Spaceships, Elvis, and Groovy inject

When I first started learning Groovy, I took to collect pretty quickly. The current trend of adopting “functional programming” practices works well in Groovy, though the names of the methods are somewhat surprising. For example, collect is like map, findAll is the equivalent of filter, and inject is the proposed replacement for reduce (or whatever the similar process is for your favorite language).

As a trivial example, in Groovy you can write:

(1..20).collect { it * 2 }      // double them all
       .findAll { it % 3 == 0 } // find the doubles divisible by 3
       .sum()                   // add them up

which is a functional style, even though Groovy is an object-oriented language.

Notice, however, that I didn’t use inject. That’s not an accident, of course. For years, while I “got” collect and findAll, I never found a usage for inject that couldn’t be done in an easier way. The inject version of the above example would be:

(1..20).collect { it * 2 }                   // double them all
       .findAll { it % 3 == 0 }              // find the doubles divisible by 3
       .inject(0) { acc, val -> acc + val }  // add them up

That seems like a lot of work compared to the sum method, especially when I always had trouble remembering exactly what the arguments to the closure meant.

That changed recently. One of the examples I use when teaching Groovy to Java developers is do some basic sorting. I like that example, because it shows not only how easy it is to replace anonymous inner classes with closures, but also because it shows how much the Groovy JDK simplifies coding.

As a preface to my example, consider making an ArrayList of strings:

def strings = 'this is a list of strings'.split()
assert strings.class == java.lang.String[]

The split method splits the string at spaces by default, and returns, sadly, a string array. What I want is a List, and converting an array into a List is a special blend of Java awkwardness and verbosity, though the code isn’t too bad once you’ve seen it.

The conversion is trivial in Groovy, however.

List strings = 'this is a list of strings'.split()
assert strings.class == java.util.ArrayList

Just replace def with the datatype you want, and Groovy will do its best to do the conversion for you. :)

To sort a list, Java has the various static sort methods in the java.util.Collections class. The sort method with no arguments does the natural, alphabetical (more properly, lexicographical, where there capital letters come before the lowercase letters) sort.

List strings = 'this is a list of strings'.split()
Collections.sort()  // natural sort (alphabetical)
assert strings == ['a', 'is', 'list', 'of', 'strings', 'this']

This sorts the strings in place (a destructive sort) and returns void, so to see the actual sort you have to print it.

How do you test this? I wrote an assert that hardwired the results, because I knew what they had to be. That’s hardly generalizable, however, and this is where inject comes in.

Have you ever looked at the definition of inject in the GroovyDocs? Here it is, from the class org.codehaus.groovy.runtime.DefaultGroovyMethods.

public static T inject(E[] self, U initialValue, @ClosureParams(value=FromString.class,options=”U,E”) Closure closure)

Iterates through the given array, passing in the initial value to the closure along with the first item. The result is passed back (injected) into the closure along with the second item. The new result is injected back into the closure along with the third item and so on until all elements of the array have been used. Also known as foldLeft in functional parlance.

Parameters:
self – an Object[]
initialValue – some initial value
closure – a closure

Returns:
the result of the last closure call

You would be forgiven for being seriously confused right now. I’ve been using Groovy since about 2007 and if I didn’t already know what inject did, I’d be seriously confused, too.

The DefaultGroovyMethods class contains lots of methods that are added to the library at runtime via Groovy metaprogramming. In this case, the inject method is added to collection (the first argument above). The second argument to inject is an initial value. An initial value to what, you say? The third argument to inject is a closure, and it takes two arguments, and the second argument to inject is the initial value of the first argument of the closure. The subsequent values to that argument are the result of the closure. The second argument to the closure is each element of the collection, in turn.

I expect that almost nobody actually read that last paragraph. Or, more likely, you started it and abandoned it somewhere in the middle. I can hardly blame you.

As usual, the indefatigable Mr. Haki comes to the rescue. Here is one of his examples:

(1..4).inject(0) { result, i ->
    println "$result + $i = ${result + i}"
    result + i
}

and the output is:

0 + 1 = 1
1 + 2 = 3
3 + 3 = 6
6 + 4 = 10

The value of result starts at the initial value (here, 0), and is assigned the result of each execution of the closure.

That’s the sort (no pun intended) of example I’d seen before, and because I always associated inject with accumulators, I never actually needed it. After all, the Groovy JDK adds a sum method already.

The key is to note that the value of “result” is actually whatever is returned from the closure (Mr. Haki’s second example illustrates this beautifully — seriously, go read his post). This actually makes it easy to use to test the sort.

Try this out for size:

List strings = 'this is a list of strings'.split()
Collections.sort(strings)
strings.inject('') { prev, curr ->
    println "prev: $prev, curr: $curr"
    assert prev <= curr
    curr  // value of 'prev' during next iteration
}

The result is:

prev: , curr: a
prev: a, curr: is
prev: is, curr: list
prev: list, curr: of
prev: of, curr: strings
prev: strings, curr: this

Since the closure returns the current value, that becomes the value of prev during the next iteration.

Actually, this can be simplified too. As of Groovy 1.8, there’s now an inject method that leaves out the initialization value. When you call it, the first two elements of the collection become the first two arguments of inject. In other words, now I can do this:

List strings = 'this is a list of strings'.split()
Collections.sort(strings)
strings.inject { prev, curr ->
    println "prev: $prev, curr: $curr"
    assert prev <= curr
    curr  // value of 'prev' during next iteration
}

The output now is:

prev: a, curr: is
prev: is, curr: list
prev: list, curr: of
prev: of, curr: strings
prev: strings, curr: this

Sweet. Now I have a real, live use case for inject. :) I promised you more, however. The title of this post refers to Elvis and spaceships, too.

Assume you want to sort the strings by length. In Java, when you can’t modify the class to be sorted (String), you use the two-argument sort method in Collections. The second argument is of type java.util.Comparator, which gives rise to the dreaded anonymous inner class monster:

List strings = 'this is a list of strings'.split()
Collections.sort(strings, new Comparator<String>()) { // R: Holy anonymous inner class, Batman!
    int compare(String s1, String s2) {               // B: Yes, Robin, with generics and everything.
        s1.size() <=> s2.size()                       // R: Gosh, gee, and a spaceship!
    }
})
assert strings == ['a', 'is', 'of', 'this', 'list', 'strings']
assert strings*.size() == [1, 2, 2, 4, 4, 7]          // R: Holy spread-dot operator, too!
                                                      // B: Dude, seriously, get a grip.

The spaceship operator returns -1, 0, or 1 when the left side is less than, equal to, or greater than the right side. It’s like a comparator, except the values are fixed to -1, 0, and 1.

The nice thing about the spread-dot operator here is that it doesn’t care whether the resulting strings are also alphabetical or not. The fact that equals length strings were also sorted alphabetically is just a side effect of the algorithm.

One of the common idioms in the Groovy JDK is to take static methods in Java and make them instance methods in Groovy. Here, the sort method is a static method in the Collections class. The Groovy JDK makes it an instance method in Collection (singular).

List strings = 'this is a list of strings'.split()
strings.sort { s1, s2 -> s1.size() <=> s2.size() }  // R: Holy closure coercion, Batman!
                                                    // B: No, the instance method takes a closure.
                                                    //    Seriously, did you remember your ADHD meds today?
assert strings*.size() == [1, 2, 2, 4, 4, 7]

The sort method in Groovy is now an instance method, which takes a one- or two-argument closure. The two-argument variety is implemented like a traditional comparator, meaning you return negative, zero, or positive as usual.

Even better, the one-argument version of sort says, in effect, transform each element into a number and Groovy will sort the numbers and use that as a way to sort the collection.

List strings = 'this is a list of strings'.split()
strings.sort { it.size() }                     // R: Everything is awesome!
assert strings*.size() == [1, 2, 2, 4, 4, 7]   // B: Shut up kid, or you'll find yourself floating home.

Here’s the best part. What if you want to sort by length, and then sort equal lengths reverse alphabetically? (I’ll use reverse alpha because the length sort also did alphabetical by accident).

Now I can use Elvis and spaceships together:

List strings = 'this is a list of strings'.split()
strings.sort { s1, s2 -> 
    s1.size() <=> s2.size() ?: s2 <=> s1
}

The Elvis operator says if the result is true by the Groovy Truth, use it, else use a default. Here it means do the length comparison and if the result is non-zero, we’re good. Otherwise do the (reverse) alphabetical comparison.

R: So that’s Elvis being carried back to his home planet by two tandem spaceships, right? Meaning it’s the fat Elvis from the 70s and not the thin Elvis from the 50s?
B: BAM! POW! OOF!

Here, finally, is a test based on inject for that sort:

strings.inject { prev, curr ->
    assert prev.size() <= curr.size()
    if (prev.size() == curr.size()) {
        assert prev >= curr
    }
    curr
}

There you have it: Elvis, spaceships, and inject all in a dozen lines of code. Now you add that to your Groovy utility belt.

(B: POW! BAM! OOF!)

The Reason The Internet Was Invented, or, Cat Pictures FTW

I’ll report about the SpringOne2GX conference soon, but I’m fighting a cold at the moment and the medication has put me in a rather bleary state. Thus, the following bit of random Groovy whimsy…

In this post, I’ll use Groovy to access a publicly available RESTful web service, parse the downloaded data, and build a GUI to hold the resulting images. The code will demonstrate the Groovy JDK, maps, the collect method, I/O processing, and the builder pattern, all to achieve its ultimate goal: looking at cat pictures.

A Flickring Light In The Darkness

There are many so-called RESTful web services available to the general public. Despite the name, however, most only support GET requests. This isn’t terribly surprising, since POST, PUT, and DELETE would require some kind of transaction, and security, and all sorts of protection against the sort of damage reminiscent of YouTube comments[1].

Examples of that sort of services include Open Weather Map (http://openweathermap.org/api), Currency Exchange Rates (http://openexchangerates.org), the Bing[2] Maps REST services (http://msdn.microsoft.com/en-us/library/ff701713.aspx), the Google Maps Geocoder (https://developers.google.com/maps/documentation/geocoding/) and the source for this article, Flickr (https://secure.flickr.com/services/developer). Each supports HTTP GET requests only.

If a RESTful web service only supports GET requests, does that make it a GETful service? If so, and it’s also stateless, does that make it a FORGETful web service? Thank you, thank you. Try the veal, and please remember to tip your wait staff.

Believe it or not, Flickr used to be the definitive site for image sharing, and some people still use it. Access requires a key, which means you need to register with Yahoo!, another web site that used to matter long, long ago when the web was shiny and new.

Why use it here? It’s available, it’s free, and, most important of all, it’s got cat pictures.

Feeling Groovy

Groovy is one of the new family of languages that compiles to bytecodes for the Java Virtual Machine. It’s simpler than Java, yet much more powerful, and has an easy learning curve for existing Java developers. It’s cool in its own right, and also forms the basis of interesting projects like Grails and Gradle.

Hey, I even wrote a Java/Groovy integration book about it.

Now on to the Groovy. Let’s assume you register and receive your special key. Save it in a file called flickr_key.txt, so you can access it this way:

String key = new File('flickr_key.txt').text

Accessing a property (like text here) in Groovy is the same as invoking the corresponding getter or setter method. Here, the Groovy JDK adds the getText method to File, which returns the contents of the file.

The base URL for the web service is:

String endPoint = 'https://api.flickr.com/services/rest?'

I included the “?” because I want to append the generated query string to the end point to get the complete URL.

The Flickr API uses a query string with a lot of parameters. Here I add them to a Groovy map.

def params = [method: 'flickr.photos.search', api_key: key,
    format: 'json', tags: 'cat', nojsoncallback: 1,
    media: 'photos', per_page: 6]

I’m calling the flickr.photos.search method with the api_key. I want JSON data back, but not as a JSON callback, and I want photo data using the cat tag. It’s not terribly complicated, but there are a lot of parameters.

The cool part is that a Groovy map like this can be converted to a query string using a very common idiom. The collect method applied to a map generates a list by applying a closure to each element of the original list. Therefore I can generate the query string thusly:

def qs = params.collect { k,v -> "$k=$v" }.join('&')

The two-argument closure takes each key/value pair and returns a string where the key equals the value. Then the join method applied to the list creates a string whose entries are separated by ampersands.

That means the whole URL is defined by "$endPoint$qs". I can then use the toURL method from the Groovy JDK to convert the string to a URL[3], and the text parameter invokes getText, returning the total response.

I can now use the cool JsonOutput.prettyPrint(txt) method to format the output and write it to a file.

File f = new File('cats.json')
if (f) f.delete()
f << jsonTxt
println JsonOutput.prettyPrint(jsonTxt)

Here’s a sample from that output:

{
    "photos": {
        "page": 1,
        "pages": 882777,
        "perpage": 6,
        "total": "5296661",
        "photo": [
            {
                "id": "15077663928",
                "owner": "125421155@N06",
                "secret": "dc605f2671",
                "server": "3837",
                "farm": 4,
                "title": "Mornings with bb kitty",
                "ispublic": 1,
                "isfriend": 0,
                "isfamily": 0
            },
            { ... },
            { ... },
            { ... },
            { ... },
            { ... }
        ]
    },
    "stat": "ok"
}

From JSON to Cat Pictures

Each photo block has the components necessary to assemble the URL to retrieve it, which is way more complicated than it needs to be but is still doable. Here’s the code for that part.

def json = new JsonSlurper().parseText(jsonTxt)
def urls = json.photos.photo.collect { p ->
    "http://farm${p.farm}.staticflickr.com/${p.server}/${p.id}_${p.secret}.jpg"
}

The JsonSlurper has a parseText method that converts the received JSON string into a Groovy map. Then walking the map by calling json.photos.photo gets all the photo objects, and the collect block converts each of them into a URL.

All that remains is to access all the photo URLs and populate a simple GUI with them. One simple way to do that is to use Groovy’s SwingBuilder class.

new SwingBuilder().edt {
    frame(title:'Cat pictures', visible: true, pack: true,
        defaultCloseOperation: WindowConstants.EXIT_ON_CLOSE,
        layout:new GridLayout(0, 2)) {
        urls[0..5].each { String url ->
            label(icon:new ImageIcon(url.toURL()))
        }
    }
}

The SwingBuilder has an edt method, which builds the UI on the event dispatch thread. The frame “method” instantiates a JFrame with the given title, makes it visible, and packs it as small as possible. The GUI uses a GridLayout to make two columns of equally-spaced elements, each of which is a JLabel that contains an ImageIcon with the individual photo URLs.

The result is a display of the most recent six pictures tagged cat in Flickr, which can result in almost anything[4].

Here’s a sample view:

Cat-pictures

Don’t ask me about the monkey. I have no idea what that’s about either.

So what have we learned?

  1. Groovy maps can be converted to query strings using a closure and the join method
  2. The Groovy JDK adds methods to Java library classes that you wished were there all along
  3. Groovy builders can make Swing programming almost, but not quite, pretty cool
  4. I can haz kitteh pics

The entire script from this article is part of a GitHub repository called IntroGroovy. Drill down to the src/main/groovy/flickr directory. You’ll need to add your own Flickr key, but after that, go for it. Just don’t be surprised, apparently, if you get monkeys or tractors mixed in with your cat pictures.

I suppose now I should go and reimplement the whole script in LOLCODE.


1. Okay, that’s a little harsh. Nothing could be as awful as YouTube comments. Or perhaps the correct response to that is, “I didn’t know YouTube supported comments.” If that’s your answer, I SO wish I was you.
2. Come on, stop laughing. I get it: “If you don’t remember what Bing is, just Google it”. Ha ha.
3. What did you expect toURL to convert the string into, a tomato? That would be pretty cool, actually. Now I wish I had that method. Of course, Groovy is an open source project…​
4. Usually it’s cats, but once I got a bunch of tractors, which confused me until I realized they were manufactured by Caterpillar.

SpringOne2GX, Day Minus 3

Yes, that’s a minus sign, because the event in question hasn’t happened yet. This is my first report from the upcoming SpringOne2GX conference, which starts in Dallas, TX next Monday, 9/8/14 (or 8/9/14 for my non-US-based friends). Since there are no talks on the opening day (just an opening keynote and reception), I’m arbitrarily designating that as Day Zero. I’m giving several talks at the conference and I thought I’d comment here on my preparations. I also have a kind of “insider” status, being a member of the No Fluff, Just Stuff conference tour (one of the organizers of SpringOne2GX*) and, as the author of Making Java Groovy, I know a lot of people in the community. That will hopefully give me opportunities to share amusing anecdotes and other non-presentation-based tidbits of information here.

*What is the proper hashtag for the conference? In past years people have recommended #springone2gx, but that’s way too many characters. The value #s12gx seems reasonable, but didn’t seem to catch on. Most people I know used #s2gx, which is probably what I’ll adopt.

Speaking of embarrassing anecdotes, let me give you one right away. Two years ago, after one of my talks at that year’s s2gx conference, Guillaume Laforge and Cedric Champeau came up to me to say hi. Guillaume, who lives near Paris, has an outrageous French accent. Every time I talk to him I keep thinking of this character from the classic movie. So I finally got up the nerve to ask him, “Just once, would you say to me, ‘your mother was a hamster, and your father smelt of elderberries!'”.

He and Cedric both just stared at me like I’d lost my mind. They had no idea what I was talking about. It wasn’t until later I realized that even if they knew Monty Python and the Holy Grail (a minimum requirement for any decent developer), they knew it in French.

Oh well.

This week I’m still polishing (yeah, right, polishing — not writing from scratch, at least as far as you know) my slides. I’m scheduled to give the following talks:

This is an introduction to Grails, combining domain classes (of course, Quest, Task, Knight, and Castle), controllers, services (using the Google geocoder), and more. Experienced Grails developers can comfortably skip it, but newbies will no doubt appreciate the introduction. I plan to add in some architectural discussions, as well as some Hibernate and/or Spring stuff if time allows.

I gave a similar talk a couple years ago, but things have changed significantly since then. As a minimal example, now the default unit testing framework is Spock, rather than the JUnit subclasses we used to use. Everything is annotation based, too. I’ll definitely talk about testing controllers, services, and domain classes (constraints), and plan to do a couple of integration tests as well, and then talk about various plugins, like the Build Test Data plugin and some other functional ones.

As I travel from company to company*, I rarely see the new REST capabilities in Grails 2.3+ being used. Last year Graeme Rocher did a wonderful presentation on them, as well as a brief discussion of the asynchronous capabilities in Grails. I hope to cover much the same (REST) ground, but rather than get into async I’m going to talk about customizing responses, using HAL for hypermedia, and so on.

*My day job is teaching technical training courses in pretty much all areas related to Java, specializing in open source topics like Spring, Hibernate/JPA, Android, Groovy, and Grails. That means I visit somewhere between 30 to 50 companies a year, and while I’m mostly talking to people who don’t know much about Grails yet, I’m always trying to find out what they are using. Between that and the NFJS conferences, I get a decent sense of what is becoming popular in the industry and what isn’t catching on.

As a sample, both Groovy and Grails are still growing at their normal slow, steady pace. Spring Boot is generating a lot of excitement among the more advanced developers, but most Spring users in the field haven’t even heard of it. I imagine this year’s conference will start to change that.

It turns out that Jeff Brown (co-author of DGG2, the Definitive Guide to Grails 2) is giving a talk entitled RESTful Grails 2 at 12:45pm on Tuesday. This is extraordinarily fortunate for me, because that means I can steal (er, reuse) whatever he says. Sweet. :)

I gave a similar talk last year, and was just getting ready when Paul King walked into the room. Paul King is one of the Greatest Groovy Experts Ever, and I experienced a flood of Imposter Syndrome as soon as he entered. Fortunately, he didn’t violently disagree with anything I said and was kind enough to share some additional information after it was over. Is it any wonder he is one of my personal heroes?

Another interesting Paul King tidbit: last time I checked, he had more commits to the Groovy code base than anyone, ever. Heck, I just checked and he did six more today.

I’ve updated the talk with new information, which should be fun. The toughest problem I’ll have is not giving away spoilers for the Groovy Puzzlers talk coming later in the conference.

(Go to the puzzlers talk. It’s fun.)

This is a big one for me. The fact that Android applications are now built using Gradle is a big deal for every Android developer, and the Android plugin for Gradle is still evolving at a rapid rate. The new IDE, Android Studio, is also changing all the time. My hope is to give the most up-to-date information on it possible. I’ll show the multi-project build they use by default, add some dependencies, run some tests (using the Robotium plugin), talk about Spring Android, and show some flavors and variants.

Best of all, I expect to try to update Android Studio right before the talk, so I have the latest possible version in the so-called Canary channel. What could possibly go wrong?

This is a revised and extended version of the same talk I gave at the Gr8 conf in Minneapolis in July. Those talks were limited to an hour, and given that lots of speakers were talking about Ratpack*, I decided to focus on Grails for the REST part. Now I should have time for both.

*Dan “Tiger” Woods is giving his Ratpack Web Framework talk at 2:30pm on Tuesday.

Wait, let’s try to improve that:

['Tiger','James','Ickey','Mirk','Sher'].each { nickname ->
    println "Dan '${nickname}' Woods"
}

Let me know if you have some better suggestions. Also, Lari Hotari is giving a talk on Ratpack and Grails 3.

Since there are no images in this post, here’s a small spoiler from the Groovy Vampires talk. Note the nice subtitle.

Making Java Groovy: Still a better love story than Twilight

If you’re coming to the conference, please drop by and say hi. I hope to write a short blog post every day of the conference, commenting on what talks I attended and which people I spoke to, but that’ll depend on how much last minute “updating” I’ll need to do on my slides.

Responses to “The Closure Of No Return”

I knew as soon as I wrote about implementing a simple prime number algorithm using Groovy that someone would find a more elegant way of solving the problem. In this post, I want to highlight some of the responses I received.

In my previous post, The Closure Of No Return, I discussed implementing an isPrime method that worked for relatively small integers. The algorithm is to try to divide the given number by all integers from two up to the square root of the number, rounded up.

My initial attempt, and the actual motivation for the blog post, was to demonstrate that if you use a return keyword inside a closure, you only return from the closure. So my initial implementation didn’t work:

// THIS DOESN'T WORK
boolean isPrime1(int x) {
    if (x == 2) return true
    int limit = Math.sqrt(x) + 1
    (2..limit).each { n ->
        // nice try, but a return in a closure
        // returns only from the closure
        if (x % n == 0) return false
    }
    return true
}

The each method takes a closure as an argument, which is like calling a completely separate method. The return statement inside the closure returns from that method, but not from the each loop itself.

In order to return the proper value, I introduced a local variable, called result, which I assigned inside the closure and then returned:

boolean isPrime2(int x) {
    if (x == 2) return true
    boolean result = true
    int limit = Math.sqrt(x) + 1
    (2..limit).each { n ->
        if (x % n == 0) {
            result = false
            // can't break out of the loop
        }
    }
    return result
}

Since I couldn’t use the break keyword inside the loop, this implementation was forced to check all the integers up to limit. I therefore switched at that point to Groovy’s for-in loop, which does allow the break:

boolean isPrime3(int x) {
    if (x == 2) return true
    boolean result = true
    int limit = Math.sqrt(x) + 1
    for (n in 2..limit) {
        if (x % n == 0) {
            result = false
            break
        }
    }
    return result
}

That was my final implementation in the blog post. I posted that one the web and waited for the corrections to roll in.

One commenter, Eric Johansson, pointed out that once I switched to the for-in loop, I could go back to using return again and eliminate the local variable result. Of course that’s true, and I missed it because of the steps I went through to write the code. I wonder how often such hysteresis loops occur in practice.

Another person commenting on the post, identified at Tim but without an email address or a link, suggested using the any method:

boolean isPrime3(int x) {
    if (x == 2) return true
    int limit = Math.sqrt(x) + 1
    !(2..limit).any { n ->
        x % n == 0
    }
}

The hope here is that the any method would stop at the first number that satisfied the closure. I wasn’t sure that was true, so I looked at the source code for Groovy, which can be found on GitHub. The any method is shown in the Groovy JDK as being part of the java.lang.Object class. One way methods are added to the Java standard library is through the metaprogramming methods in
org.codehaus.groovy.runtime.DefaultGroovyMethods. I looked in that class and found the any implementation, which adds the any method to the java.lang.Object class:

// from DefaultGroovyMethods
public static boolean any(Object self, Closure closure) {
    BooleanClosureWrapper bcw = new BooleanClosureWrapper(closure);
    for (Iterator iter = InvokerHelper.asIterator(self); iter.hasNext();) {
        if (bcw.call(iter.next())) return true;
    }
    return false;
}

If I’m interpreting this correctly, the iterator walks through each element one by one and returns true the first time an element satisfies the closure. In other words, yes, this approach would work and leave the loop at the proper time.

Soren (sorry, I don’t know how to put the line through the “o”) Berg Glasius (http://twitter.com/sbglasius) suggested using the every method instead, which would also probably do the job.

Both great developer Jochen Theodorou and the indefatigable Tim Yates immediately identified the obvious alternative that I missed. Here’s Tim’s solution (which is a slight variation on Jochen’s):

boolean isPrime5(int x) {
    int limit = Math.sqrt(x) + 1
    x == 2 || !(2..limit).find { n -> x % n == 0 }
}

I remembered the findAll method, which I used in my tests, but I forgot all about the find method, which returns the first element that satisfies the closure. That’s pretty ironic, too, because (in Grails especially) I normally accidentally use find when I meant findAll.

(As an aside, as soon as I saw Tim’s name, I assumed his solution was going to be both elegant and have the take method in there somewhere. I was half right.)

The only problem I have with these solutions is that while they work, they invalidate the real reason for my blog post, which was to emphasize that a return inside a closure returns only from the closure. Still, that last one feels like a really good way to solve the actual problem, so I’m going to use that in the future.

The best part is that everyone who suggested an alternative was quite friendly about it. I may have felt a bit foolish for missing these alternatives, but that’s my own pressure on myself. One of the best features about the Groovy community is how helpful and easy to work with everyone is, which I really like. Hopefully by posting those solutions here, I’m paying some of that forward.

(Thanks to everyone who responded, whether I quoted them here or not. I really appreciate the feedback.)

The Closure Of No Return

Even in a language like Groovy that is normally so clean and intuitive, there are traps for the unwary. I fell into one again today (in front of a room full of students), and I think it’s high time I documented it, at least so I’ll remember it for next time.

I’m teaching a Groovy / Grails class this week, and one of the problems I posed to the students was to write an isPrime method that worked for integers less than, say, 1000. The easiest brute force way to solve the problem is to divide the given number by every integer from 2 up to the square root of the number, rounded up.

Here’s a first try:

// NOTE: THIS DOESN'T WORK
boolean isPrime1(int x) {
    if (x == 2) return true
    int limit = Math.sqrt(x) + 1
    (2..limit).each { n ->
        if (x % n == 0) return false // Danger, Will Robinson!
    }
    return true
}

assert (2..20).findAll { isPrime1(it) } == (2..20) // Wait, what??

In other words, the method returns true for every number, whether it’s prime or not.

The problem is the return statement inside the each loop. The argument to the each method is a closure, and returning from a closure returns only from the closure, not the whole method.

I should emphasize that, so maybe next time I’ll remember it:

A return inside a closure returns only from the closure, not from the method that called it.

It’s actually not that hard to understand, when I think it through. Calling the closure is like calling another method. When I use the return keyword inside the closure, it’s like returning from that method, so it only returns from there and not from the each method itself.

So, how do I fix this? One way is to define a boolean variable outside the closure and then set it inside, but that leads to a minor problem as well.

boolean isPrime2(int x) {
    if (x == 2) return true
    boolean result = true // local var to set in closure
    int limit = Math.sqrt(x) + 1
    (2..limit).each { n ->
        if (x % n == 0) {
            result = false
            // don't you wish you could break here?
        }
    }
    return result
}

assert (2..20).findAll { isPrime2(it) } == 
    [2, 3, 5, 7, 11, 13, 17, 19] // works, but no break

I defined a boolean variable called result and initialized it to true. Then, inside the closure, if a number is not prime, I set the variable to false, and returned the variable at the end.

As the comment shows, however, I can’t break out of the loop. You can only use break inside loops, and some reason (I don’t know why), each loops don’t qualify. I’m forced to follow the loop to the end.

There is a way to solve that problem, too. I just need to replace the each loop with Groovy’s for-in loop.

boolean isPrime3(int x) {
    if (x == 2) return true
    boolean result = true
    int limit = Math.sqrt(x) + 1
    for (n in 2..limit) {
        if (x % n == 0) {
            result = false
            break  // yay!
        }
    }
    return result
}

assert (2..20).findAll { isPrime3(it)} == 
    [2, 3, 5, 7, 11, 13, 17, 19] // works

Now it works. This is also one reason to sometimes favor the for-in loop over the each iterator, but usually they’re interchangeable.

As usual, I couldn’t leave well enough alone. A common idiom in the Groovy JDK is to use metaprogramming to replace a static method in one class with an instance method in another. For example, java.lang.Math has the static abs method, but the Groovy JDK adds abs as an instance method to java.lang.Number. It’s not hard to do the same thing here:

Number.metaClass.isPrime = { ->
    Integer x = delegate as Integer
    if (x == 2) return true
    boolean result = true
    int limit = Math.sqrt(x) + 1
    for (n in 2..limit) {
        if (x % n == 0) {
            result = false
            break
        }
    }
    return result
}

assert (2..20).findAll { it.isPrime() } == [2, 3, 5, 7, 11, 13, 17, 19]

The delegate of the closure is the number the isPrime method was called on, so if I assign it to x I can copy and paste my implementation from above.

The truly fun part was that I assigned this problem to the students, and then decided to live code the implementation in front of them. Of course I went for the each implementation, falling right into the trap. Since I’ve seen this problem many times, I recognized it pretty quickly and therefore wrote the other two implementations after that. I’d like to believe that the students got some value out of seeing me mess it up and then fix it. If nothing else, they learned the value of test cases. Without those assert statements, I probably wouldn’t have noticed the error in the first place.

Baruch Sadogursky (@jbaruch on Twitter) is doing a presentation on Groovy and Grails Puzzlers at Gr8conf in Minneapolis at the end of July, and he’s always interested in issues like this, so I made sure to send it along to him. If you know of any others, I’m sure he’d appreciate them as well.

Groovy Groundhogs, revisited

For those people in the U.S. who are concerned that this morning Punxsutawney Phil, the Seer of Seers, Prognosticator of Prognosticators, emerged, reluctantly, but alertly, and stated in groundhog-ese, “I definitely see a shadow”, let me allay your fears using Groovy:

Calendar cal = Calendar.instance

cal.set(year:2014, month:Calendar.FEBRUARY, date:2)
def groundhogDay = cal.time

cal.set(year:2014, month:Calendar.MARCH, date:20)
def firstDayOfSpring = cal.time

int days = firstDayOfSpring - groundhogDay
println """There are $days days between Groundhog Day and the first Day of Spring.
That's ${(int) (days/7)} weeks and ${days % 7} days.
"""

The result of that script is:

There are 46 days between Groundhog Day and the first Day of Spring.
That's 6 weeks and 4 days.

In other words, “six more weeks of winter” is a good thing, because otherwise we’d have to wait an extra four days.

A few notes are in order:

  • Here is the official Groundhog Day site.
  • According to the Wikipedia page, Phil has been predicting the weather since 1887. His accuracy rate is only about 39%, however.
  • Some kind of Groundhog Day celebration has been recorded since as far back as 1841. It bears similarities to the pagan festival of Imbolc, among others.
  • If you don’t like using the Calendar class (and who does?), the Groovy JDK also adds a parse method directly to java.util.Date:
    Date groundhogDay = Date.parse('MM/dd/yyyy', '02/02/2014')
    Date firstDayOfSpring = Date.parse('MM/dd/yyyy', '03/20/2014')
    

    and the rest is the same.

  • The Vernal Equinox this year occurs on March 20, 2014, at 1:57pm EDT.
  • I considered posting this blog entry over and over as some sort of weak tribute to the movie, but at least I spared you that. :)

There’s more, but I have to go get ready for our Super Bowl party, where the announcers have been talking about the weather all week and will no doubt be disappointed that the temperature will be in the 40s (Fahrenheit).

Groovy Weather: A New Groovy Example at Java.net

One of the main goals of Making Java Groovy is to show Java developers how much Groovy can make their lives easier. To that end, I just published a blog post (through Manning’s account) over a Java.net entitled, Groovy Weather: POGOs, Gson, and Open Weather. The blog post comes with a coupon code for 45% off the book. :)

(Spoiler: it’s “kousenjn“, but if you’re going to use it at least drop by the blog link.)

Another spoiler: it’s freakin’ cold outside. That’s partly what my blog post is about — consuming JSON data from Open Weather Map and displaying it at the console. That’s not terribly difficult, but the real value added comes from using Google’s Gson parser to convert the JSON objects into Groovy.

If you’re new to Groovy, the blog post shows a lot of detail. Regular readers of my blog, however, probably are at least comfortable with the language, so I thought I’d summarize the good parts here.

First, I need to show the JSON data, so I know what to map to.

(Er, “to what to map”? Ugh. Winston Churchill was once criticized for ending a sentence with a preposition. His reply was, “Madam, that is nonsense up with which I will not put.” That’s one of those possibly apocryphal stories that I don’t want to try to verify because I like it too much.)

Here’s the trivial Groovy expression to download the data:

import groovy.json.*

String url = 'http://api.openweathermap.org/data/2.5/weather?q=marlborough,ct'
String jsonTxt = url.toURL().text
println JsonOutput.prettyPrint(jsonTxt)

I’m using the static prettyPrint method of JsonOutput to make viewing the results easier. As a minor complaint, I have to mention that every time I use prettyPrint, I’m surprised it doesn’t actually print. It just formats the JSON data. I still need to print the result. That seems rather misleading to me.

Anyway, here’s the result:

{
    "coord": {
        "lon": -72.46,
        "lat": 41.63
    },
    "sys": {
        "message": 0.193,
        "country": "United States of America",
        "sunrise": 1389096985,
        "sunset": 1389130583
    },
    "weather": [
        {
            "id": 800,
            "main": "Clear",
            "description": "Sky is Clear",
            "icon": "01d"
        }
    ],
    "base": "gdps stations",
    "main": {
        "temp": 260.41,
        "humidity": 33,
        "pressure": 1025,
        "temp_min": 258.71,
        "temp_max": 262.15
    },
    "wind": {
        "speed": 1.54,
        "deg": 0
    },
    "clouds": {
        "all": 0
    },
    "dt": 1389130569,
    "id": 4835003,
    "name": "Hartford",
    "cod": 200
}

The current temperature is buried inside the object, in the “temp” property of the “main” subobject. I could just parse this using a JsonSlurper, but instead I decided to map the whole thing using Gson.

Gson wants a class structure that maps to the JSON hierarchy. Here’s mine, which I just stuffed into a single Groovy class called Model.groovy.

class Model {
    Long dt
    Long id
    String name
    Integer cod

    Coordinates coord
    Main main
    System sys
    Wind wind
    Clouds clouds
    Weather[] weather
}

class Main {
    BigDecimal temp
    BigDecimal humidity
    BigDecimal pressure
    BigDecimal temp_min
    BigDecimal temp_max
}

class Coordinates {
    BigDecimal lat
    BigDecimal lon

    String toString() { "($lat, $lon)" }
}

class Weather {
    Integer id
    String main
    String description
    String icon
}

class System {
    String message
    String country
    Long sunrise
    Long sunset
}

class Wind {
    BigDecimal speed
    BigDecimal deg
}

class Clouds {
    BigDecimal all
}

I added the data types based on reading the docs, which are pretty thin, and following the nested structure of the JSON data. The names of the classes aren’t important. It’s the property names that have to match the keys in the JSON maps for deserialization to work.

Using Gson is almost trivial. All I need is the fromJson method in the Gson class:

import groovy.transform.*
import com.google.gson.Gson

@ToString(includeNames=true)
class Model {
  ...
}

String url = 'http://api.openweathermap.org/data/2.5/weather?q=marlborough,ct'
String jsonTxt = url.toURL().text
Gson gson = new Gson()
println gson.fromJson(jsonTxt, Model)

Everything is converted to the Groovy class structure, just as expected.

Before printing the results, though, I need to do some data manipulation. If you looked at the current temperature value, you might have noticed it’s in Kelvin, of all things. As a US-based developer, I need to convert that to Fahrenheit.

def convertTemp(temp) {
    9 * (temp - 273.15) / 5 + 32
}

The time fields are based on “Unix time”, which measures seconds in the current epoch (beginning January 1, 1970 GMT). Java’s Date class has a constructor that takes a long representing milliseconds from the beginning of the epoch.

def convertTime(t) {
    new Date(t*1000)  // Unix time in sec, Java time in ms
}

Finally, the wind speed is in “meters per second” and I want “miles per hour”. When I was in high school (and dinosaurs roamed the Earth), I learned about the Factor Label method, which meant I could memorize a single length conversion and calculate any other.

def convertSpeed(mps) {
    // 1 m/sec * 60 sec/min * 60 min/hr * 100 cm/m * 1 in/2.54 cm * 1 ft/12 in * 1 mi/5280 ft
    mps * 60 * 60 * 100 / 2.54 / 12 / 5280
}

I added all of these to the Model class, and some getters to use them.

class Model {
  // ... as before ...

  def getTime() { convertTime dt }
  def getTemperature() { convertTemp main.temp }
  def getLow() { Math.floor(convertTemp(main.temp_min)) }
  def getHigh() { Math.ceil(convertTemp(main.temp_max)) }
  def getSunrise() { convertTime sys.sunrise }
  def getSunset() { convertTime sys.sunset }
  def getSpeed() { convertSpeed wind.speed }
}

Finally, here’s my nice, formatted toString method to print the results (also a part the Model class):

String toString() {
    """
    Name         : $name
    Time         : $time
    Location     : $coord
    Weather      : ${weather[0].main} (${weather[0].description})
    Icon         : http://openweathermap.org/img/w/${weather[0].icon}.png
    Current Temp : $temperature F (high: $high F, low: $low F)
    Humidity     : ${main.humidity}%
    Sunrise      : $sunrise
    Sunset       : $sunset
    Wind         : $speed mph at ${wind.deg} deg
    Cloudiness   : ${clouds.all}%
    """
}

I should mention that the Weather attribute of the Model class is a collection. Presumably that’s for when there are multiple weather stations associated with a given location. In the source code repository (linked below), I used Groovy’s each method to iterate over them all. Here I’m just using the first one.

The driver for the system is:

import com.google.gson.Gson

class OpenWeather {
    String base = 'http://api.openweathermap.org/data/2.5/weather?q='
    Gson gson = new Gson()

    String getWeather(city='Marlborough', state='CT') {
        String jsonTxt = "$base$city,$state".toURL().text
        gson.fromJson(jsonTxt, Model).toString()
    }
}

I like Groovy’s default arguments for methods. If I invoke the getWeather method without arguments (or as the weather property in the usual idiom), then the result is for “Marlborough, CT”. Otherwise I supply a city and a state and they’re used instead.

Clearly this needs to be tested, or at least the converters so. Here’s a Spock test for them:

import spock.lang.Specification

class ModelSpec extends Specification {
    Model model = new Model()

    def 'convertTemp converts from Kelvin to F'() {
        expect:
        32 == model.convertTemp(273.15)
        212 == model.convertTemp(373.15)
    }

    def 'convertSpeed converts from meters/sec to miles/hour'() {
        expect:
        (2.23694 - model.convertSpeed(1)).abs() < 0.00001
    }

    def 'convertTime converts from Unix time to java.util.Date'() {
        given:
        Calendar cal = Calendar.instance
        cal.set(1992, Calendar.MAY, 5)
        Date d = cal.time
        long time = d.time / 1000  // Java time in ms, Unix time in sec

        when:
        Date date = model.convertTime(time)

        then:
        d - date < 1
    }
}

The mps to mph value I got from putting “1 meter per second in mph” into Google, which gave me the proper result.

Just to make sure the calls are working properly, here are a couple of tests for the Model class.

import spock.lang.Specification

class OpenWeatherSpec extends Specification {
    OpenWeather ow = new OpenWeather()

    def 'default city and state return weather string'() {
        when:
        String result = ow.weather
        println result

        then:
        result  // not null is true in Groovy
        result.contains('41.63')
        result.contains('-72.46')
    }

    def "The weather is always great in Honolulu"() {
        when:
        String result = ow.getWeather('Honolulu', 'HI')
        println result

        then:
        result
        result.contains('21.3')
        result.contains('-157.86')
    }
}

Here’s a script to run the whole system:

OpenWeather ow = new OpenWeather()
println ow.weather  // called Marlborough, CT, but really Hartford

// Home of Paul King, co-author of _Groovy in Action_ and my personal hero
println ow.getWeather('Brisbane','Australia')

// Home of Guillaume Laforge, head of the Groovy project
// (also one of my heroes, along with Dierk Koenig, Graeme Rocher, Tom Brady, David Ortiz, ...)
println ow.getWeather('Paris','France')

// Have to check the weather in Java, right?
println ow.getWeather('Java','Indonesia')

// Any weather stations in Antarctica?
println ow.getWeather('', 'Antarctica')

Here are the results at the moment (Jan 7, 2014, at about 5:45pm EST):

Name         : Marlborough
Time         : Tue Jan 07 17:43:10 EST 2014
Location     : (41.63, -72.46)
Weather      : Clear (Sky is Clear)
Icon         : http://openweathermap.org/img/w/01n.png
Current Temp : 8.312 F (high: 11.0 F, low: 5.0 F)
Humidity     : 35%
Sunrise      : Tue Jan 07 07:16:25 EST 2014
Sunset       : Tue Jan 07 16:36:23 EST 2014
Wind         : 3.4448818898 mph at 258 deg
Cloudiness   : 0%

Name         : Brisbane
Time         : Tue Jan 07 17:36:03 EST 2014
Location     : (-27.47, 153.02)
Weather      : Clouds (broken clouds)
Icon         : http://openweathermap.org/img/w/04n.png
Current Temp : 78.566 F (high: 82.0 F, low: 77.0 F)
Humidity     : 46%
Sunrise      : Mon Jan 06 14:00:36 EST 2014
Sunset       : Tue Jan 07 03:47:48 EST 2014
Wind         : 10.51360057266 mph at 121 deg
Cloudiness   : 80%

Name         : Paris
Time         : Tue Jan 07 17:39:23 EST 2014
Location     : (48.85, 2.35)
Weather      : Clear (Sky is Clear)
Icon         : http://openweathermap.org/img/w/01n.png
Current Temp : 52.682 F (high: 54.0 F, low: 51.0 F)
Humidity     : 81%
Sunrise      : Tue Jan 07 02:42:36 EST 2014
Sunset       : Tue Jan 07 11:11:33 EST 2014
Wind         : 8.052970651396 mph at 220 deg
Cloudiness   : 0%

Name         : Batununggal
Time         : Tue Jan 07 17:43:35 EST 2014
Location     : (-6.96, 107.65)
Weather      : Clouds (scattered clouds)
Icon         : http://openweathermap.org/img/w/03n.png
Current Temp : 68.5904 F (high: 69.0 F, low: 68.0 F)
Humidity     : 91%
Sunrise      : Mon Jan 06 17:40:32 EST 2014
Sunset       : Tue Jan 07 06:10:58 EST 2014
Wind         : 2.9303865426 mph at 200 deg
Cloudiness   : 44%

Name         :
Time         : Tue Jan 07 17:43:35 EST 2014
Location     : (-78.33, 20.61)
Weather      : Clear (Sky is Clear)
Icon         : http://openweathermap.org/img/w/01d.png
Current Temp : -20.2936 F (high: -20.0 F, low: -21.0 F)
Humidity     : 38%
Sunrise      : Tue Jan 07 16:43:35 EST 2014
Sunset       : Wed Jan 08 04:43:35 EST 2014
Wind         : 15.770400858984 mph at 15.0058 deg
Cloudiness   : 0%

So, yeah, it’s cold outside. I also freely admit it’s a little weird seeing those non-US temperatures converted to Fahrenheit.

The Gradle build file for this system is almost trivial:

apply plugin:'groovy'

repositories {
    mavenCentral()
}

dependencies {
    compile 'org.codehaus.groovy:groovy-all:2.2.1'
    compile 'com.google.code.gson:gson:2.2.4'
    testCompile 'org.spockframework:spock-core:0.7-groovy-2.0'
}

All of this code can be found in the book’s Github repo. I added this to Chapter 2: Groovy by Example. Chapter 2 now uses a multi-project Gradle build, so it doesn’t look exactly that the one shown here, but it works the same way.

This process is easy enough to replicate for any RESTful service that returns JSON data. The time consuming part, if any, is writing the POGO structure to capture the JSON tree, but at least with POGOs the result is pretty simple.

————-

Now for a few personal notes. First, the latest Amazon review is simply beautiful. It’s from a person who knew Java and didn’t expect much from Groovy, but found out how much it could help. That’s exactly what I’m trying to achieve. In fact, I’ve been amazed that every single Amazon reviewer, no matter how much he or she did or didn’t like specific aspects of the book, definitely understood the goal and appreciated it.

Second, I have a few new projects in the works, but I think I can mention one of them. I’m currently generating a series of screencasts for the book, which I’m calling “Making Java Groovy: The Director’s Cut”. The goal is to discuss why the book was written as it was — what was included, what was left out, what has changed since its publication in the Fall of 2013, and anything else I think might be interesting or helpful. Everyone at Manning has been very supportive of the effort so far. I hope to make at least the first set of screencasts available soon.

Third, I’ll be giving a talk at the Boston Grails Users’ Group on Thursday, January 9, at 7pm. I hope to bring some books to add to the raffle, along with an admission to the NFJS event in the Boston area in March. If you’re in the area, please drop by and say hello.

Fourth, if you’ve read this far, you deserve a reward! If you use the coupon code “kousen37” at Manning.com, you can get 37% off any book at Manning. :)

Oh, and in case you didn’t hear me celebrating, I’m now officially a 2013 JavaOne Rock Star. Thank you very much to everyone who recommended me.

Finally, those of you battling the so-called “polar vortex”, stay warm! It’s cold here, but at least we’re still above 0 (F, not C). Next week I’m planning to brave the winds of Caradhras, otherwise known as central Minnesota. Brr.

As the saying goes, follow the great advice found on the side of a mayonnaise jar: “Keep cool but don’t freeze”.

Follow

Get every new post delivered to your Inbox.

Join 1,372 other followers

%d bloggers like this: