy's metaprogramming toolbox. In the Author GORM object, constraints is a closure, that invokes a name method with one parameter of name size which has the value of the range between 10 and 15. It could also be written less DSL'y as:
class Author {
   String name 

   static constraints = {
       name(size: 10..15)
   }
}
Either way, behind the scenes, Grails looks for a constraints closure and assigns its delegate to a special object that synthesizes the constraints logic. In pseudo code, it would be something like this...
// Set the constraints delegate

constraints.delegate = new ConstraintsBuilder();  // delegate is assigned before the closure is executed.

class ConstraintsBuilder = {
  //
  // ...
  // In every Groovy object methodMissing() is invoked when a method that does not exist on the object is invoked
  // In this case, there is no name() method so methodMissing will be invoked. 
  // ...
  def methodMissing(String methodName, args) {
     
     // We can get the name variable here from the method name
     // We can get that size is 10..15 from the args
     ...
     // Go and do stuff with hibernate to enforce constraints
  }
}
So there you have it. Closures are very powerful, they can delegate out to objects that can be set dynamically at runtime. That plays an important part in Groovy's meta programming capabilities which mean that Groovy can have some very expressive DSLs.