Java 8 Streams and JPA

Since the definition of the JPA Standard predates the release of Java 8 it is not surprising that the JPA API is only based on collections. There is no way to obtain a Java 8 Stream out of a query object.

For those out there using Hibernate as their JPA provider there is an interesting way to create a stream for your JPQL queries.

Hibernate Scrollable Results

Hibernate supports many more features than those offered by the JPA Standard only. Hibernate’s Query class allows us to obtain an iterator-like object for a JQPL query. This iterator-like class is named ScrollableResults.

You can see a great example of this feature in this blog post about Reading Large Result Sets with Hibernate and MySQL. I have taken the liberty of copying one of their examples:

Query query = session.createQuery(query);
query.setReadOnly(true);
query.setFetchSize(100);
ScrollableResults results = query.scroll(ScrollMode.FORWARD_ONLY);
while (results.next()) {
   Object row = results.get();
   // process the entity here
}
results.close();

Creating an Iterator Wrapper around Scrollable Results

Let’s take this a bit further now by implementing a Java Iterator wrapper around the Hibernate’s scrollable results object as shown below:

class ScrollableResultIterator<T> implements Iterator<T> {
   private final ScrollableResults results;
   private final Class<T> type;

   ScrollableResultIterator(ScrollableResults results, Class<T> type) {
      this.results = results;
      this.type = type;
   }

   @Override
   public boolean hasNext() {
      return results.next();
   }

   @Override
   public T next() {
      return type.cast(results.get(0));
   }
}

Gaining Access to Hibernate Session through JPA Entity Manager

As you can see the examples above use a Hibernate session object to obtain an instance of a Hibernate query object. Supposing we are using JPA, we typically have access to an EntityManager, but not to any particular implementation objects of the underlaying provider.

To gain access to the Hibernate’s Session object we can use a special method in the entity manger.

Session session = entityManager.unwrap(Session.class);

Be warned that at this point we are escaping from the safety of vendor agnostic code. In other words, if we ever wanted to use another provider like OpenJPA or EclipseLink, we would be forced to find a different alternative to implement our code here since those other providers won’t offer anything like Hibernate’s Session and ScrollableResults.

From Iterators to Spliterators to Streams

Our next step consists in obtaining a Java 8 stream out of this iterator. Lukas Eder recently wrote an interesting article about the difficulties to obtain a stream out of an iterable object which will be pretty handy here.

To achieve our goal we now need to use two utility classes in Java 8 named StreamSupport and Spliterators.

For instance, we could take an iterator, like the one we defined above and transform it into a Spliterator by doing:

private Spliterator<T> toSplitIterator(ScrollableResults scroll, Class<T> type){
   return Spliterators.spliteratorUnknownSize(
      new ScrollableResultIterator<>(scroll, type),
         Spliterator.DISTINCT | Spliterator.NONNULL | 
         Spliterator.CONCURRENT | Spliterator.IMMUTABLE
      );
}

This Splititerator is an intermediate product that we can now use to create a Java 8 stream out of it:

StreamSupport.stream(spliterator, false);

At this point we have built enough to create our first stream out of a JPQL query. We can do it as follows:

public Stream<T> getResultStream(
      String sql, 
      Integer fetchSize, 
      Map<String,Object> parameters) 
{
   Query query = session.createQuery(sql);
   if (fetchSize != null) {
      query.setFetchSize(fetchSize);
   }
   for (Map.Entry<String, Object> parameter : parameters.entrySet()) {
      query.setParameter(parameter.getKey(), parameter.getValue());
   }
   query.setReadOnly(true);
   ScrollableResults scroll = query.scroll(ScrollMode.FORWARD_ONLY);
   return StreamSupport.stream(toSplitIterator(scroll, type), false)
            .onClose(scroll::close);
}

A JPA Stream API

We can do a little bit better by defining the basics of the type of query object we are currently missing in JPA Standard API:

public interface StreamQuery<T> {
   Stream<T> getResultStream();
   StreamQuery<T> setParameter(String name, Object value);
   StreamQuery<T> setFetchSize(int fetchSize);
}

And putting together all we have learned so far we could create this implementation of our new StreamQuery interface:

public class HibernateQueryStream<T> implements StreamQuery<T> {

   private final Session session;
   private final String sql;
   private final Class<T> type;
   private final Map<String, Object> parameters = new HashMap<>();
   private Integer fetchSize;

   public HibernateQueryStream(
      EntityManager entityManager, 
      String sql, 
      Class<T> type) 
   {
     this.session = entityManager.unwrap(Session.class);
      this.sql = sql;
      this.type = type;
   }

   @Override
   public StreamQuery<T> setParameter(String name, Object value) {
      parameters.put(name, value);
      return this;
   }
   
   @Override
   public StreamQuery<T> setFetchSize(int fetchSize) {
      this.fetchSize = fetchSize;
      return this;
   }

   @Override
   public Stream<T> getResultStream() {
      Query query = session.createQuery(sql);
      if (fetchSize != null) {
         query.setFetchSize(fetchSize);
      }
      query.setReadOnly(true);
      for (Map.Entry<String, Object> parameter : parameters.entrySet()) {
         query.setParameter(parameter.getKey(), parameter.getValue());
      }
      ScrollableResults scroll = query.scroll(ScrollMode.FORWARD_ONLY);
      return StreamSupport.stream(toSplitIterator(scroll, type), false)
               .onClose(scroll::close);
   }
   
   private Spliterator<T> toSplitIterator(ScrollableResults scroll, Class<T> type){
      return Spliterators.spliteratorUnknownSize(
         new ScrollableResultIterator<>(scroll, type),
            Spliterator.DISTINCT | Spliterator.NONNULL | 
            Spliterator.CONCURRENT | Spliterator.IMMUTABLE
      );
   }

   private static class ScrollableResultIterator<T> implements Iterator<T> {

      private final ScrollableResults results;
      private final Class<T> type;
      
      ScrollableResultIterator(ScrollableResults results, Class<T> type) {
         this.results = results;
         this.type = type;
      }
      
      @Override
      public boolean hasNext() {
         return results.next();
      }
      
      @Override
      public T next() {
         return type.cast(results.get(0));
      }
   }
}

The Repository Layer

On top of this component we can now build our repository layer.

@Repository
public class StreamRepository {

   @PersistenceContext(unitName="demo")
   private EntityManager entityManager;

   public Stream<Order> getOrderHistory(String email) {
      String jpql = "SELECT o FROM Order o WHERE o.customer.email=:email";
      StreamQuery<Order> query = new HibernateQueryStream<>(
          entityManager, 
          jpql, 
          Order.class
      );
      return query.getResultStream();
   }
}

And from here the rest is a piece of cake:

orderRepository.findOrderHistory(email)
   .filter(order -> order.total() > 100)
   .map(Order::getOrderLines)
   .flatMap(List::stream)
   .map(OrderLine::getTotal)
   .reduce(0, (x,y) -> x + y);

It is important to highlight that the Hibernate session must be alive by the time we actually consume the stream, because it is at that point that the entities will be mapped. So, make sure that wherever you consume the stream the Hibernate session or you entity manager context is still alive.

Further Reading

Functional Programming with Java 8 Functions

Edwin Dalorzo I have finally had the opportunity to work in a couple of projects being entirely developed with Java 8 and during the past months I’ve experienced, first hand, many of the new functional programming features in the language. So I decided to write a series of articles about some of these things I’ve being learning and how some of the well known functional programming constructs can be translated into Java 8.

Preliminaries

Ok, let’s start with something simple. The following is a lambda expression (i.e. an anonymous function) that takes an argument x and increments it by one. In other words a function that receives, apparently an integer, and returns a new incremented integer:

x -> x + 1

And what is the type of this function in Java?

Well, the answer is that it depends. In Java the same lambda expression could be bound to variables of different types. For instance, the following two are valid declarations in Java:

Function<Integer,Integer> add1 = x -> x + 1;
Function<String,String> concat1 = x -> x + 1;

The first one increments an integer x by one, whereas the second one concatenates the integer 1 to any string x.

And how can we invoke these functions?

Well, now that they are bound to a reference we can treat them pretty much like we treat any object:

Integer two = add1.apply(1); //yields 2
String answer = concat1.apply("0 + 1 = "); //yields "0 + 1 = 1"

So, as you can see every function has a method apply that we use to invoke it and pass it an argument.

And what if I already have a method that does that, can I use it as a function?

Yes, since an alternative way to create functions is by using methods we had already defined and that are compatible with our function definition.

Suppose that we have the following class definition with methods as defined below:

public class Utils {
   public static Integer add1(Integer x) { return x + 1; }
   public static String concat1(String x) { return x + 1; }
}

As you can see the methods in this class are compatible with our original function definitions, so we could use them as “method references” to create the same functions we did before with lambda expressions.

Function<Integer,Integer> add1 = Utils::add1;
Function<String,String> concat1 = Utils::concat1;

These two are just the same thing as the ones we did before.

High Order Programming

The cool thing about functions is that they encapsulate behavior, and now we can take a piece of code, put it into a function and pass it around to other functions or methods for them to use it. This type of functions that operate on (or produce new) functions are typically called high order functions and the programming style based on exploiting this powerful feature is called, unsurprisingly, high order programming.

About Functions that Create Functions

Let’s see a couple of examples of how we can do this using function objects. Let’s consider the following example:

Function<Integer, Function<Integer,Integer>> makeAdder = x -> y -> x + y;

Above we have a function called makeAdder that takes an integer x and creates a new function that takes an integer y and when invoked adds x to y. We can tell this is a high order function because it produces a new function.

Now, we use this to create our original add1 function:

Function<Integer,Integer> add1 = makeAdder.apply(1);
Function<Integer,Integer> add2 = makeAdder.apply(2);
Function<Integer,Integer> add3 = makeAdder.apply(3);

With our new high order function, however, we could also create add2, add3, ..., addn, right?

Can’t we define this in a simpler way as we did before with the Utils class methods?

Yes, we can. Consider the following addition to the Utils class:

public class Utils {
    public static Function<Intger, Integer> adder(Integer x) {
       return y -> x + y;
    }
}

This signature is a bit simpler to read than that in our lambda expression, but as you can see it is pretty much the same thing. The function continues to receive the same number and type of arguments and continues to return the same type of result.

We can now use this simpler function factory to create our makeAdder and add1 functions again:

Function<Integer, Function<Integer,Integer>> makeAdder = Utils::adder;
Function<Integer,Integer> add1 = makeAdder.apply(1);

And there you have it, this is exactly the same thing as before.

About Functions that Receive Functions as Arguments

Let’s suppose we had the following two functions defined:

Function<Integer,Integer> add1 = x -> x + 1;
Function<Integer,Integer> mul3 = x -> x * 3;

Now, naively, we could invoke this two functions together to increment and multiply a number by 3, right?. Like this:

Integer x = 10;
Integer res = mul3.apply(add1.apply(x)); //yields 33

But what if we created a function that did both things instead?

Consider the following pseudocode:

(f,g) -> x -> g( f(x) )

This would be a function that takes two other unary functions and creates yet another function that applies the original two in certain order. This is a classical example of what is called function composition.

In some languages there is even an operator to compose two functions in this way:

h = f o g

Where o would be a binary operator that would compose functions f and g pretty much as we did in pseudocode above and produce a new function h.

How can we do function composition in Java?

I can think of two ways to do this in Java, one more difficult than the other. Let’s start with the more difficult first, because that will let us appreciate the value of the simpler solution later on.

Function Composition Strategy 1

First, we must start by realizing that the function in pseudocode above is a binary function (i.e. a function that receives two arguments). But all our examples so far have dealt only with unary functions.

It would seem this is not important, but in Java it is, since functions of different arities have different target functional interfaces. In Java, a function that receives two arguments is called BinaryOperator.

For instance, using a BinaryOperator we could implement a sum operator:

BinaryOperator<Integer> sum = (a,b) -> a + b;
Integer res = sum.apply(1,2); // yields 3

Well, just as easily we could implement the compose operator, right? Only that in this case, instead of two simple integers, we receive two unary functions:

BinaryOperator<Function<Integer,Integer>> compose;
compose = (f,g) -> x -> g.apply(f.apply(x));

Now we can easily use this to fabricate a compound function that adds 1 and multiplies by 3, one after the other.

Function<Integer,Integer> h = compose.apply(add1,mul3);
Integer res = h.apply(10); //yields 33	

And now we can beautifully, and in really simple way, combine two unary integer functions.

Function Composition Strategy 2

Now, function composition is something so common that it would have been a mistake if the Java Expert Group would have not considered it in their API design, and so, to make our lives simpler, all Function objects have a method called compose that allows us to very easily compose two functions together.

The following code produces the exact same result as above:

Function<Integer,Integer> h = mul3.compose(add1);
Integer res = h.apply(10);	

Partial Function Application or Currying

In most functional programming languages it is possible to create partially applied functions. That is, if a function is receiving multiple arguments, we can partially invoke the function providing just a few arguments and receive a partially applied function out of it. This is typically called currying.

Although you have not noticed it, we have already covered that in this article, but now we are going to make it much more evident :-)

So, consider the following pseudo code

sum = x -> y -> x + y

Then we say that sum is a function that accepts one parameter x and fabricates another anonymous function that in turn accepts one parameter y that, when invoked, sums x and y.

In many functional programming languages a construct like this can be invoked as if this was just one simple function by doing:

sum 10 5 //yields 15

But the truth is that this is just syntactic sugar to do:

sum(10)(5) //yields 15

Since sum is a function that returns a function.

The beauty of this idiom is that now we could partially apply sum:

plus10 = sum 10

And now plus10 is a partially applied function of sum. Can you see now where we had already talked about this?

plus10(5) //yields 15

Can we do this with Java?

The truth is that we have already done it above, we just probably did not notice. Unfortunately, in Java we do not have the syntactic sugar that some other language have, and therefore this is a bit more verbose:

Function<Integer,Function<Integer,Integer>> sum = x -> y -> x + y;

Well, you can see sum is declared in a “currified” way. And now we can partially apply it:

Function<Integer, Integer> plus10 = sum.apply(10);
Integer res = plus10.apply(5); //yields 15

Unary, Binary, Ternary and n-ary Functions

So, as mentioned above, Java uses different functional interfaces for different function arities. And so Function<T,R> is a functional interface for any unary function where the domain and the codomain of the function may be of different types.

For instance, we could define a function that receives a string value and parses it as an integer:

Function<String,Integer> atoi = s -> Integer.valueOf(s);

But most of our examples above are for integer functions whose argument and return value are of this same type. For those cases we could alternatively use the UnaryOperator<T> instead. This is just a Function<T,T>.

Thus, some our declarations above could be slightly simplified with this functional interface:

UnaryOperator<Integer> add1 = n -> n + 1;
UnaryOperator<String> concat1 = s -> s + 1;
Function<Integer, UnaryOperator<Integer>> sum = x -> y -> x + y;
UnaryOperator<Integer> sum10 = sum.apply(10);

I have already written another article that explains Why Java 8 has Interface Pollution like this in case you are interested in an explanation.

Value Types and Primitive Type Functions?

Evidently using a type like Integer incurs into the costs of boxing and unboxing when our functions have to deal with a primitive type like int.

As you know, Java does not support value types as type arguments in generic declarations, so to deal with this problem we can use alternative functional interfaces like ToIntFunction, IntFunction or IntUnaryOperator.

Or we can define our own primitive function.

interface IntFx {
	public int apply(int value);
}

Then we can do:

IntFx add1 = n -> n + 1;
		
IntFunction<IntFx> sum = x -> y -> x + y;
IntFx sum10 = sum.apply(10);
sum10.apply(4); //yields 14

Similar functional interfaces can be found in the Java API for types double and long as well. This topic is also covered in the alternative article mentioned above about interface pollution.

And that’s it for the time being. I hope that in a future article I can take this a bit further and show some practical applications derived from my experience in actual projects.

Further Reading

Why There Is Interface Pollution in Java 8

Edwin DalorzoI was reading this interesting post about The Dark Side of Java 8. In it, Lukas Eder, the author, mentions how bad it is that in the JDK 8 the types are not simply called functions. For instance, in a language like C#, there is a set of predefined function types accepting any number of arguments with an optional return type (Func and Action each one going up to 16 parameters of different types T1, T2, T3, …, T16), but in the JDK 8 what we have is a set of different functional interfaces, with different names and different method names, and whose abstract methods represent a subset of well know function signatures/function arities (i.e. nullary, unary, binary, ternary, etc).

The Type Erasure Issue

So, in a way, both languages suffer from some form of interface pollution (or delegate pollution in C#). The only difference is that in C# they all have the same name. In Java, unfortunately, due to type erasure, there is no difference between Function<T1,T2> and Function<T1,T2,T3> or Function<T1,T2,T3,...Tn>, so evidently, we couldn’t simply name them all the same way and we had to come up with creative names for all possible types of function combinations.

Don’t think the expert group did not struggle with this problem. In the words of Brian Goetz in the lambda mailing list:

[…] As a single example, let’s take function types. The lambda strawman offered at devoxx had function types. I insisted we remove them, and this made me unpopular. But my objection to function types was not that I don’t like function types — I love function types — but that function types fought badly with an existing aspect of the Java type system, erasure. Erased function types are the worst of both worlds. So we removed this from the design.

But I am unwilling to say “Java never will have function types” (though I recognize that Java may never have function types.) I believe that in order to get to function types, we have to first deal with erasure. That may, or may not be possible. But in a world of reified structural types, function types start to make a lot more sense […]

So, how does this affect us as developers? The following is a categorization of some of the most important new functional interfaces (and some old ones) in the JDK 8 organized by function return type and the number of expected arguments in the interface method.

Functions with Void Return Type

In the realm of functions with a void return type, we have the following:

Type of Function Lambda Expression Known Functional Interfaces
Nullary
				() -> doSomething()
			
Runnable
Unary
				foo  -> System.out.println(foo)
			
Consumer
IntConsumer
LongConsumer
DoubleConsumer
Binary
				(console,text) -> console.print(text)
			
BiConsumer
ObjIntConsumer
ObjLongConsumer
ObjDoubleConsumer
n-ary
				(sender,host,text) -> sender.send(host, text)
			
Define your own

Functions with Some Return Type T

In the realm of functions with a return type T, we have the following:

Type of Function Lambda Expression Known Functional Interfaces
Nullary
				() -> "Hello World"
			
Callable
Supplier
BooleanSupplier
IntSupplier
LongSupplier
DoubleSupplier
Unary
				n -> n + 1
				n -> n >= 0
			
Function
IntFunction
LongFunction
DoubleFunction
IntToLongFunction
IntToDoubleFunction
LongToIntFunction
LongToDoubleFunction
DoubleToIntFunction
DoubleToLongFunction
UnaryOperator
IntUnaryOperator
LongUnaryOperator
DoubleUnaryOperator
Predicate
IntPredicate
LongPredicate
DoublePredicate
Binary
				(a,b) -> a > b ? 1 : 0
				(x,y) -> x + y
				(x,y) -> x % y == 0
			
Comparator
BiFunction
ToIntBiFunction
ToLongBiFunction
ToDoubleBiFunction
BinaryOperator
IntBinaryOperator
LongBinaryOperator
DoubleBinaryOperator
BiPredicate
n-ary
				(x,y,z) -> 2 * x + Math.sqrt(y) - z
			
Define your own

An advantage of this approach is that we can define our own interface types with methods accepting as many arguments as we would like, and we could use them to create lambda expressions and method references as we see fit. In other words, we have the power to pollute the world with yet even more new functional interfaces. Also we can create lambda expressions even for interfaces in earlier versions of the JDK or for earlier versions of our own APIs that defined SAM types like these. And so now we have the power to use Runnable and Callable as functional interfaces.

However, these interfaces become more difficult to memorize since they all have different names and methods.

Still, I am one of those wondering why they didn’t solve the problem as in Scala, defining interfaces like Function0, Function1, Function2, …, FunctionN. Perhaps, the only argument I can come up with against that is that they wanted to maximize the possibilities of defining lambda expressions for interfaces in earlier versions of the APIs as mentioned before.

Lack of Value Types

So, evidently type erasure is one driving force here. But if you are one of those wondering why we also need all these additional functional interfaces with similar names and method signatures and whose only difference is the use of a primitive type, then let me remind you that in Java we also lack of value types like those in a language like C#. This means that the generic types used in our generic classes can only be reference types, and not primitive types.

In other words, we can’t do this:

List<int> numbers = asList(1,2,3,4,5);

But we can indeed do this:

List<Integer> numbers = asList(1,2,3,4,5);

The second example, though, incurs in the cost of boxing and unboxing of the wrapped objects back and forth from/to primitive types. This can become really expensive in operations dealing with collections of primitive values. So, the expert group decided to create this explosion of interfaces to deal with the different scenarios. To make things “less worse” they decided to only deal with three basic types: int, long and double.

Quoting the words of Brian Goetz in the lambda mailing list:

More generally: the philosophy behind having specialized primitive streams (e.g., IntStream) is fraught with nasty tradeoffs. On the one hand, it’s lots of ugly code duplication, interface pollution, etc. On the other hand, any kind of arithmetic on boxed ops sucks, and having no story for reducing over ints would be terrible. So we’re in a tough corner, and we’re trying to not make it worse.

Trick #1 for not making it worse is: we’re not doing all eight primitive types. We’re doing int, long, and double; all the others could be simulated by these. Arguably we could get rid of int too, but we don’t think most Java developers are ready for that. Yes, there will be calls for Character, and the answer is “stick it in an int.” (Each specialization is projected to ~100K to the JRE footprint.)

Trick #2 is: we’re using primitive streams to expose things that are best done in the primitive domain (sorting, reduction) but not trying to duplicate everything you can do in the boxed domain. For example, there’s no IntStream.into(), as Aleksey points out. (If there were, the next question(s) would be “Where is IntCollection? IntArrayList? IntConcurrentSkipListMap?) The intention is many streams may start as reference streams and end up as primitive streams, but not vice versa. That’s OK, and that reduces the number of conversions needed (e.g., no overload of map for int -> T, no specialization of Function for int -> T, etc.)

We can see that this was a difficult decision for the expert group. I think few would agree that this is cool, and most of us would most likely agree it was necessary.

The Checked Exceptions Issue

There was a third driving force that could have made things even worse, and it is the fact that Java supports two type of exceptions: checked and unchecked. The compiler requires that we handle or explicitly declare checked exceptions, but it requires nothing for unchecked ones. So, this creates an interesting problem, because the method signatures of most of the functional interfaces do not declare to throw any exceptions. So, for instance, this is not possible:

Writer out = new StringWriter();
Consumer<String> printer = s -> out.write(s); //oops! compiler error

It cannot be done because the write operation throws a checked exception (i.e. IOException) but the signature of the Consumer method does not declare it throws any exception at all. So, the only solution to this problem would have been to create even more interfaces, some declaring exceptions and some not (or come up with yet another mechanism at the language level for exception transparency). Again, to make things “less worse” the expert group decided to do nothing in this case.

In the words of Brian Goetz in the lambda mailing list:

Yes, you’d have to provide your own exceptional SAMs. But then lambda conversion would work fine with them.

The EG discussed additional language and library support for this problem, and in the end felt that this was a bad cost/benefit tradeoff.

Library-based solutions cause a 2x explosion in SAM types (exceptional vs not), which interact badly with existing combinatorial explosions for primitive specialization.

The available language-based solutions were losers from a complexity/value tradeoff. Though there are some alternative solutions we are going to continue to explore — though clearly not for 8 and probably not for 9 either.

In the meantime, you have the tools to do what you want. I get that you prefer we provide that last mile for you (and, secondarily, your request is really a thinly-veiled request for “why don’t you just give up on checked exceptions already”), but I think the current state lets you get your job done.

So, it’s up to us, the developers, to craft yet even more interface explosions to deal with these in a case-by-case basis:

interface IOConsumer<T> {
   void accept(T t) throws IOException;
}

static<T> Consumer<T> exceptionWrappingBlock(IOConsumer<T> b) {
   return e -> {
	try { b.accept(e); }
	catch (Exception ex) { throw new RuntimeException(ex); }
   };
}

In order to do:

Writer out = new StringWriter();
Consumer<String> printer = exceptionWrappingBlock(s -> out.write(s));

Probably, in the future (maybe JDK 9) when we get Support for Value Types in Java and Reification, we will be able to get rid of (or at least no longer need to use anymore) these multiple interfaces.

In summary, we can see that the expert group struggled with several design issues. The need, requirement or constraint to keep backwards compatibility made things difficult, then we have other important conditions like the lack of value types, type erasure and checked exceptions. If Java had the first and lacked of the other two the design of JDK 8 would probably have been different. So, we all must understand that these were difficult problems with lots of tradeoffs and the EG had to draw a line somewhere and make a decisions.

So, when we find ourselves in the dark side of Java 8, probably we need to remind ourselves that there is a reason why things are dark in that side of the JDK :-)

Further Reading

  • Related Answer in Stackoverflow
  • Lambda Straw-Man Proposal
  • Neal Gafter on The Future of Java
  • Memoized Fibonacci Numbers with Java 8

    Edwin DalorzoSince today is Fibonacci Day, I decided that it would be interesting to publish something related to it.

    I believe one of the first algorithms we all see when learning non-linear recursion is that of calculating a Fibonacci number. I found a great explanation on the subject in the book Structure and Interpretation of Computer Programs [SIC] and I dedicated some time to playing with the Fibonacci algorithm just for fun. While doing so I found an interesting way to improve the classical recursive algorithm by using one of the new methods (added in Java 8) in the Map interface and which I used here to implement a form of memoization.

    Classical Recursive Fibonacci

    In the classical definition of Fibonacci we learn that:

    fib(n) = \left\{ \begin{array}{ll} 0 & \mbox{if n=0}\\1 & \mbox{if n=1}\\fibn(n-1)+fib(n-2) & \mbox{otherwise} \end{array} \right.

    We program this very easily in Java:

    public static long fibonacci(int x) {
       if(x==0 || x==1)
          return x;
       return fibonacci(x-1) + fibonacci(x-2);
    }
    

    Now the problem with this algorithm is that, with the exception of the base case, we recursively invoke our function twice and interestingly one of the branches recalculates part of other branch every time we invoke the function. Consider the following image (taken from SIC) that represents an invocation to fibonacci(5).

    Clearly the branch to the right is redoing all the work already done during the recursive process carried out by the left branch. Can you see how many times fibonacci(2) was calculated? The problem gets worse as the function argument gets bigger. In fact this problem is so serious that the calculation of a small argument like fibonacci(50) might take quite a long time.

    Memoized Recursive Fibonacci

    However, there is a way to improve the performance of the original recursive algorithm (I mean without having to resort to a constant-time algorithm using, for instance, Binet’s formula).

    The serious problem we have in the original algorithm is that we do too much rework. So, we could alleviate the problem by using memoization, in other words by providing a mechanism to avoid repeated calculations by caching results in a lookup table that can later be used to retrieve the values of already processed arguments.

    In Java we could try to store the Fibonacci numbers in a hast table or map. In the case of the left branch we’ll have to run the entire recursive process to obtain the corresponding Fibonacci sequence values, but as we find them, we update the hash table with the results obtained. This way, the right branches will only perform a table lookup and the corresponding value will be retrieved  from the hash table and not through a recursive calculation again.

    Some of the new methods in the class Map , in Java 8, simplify a lot the writing of such algorithm, particularly the method computeIfAbsent(key, function). Where the key would be the number for which we would like to look up the corresponding Fibonacci number and the function would be a lambda expression capable of triggering the recursive calculation if the corresponding value is not already present in the map.

    So, we can start by defining a map and putting the values in it for the base cases, namely, fibonnaci(0) and fibonacci(1):

    private static Map<Integer,Long> memo = new HashMap<>();
    static {
       memo.put(0,0L); //fibonacci(0)
       memo.put(1,1L); //fibonacci(1)
    }
    

    And for the inductive step all we have to do is redefine our Fibonacci function as follows:

    public static long fibonacci(int x) {
       return memo.computeIfAbsent(x, n -> fibonacci(n-1) + fibonacci(n-2));
    }
    

    As you can see, the method computeIfAbsent will use the provided lambda expression to calculate the Fibonacci number when the number is not present in the map, this recursive process will be triggered entirely for the left branch, but the right branch will use the momoized values. This represents a significant improvement.

    Based on my subjective observation, this improved recursive version was outstandingly faster for a discrete number like fibonacci(70). With this algorithm we can safely calculate up to fibonacci(92) without running into long overflow. Even better, to be sure that our algorithm would never cause overflows without letting the user know we could also use one of the new methods in Java 8 added to the Math class and which throws an ArithmeticException when overflow occurs. So we could change our code as follows:

    public static long fibonacci(int x) {
       return memo.computeIfAbsent(x, n -> Math.addExact(fibonacci(n-1),
                                                         fibonacci(n-2)));
    }
    

    This method would start failing for fibonacci(93). If we need to go over 92 we would have to use BigInteger in our algorithm, instead of just long.

    Notice that the memozied example uses mutations, therefore, in order to use this code in a multithreaded environment we would first need to add some form of synchronization to the proposed code, or use a different map implementation, perhaps a ConcurrentHashMap, which evidently, may impact performance as well. Arguably, this would still be better than the original recursive algorithm.

    Java 8 Optional Objects

    Edwin DalorzoIn this post I present several examples of the new Optional objects in Java 8 and I make comparisons with similar approaches in other programming languages, particularly the functional programming language SML and  the JVM-based programming language Ceylon, this latter currently under development by Red Hat.

    I think it is important to highlight that the introduction of optional objects has been a matter of debate. In this article I try to present my perspective of the problem and I do an effort to show arguments in favor and against the use of optional objects. It is my contention that in certain scenarios the use of optional objects is valuable, but ultimately everyone is entitled to an opinion and I just hope this article helps the readers to make an informed one just as writing it helped me understand this problem much better.

    About the Type of Null

    In Java we use a reference type to gain access to an object, and when we don’t have a specific object to make our reference point to, then we set such reference to null to imply the absence of a value.

    In Java null is actually a type, a special one: it has no name, we cannot declare variables of its type, or cast any variables to it, in fact there is a single value that can be associated with it (i.e. the literal null), and unlike any other types in Java, a null reference can be safely assigned to any other reference types (See JLS  3.10.7 and 4.1).

    The use of null is so common that we rarely meditate on it: field members of objects are automatically initialized to null and programmers typically initialize reference types to null when they don’t have an initial value to give them and, in general, null is used everywhere to imply that, at certain point, we don’t know or we don’t have a value to give to a reference.

    About the Null Pointer Reference Problem

    Now, the major problem with the null reference is that if we try to dereference it then we get the ominous and well known NullPointerException.

    When we work with a reference obtained from a different context than our code (i.e. as the result of a method invocation or when we receive a reference as an argument in a method we are working on),  we all would like to avoid this error that has the potential to make our application crash, but often the problem is not noticed early enough and it finds its way into production code where it waits for the right moment to fail (which is typically a Friday at the end of the month, around 5 p.m. and just when you are about to leave the office to go to the movies with your family or drink some beers with your friends). To make things worse, the place where your code fails is rarely the place where the problem originated, since your reference could have been set to null far away from the place in your code where you intended to dereference it. So, you better cancel those plans for the Friday night…

    It’s worth mentioning that this concept of null references was first introduced by Tony Hoare, the creator of ALGOL, back in 1965. The consequences were not so evident in those days, but he later regretted his design and he called it “a billion dollars mistake“, precisely referring to the uncountable amount of hours that many of us have spent, since then, fixing this kind null dereferencing problems.

    Wouldn’t it be great if the type system could tell the difference between a reference that, in a specific context, could be potentially null from one that couldn’t? This would help a lot in terms of type safety because the compiler could then enforce that the programmer do some verification for references that could be null at the same time that it allows a direct use of the others. We see here an opportunity for improvement in the type system. This could be particularly useful when writing the public interface of APIs because it would increase the expressive power of the language, giving us a tool, besides documentation, to tell our users that a given method may or may not return a value.

    Now, before we delve any further, I must clarify that this is an ideal that modern languages will probably pursue (we’ll talk about Ceylon and Kotlin later), but it is not an easy task to try to fix this hole in a programming language like Java when we intend to do it as an afterthought. So, in the coming paragraphs I present some scenarios in which I believe the use of optional objects could arguably alleviate some of this burden. Even so, the evil is done, and nothing will get rid of null references any time soon, so we better learn to deal with them. Understanding the problem is one step and it is my opinion that these new optional objects are just another way to deal with it, particularly in certain specific scenarios in which we would like to express the absence of a value.

    Finding Elements

    There is a set of idioms in which the use of null references is potentially problematic. One of those common cases is when we look for something that we cannot ultimately find. Consider now the following simple piece of code used to find the first fruit in a list of fruits that has a certain name:

    public static Fruit find(String name, List<Fruit> fruits) {
       for(Fruit fruit : fruits) {
          if(fruit.getName().equals(name)) {
             return fruit;
          }
       }
       return null;
    }
    

    As we can see, the creator of this code is using a null reference to indicate the absence of a value that satisfies the search criteria (7). It is unfortunate, though, that it is not evident in the method signature that this method may not return a value, but a null reference..

    Now consider the following code snippet, written by a programmer expecting to use the result of the method shown above:

    List<Fruit> fruits = asList(new Fruit("apple"),
                                new Fruit("grape"),
                                new Fruit("orange"));
    
    Fruit found = find("lemon", fruits);
    //some code in between and much later on (or possibly somewhere else)...
    String name = found.getName(); //uh oh!
    

    Such simple piece of code has an error that cannot be detected by the compiler, not even by simple observation by the programmer (who may not have access to the source code of the find method). The programmer,  in this case, has naively failed to recognize the scenario in which the find method above could return a null reference to indicate the absence of a value that satisfies his predicate. This code is waiting to be executed to simply fail and no amount of documentation is going to prevent this mistake from happening and the compiler will not even notice that there is a potential problem here.

    Also notice that the line where the reference is set to null (5) is different from the problematic line (7). In this case they were close enough, in other cases this may not be so evident.

    In order to avoid the problem what we typically do is that we check if a given reference is null before we try to dereference it. In fact, this verification is quite common and in certain cases this check could be repeated so many times on a given reference that Martin Fowler (renown for hist book on refactoring principles) suggested that for these particular scenarios such verification could  be avoided with the use of what he called a Null Object. In our example above, instead of returning null, we could have returned a NullFruit object reference which is an object of type Fruit that is hollowed inside and which, unlike a null reference, is capable of properly responding to the same public interface of a Fruit.

    Minimum and Maximum

    Another place where this could be potentially problematic is when reducing a collection to a value, for instance to a maximum or minimum value. Consider the following piece of code that can be used to determine which is the longest string in a collection.

    public static String longest(Collection<String> items) {
       if(items.isEmpty()){
          return null;
       }
       Iterator<String> iter = items.iterator();
       String result = iter.next();
       while(iter.hasNext()) {
           String item = iter.next();
           if(item.length() > result.length()){
              result = item;
           }
       }
       return result;
    }
    

    In this case the question is what should be returned when the list provided is empty? In this particular case a null value is returned, once again, opening the door for a potential null dereferencing problem.

    The Functional World Strategy

    It’s interesting that in the functional programming paradigm, the statically-typed programming languages evolved in a different direction. In languages like SML or Haskell there is no such thing as a null value that causes exceptions when dereferenced. These languages provide a special data type capable of holding an optional value and so it can be conveniently used to also express the possible absence of a value.  The following piece of code shows the definition of the SML option type:

    datatype 'a option = NONE | SOME of 'a
    

    As you can see, option is a data type with two constructors, one of them stores nothing (i.e. NONE) whereas the other is capable of storing a polymorphic value of some value type 'a (where 'a is just a placeholder for the actual type).

    Under this model, the piece of code we wrote before in Java, to find a fruit by its name, could be rewritten in SML as follows:

    fun find(name, fruits) =
       case fruits of
            [] => NONE
          | (Fruit s)::fs => if s = name
                             then SOME (Fruit s)
                             else find(name,fs)
    

    There are several ways to achieve this in SML, this example just shows one way to do it. The important point here is that there is no such thing as null, instead a value NONE is returned when nothing is found (3), and a value SOME fruit is returned otherwise (5).

    When a programmer uses this find method, he knows that it returns an option type value and therefore the programmer is forced to check the nature of the value obtained to see if it is either NONE (6) or SOME fruit (7), somewhat like this:

    let
       val fruits = [Fruit "apple", Fruit "grape", Fruit "orange"]
       val found = find("grape", fruits)
    in
       case found of
           NONE => print("Nothing found")
         | SOME(Fruit f) => print("Found fruit: " ^ f)
    end
    

    Having to check for the true nature of the returned option makes it impossible to misinterpret the result.

    Java Optional Types

    It’s a joy that finally in Java 8 we’ll have a new class called Optional that allows us to implement a similar idiom as that from the functional world. As in the case of of SML, the Optional type is polymorphic and may contain a value or be empty. So, we could rewrite our previous code snippet as follows:

    public static Optional<Fruit> find(String name, List<Fruit> fruits) {
       for(Fruit fruit : fruits) {
          if(fruit.getName().equals(name)) {
             return Optional.of(fruit);
          }
       }
       return Optional.empty();
    }
    

    As you can see, the method now returns an Optional reference (1), if something is found, the Optional object is constructed with a value (4), otherwise is constructed empty (7).

    And the programmer using this code would do something as follows:

    List<Fruit> fruits = asList(new Fruit("apple"),
                                new Fruit("grape"),
                                new Fruit("orange"));
    
    Optional<Fruit> found = find("lemon", fruits);
    if(found.isPresent()) {
       Fruit fruit = found.get();
       String name = fruit.getName();
    }
    

    Now it is made evident in the type of the find method that it returns an optional value (5), and the user of this method has to program his code accordingly (6-7).

    So we see that  the adoption of this functional idiom is likely to make our code safer, less prompt to null dereferencing problems and as a result more robust and less error prone. Of course, it is not a perfect solution because, after all, Optional references can also be erroneously set to null references, but  I would expect that programmers stick to the convention of not passing null references where an optional object is expected, pretty much as we today consider a good practice not to pass a null reference where a collection or an array is expected, in these cases the correct is to pass an empty array or collection. The point here is that now we have a mechanism in the API that we can use to make explicit that for a given reference we may not have a value to assign it and the user is forced, by the API, to verify that.

    Quoting an article I reference later about the use of optional objects in the Guava Collections framework: “Besides the increase in readability that comes from giving null a name, the biggest advantage of Optional is its idiot-proof-ness. It forces you to actively think about the absent case if you want your program to compile at all, since you have to actively unwrap the Optional and address that case”.

    Other Convenient Methods

    As of the today, besides the static methods of and empty explained above, the Optional class contains the following convenient instance methods:

    ifPresent() Which returns true if a value is present in the optional.
    get() Which returns a reference to the item contained in the optional object, if present, otherwise throws a NoSuchElementException.
    ifPresent(Consumer<T> consumer) Which passess the optional value, if present, to the provided Consumer (which could be implemented through a lambda expression or method reference).
    orElse(T other) Which returns the value, if present, otherwise returns the value in other.
    orElseGet(Supplier<T> other) Which returns the value if present, otherwise returns the value provided by the Supplier (which could be implemented with a lambda expression or method reference).
    orElseThrow(Supplier<T> exceptionSupplier) Which returns the value if present, otherwise throws the exception provided by the Supplier (which could be implemented with a lambda expression or method reference).

    Avoiding Boilerplate Presence Checks

    We can use some of the convenient methods mentioned above to avoid the need of having to check if a value is present in the optional object. For instance, we may want to use a default fruit value if nothing is found, let’s say that we would like to use a “Kiwi”. So we could rewrite our previous code like this:

    Optional<Fruit> found = find("lemon", fruits);
    String name = found.orElse(new Fruit("Kiwi")).getName();
    

    In this other example, the code prints the fruit name to the main output, if the fruit is present. In this case, we implement the Consumer with a lambda expression.

    Optional<Fruit> found = find("lemon", fruits);
    found.ifPresent(f -> { System.out.println(f.getName()); });
    

    This other piece of code uses a lambda expression to provide a Supplier which can ultimately provide a default answer if the optional object is empty:

    Optional<Fruit> found = find("lemon", fruits);
    Fruit fruit = found.orElseGet(() -> new Fruit("Lemon"));
    

    Clearly, we can see that these convenient methods simplify a lot having to work with the optional objects.

    So What’s Wrong with Optional?

    The question we face is: will Optional get rid of null references? And the answer is an emphatic no! So, detractors immediately question its value asking: then what is it good for that we couldn’t do by other means already?

    Unlike functional languages like SML o Haskell which never had the concept of null references, in Java we cannot simply get rid of the null references that have historically existed. This will continue to exist, and they arguably have their proper uses (just to mention an example: three-valued logic).

    I doubt that the intention with the Optional class is to replace every single nullable reference, but to help in the creation of more robust APIs in which just by reading the signature of a method we could tell if we can expect an optional value or not  and force the programmer to use this value accordingly. But ultimately, Optional will be just another reference and subject to same weaknesses of every other reference in the language. It is quite evident that Optional is not going to save the day.

    How these optional objects are supposed to be used or whether they are valuable or not in Java has been the matter of a heated debate in the project lambda mailing list. From the detractors we hear interesting arguments like:

    • The fact that other alternatives exist ( i.e. the Eclipse IDE supports a set of proprietary annotations for static analysis of nullability, the JSR-305 with annotations like @Nullable and @NonNull).
    • Some would like it to be usable as in the functional world, which is not entirely possible in Java since the language lacks many features existing in functional programming languages like SML or Haskell (i.e. pattern matching).
    • Others argue about how it is impossible to retrofit preexisting code to use this idiom (i.e. List.get(Object)which will continue to return null).
    • And some complain about the fact that the lack of language support for optional values creates a potential scenario in which Optional could be used inconsistently in the APIs, by this creating incompatibilities, pretty much like the ones we will have with the rest of the Java API which cannot be retrofitted to use the new Optional class.
    • A compelling argument is that if the programmer invokes the get method in an optional object, if it is empty, it will raise a NoSuchElementException, which is pretty much the same problem that we have with nulls, just with a different exception.

    So, it would appear that the benefits of Optional are really questionable and are probably constrained to improving readability and enforcing public interface contracts.

    Optional Objects in the Stream API

    Irrespective of the debate, the optional objects are here to stay and they are already being used in the new Stream API in methods like findFirstfindAnymax and min. It could be worth mentioning that  a very similar class has been in used in the successful Guava Collections Framework.

    For instance, consider the following example where we extract from a stream the last fruit name in alphabetical order:

    Stream<Fruit> fruits = asList(new Fruit("apple"),
                                  new Fruit("grape")).stream();
    Optional<Fruit> max = fruits.max(comparing(Fruit::getName));
    if(max.isPresent()) {
       String fruitName = max.get().getName(); //grape
    }
    

    Or this another one in which we obtain the first fruit in a stream

    Stream<Fruit> fruits = asList(new Fruit("apple"),
                                  new Fruit("grape")).stream();
    Optional<Fruit> first = fruits.findFirst();
    if(first.isPresent()) {
       String fruitName = first.get().getName(); //apple
    }
    

    Ceylon Programming Language and Optional Types

    Recently I started to play a bit with the Ceylon programming language since I was doing a research for another post that I am planning to publish soon in this blog. I must say I am not a big fan of Ceylon, but still I found particularly interesting that in Ceylon this concept of optional values is taken a bit further, and the language itself offers some syntactic sugar for this idiom. In this language we can mark any type with a ? (question mark) in order to indicate that its type is an optional type.

    For instance, this find function would be very similar to our original Java version, but this time returning an optional Fruit? reference (1). Also notice that a null value is compatible with the optional Fruit? reference (7).

    Fruit? find(String name, List<Fruit> fruits){
       for(Fruit fruit in fruits) {
          if(fruit.name == name) {
             return fruit;
          }
       }
       return null;
    }
    

    And we could use it with this Ceylon code, similar to our last Java snippet in which we used an optional value:

    List<Fruit> fruits = [Fruit("apple"),Fruit("grape"),Fruit("orange")];
    Fruit? fruit = find("lemon", fruits);
    print((fruit else Fruit("Kiwi")).name);
    

    Notice the use of the else keyword here is pretty similar to the method orElse in the Java 8 Optional class. Also notice that the syntax is similar to the declaration of C# nullable types, but it means something totally different in Ceylon. It may be worth mentioning that Kotlin, the programming language under development by Jetbrains, has a similar feature related to null safety (so maybe we are before a trend in programming languages).

    An alternative way of doing this would have been like this:

    List<Fruit> fruits = [Fruit("apple"),Fruit("grape"),Fruit("orange")];
    Fruit? fruit = find("apple", fruits);
    if(exists fruit){
       String fruitName = fruit.name;
       print("The found fruit is: " + fruitName);
    } //else...
    

    Notice the use of the exists keyword here (3) serves the same purpose as the isPresent method invocation in the Java Optional class.

    The great advantage of Ceylon over Java is that they can use this optional type in the APIs since the beginning, within the realm of their language they won’t have to deal with incompatibilities, and it can be fully supported everywhere (perhaps their problem will be in their integration with the rest of the Java APIs, but I have not studied this yet).

    Hopefully, in future releases of Java, this same syntactic sugar from Ceylon and Kotlin will also be made available in the Java programming language, perhaps using, under the hood, this new Optional class introduced in Java 8.

    Further Reading

    Java Streams API Preview

    Edwin DalorzoFor today’s post I want to share a series of examples that I have developed while trying the latest features in the Java 8 Stream API. In my last post I did a comparison of the features with those available in LINQ, and for today’s post I have decided to go a bit further and try to use the API to work with a small domain model. The examples developed here are based on the same examples presented in the LambdaJ Project.

    The Data Model

    For the examples I will use the following domain model:

    lambda-model

    You can see a full implementation of all the examples developed here by downloading the following Gist.

    For the examples presented below assume that  in the context of the code there are three streams always available:

    • Stream<Person> persons.
    • Stream<Car> cars.
    • Stream<Sale> sales.

    Challenge 1: Print All Car Brands

    From a collection of cars print all car brands.

    StringJoiner brands = cars.map(Car::getBrand)
                              .collect(toStringJoiner(","));
    String allBrands = brands.toString();
    

    For this example I have also use the new StringJoiner class.

    Challenge 2: Select all Sales on Toyota

    From a collection of sales, select all those that are related to Toyota cars.

    List<Sale> toyotaSales;
    
    toyotaSales = sales.filter(s -> s.getCar().getBrand().equals("Toyota"))
                       .collect(toList());
    
    toyotaSales.forEach(System.out::println);
    

    For this example I could have also used the forEach method in the stream to get all the sales printed. I did it this way just to illustrate that it is possible to collect all items in the stream into a list and from there we can process them. But ideally, I should have processed the items directly in the stream.

    Challenge 3: Find Buys of the Youngest Person

    From a collection of sales, find all those that are from the youngest buyer.

    ToIntFunction<Entry<Person, List<Sale>>> byAge;
    byAge = e -> e.getKey().getAge();
    byYoungest = sales.collect(groupingBy(Sale::getBuyer))
                      .entrySet()
                      .stream()
                      .sorted(comparing(byAge))
                      .map(Entry::getValue)
                      .findFirst();
    if(byYoungest.isPresent()) {
     System.out.println(byYoungest.get());
    }
    

    Challenge 4: Find Most Costly Sale

    From a collection of sales, find the most costly of all of them.

    Optional<Sale> mostCostlySale;
    Comparator<Sale> byCost = comparing((ToDoubleFunction<Sale>)Sale::getCost)
                              .reverseOrder();
    
    mostCostlySale = sales.sorted( byCost )
                          .findFirst();
    
    if(mostCostlySale.isPresent()) {
    	System.out.println(mostCostlySale.get());
    }
    

    Challenge 5: Sum of Sales from Male Buyers & Sellers

    From a collection of sales find the sum of all buys/sells made by men.

    double sum = sales.filter(s -> s.getBuyer().isMale()
                                   && s.getSeller().isMale())
                      .mapToDouble(Sale::getCost)
                      .sum();
    

    Challenge 6: Find the Age of the Youngest Buyer

    From a collection of sales, find the age of the youngest buyer who bought for more than 40,000.

    OptionalInt ageOfYoungest;
    
    ageOfYoungest = sales.filter(sale -> sale.getCost() > 40000)
                         .map(Sale::getBuyer)
                         .mapToInt(Person::getAge)
                         .sorted()
                         .findFirst();
    
    if(ageOfYoungest.isPresent()) {
    	System.out.println(ageOfYoungest.getAsInt());
    }
    

    Challenge 7: Sort Sales by Cost

    Sort a collection of sales by cost.

    Comparator<Sale> byCost= comparing((ToDoubleFunction<Sale>) Sale::getCost);
    List<Sale> sortedByCost;
    
    sortedByCost = sales.sorted( byCost )
                        .collect(toList());
    

    Challenge 8: Index Cars by Brand

    From a collection of cars, index cars by the their brand.

    Map<String,List<Car>> byBrand;
    byBrand = cars.collect( groupingBy(Car::getBrand ));
    

    Challenge 9: Find Most Bought Car

    From a collection of sales find the most bought car.

    ToIntFunction<Entry<Car,List<Sale>>> toSize = (e -> e.getValue().size());
    
    Optional<Car> mostBought;
    
    mostBought = sales.collect( groupingBy(Sale::getCar) )
                      .entrySet()
                      .stream()
                      .sorted( comparing(toSize).reverseOrder() )
                      .map(Entry::getKey)
                      .findFirst();
    
    if(mostBought.isPresent()) {
       System.out.println(mostBought.get());
    }
    

    Related Posts

    Futher Reading

    Java Streams Preview vs .Net High-Order Programming with LINQ

    Edwin DalorzoIn today’s post I want to share with you a comparison that I did this weekend between the high-order programming features from LINQ and those being incorporated into the new Java 8 streams API.  It is important to highlight that the Java Stream API is still work in progress. I have been downloading and building the source code of Project Lambda for almost a year now and I have seen how the API evolves over time, but the improvements are more frequent as we approach the general availability date (planed for September 2013).

    Before I delve into the comparisons, I must say that as of today, the current implementation of the Java Stream API is far away from all the high-order programming functionality offered by LINQ. It has already been announced in the mailing list that there was not going to be enough time for all improvements originally planned and the JEP 108: Collection Enhancements for Third-party Libraries has been deferred to future releases of Java. I am not sure how much of the Stream API is going to be affected by this decision but I hope that in the remaining months, prior to the release, they do much more improvements in the current API, at least, bring it closer to LINQ.

    At any rate, as of today, it appears that some basic functionality is already covered (filter, map, reduce, sort, etc.), but much more work is still needed for this API to be decently compared to LINQ in terms of functionality and my comparison here only focuses particularly in the high-order programming features. So, I really hope that it is only too early to write this article, and in the coming months things are going to improve dramatically.

    Also, there are some inherent problems that have been evidently difficult to overcome. One of those affecting the design of the API is the lack of value types in Java. It appears that the expert group have been struggling with this, and their solution has been to make copies of some interfaces, defining some to deal with primitive types (i.e. IntStream, LongStream, DoubleStream) and others with reference types (i.e. Stream). This has also caused a diaspora of copies of other functional interfaces (aka interface polution), particularly in the case of the interface Function the problem is much more evident (i.e. ToIntFunction, ToLongFunction, ToDoubleFunction and Function).

    Even when all this repetition is done to alleviate the performance problems inherent to boxing and unboxing of primitive types, this workaround has undesirable side effects: first, the need to overload methods for everyone of these interfaces. This overloading is undesirable because it makes it impossible to use a clean lambda expression for these cases, since the compiler is incapable to determine which of all the argument functional interfaces is the one being implemented by the lambda. So, to avoid compiler errors we are  forced to use awful type castings in the code, which make it more verbose and look pretty bad. And a secondary side effect is the incompatibility between primitive type and reference type collections (i.e. an Stream<Integer> is not the same as an IntStream).

    About the Comparison

    For this exercise I took the code examples from the book .Net 4.0 Generics a Beginners Guide by Sudipta Mukherjee, particularly his examples from Chapter 4: LINQ to Objects. I did not do all the examples simply because there are many features still lacking in the Java Stream API which make it impossible to replicate the same functionality.

    The author of the book makes all these examples to demonstrate LINQ high-order programming functionality. It is not  my intention to write an article on LINQ and all it can do, but a comparison of what I could do with the high-order programming features in the Stream API today if I wanted to implement similar idioms as those I can so easily implement using LINQ functionality. Perhaps I may have misinterpreted the author in his use of all these examples to illustrate LINQ, or perhaps, the author only intended to illustrate just one aspect of this broad technology. Therefore, this article is  based on my understanding of his view of this part of LINQ in order to make a comparison, but I can understand that LINQ is more than just a bunch of high-order functions.

    Now some may argue that there is no intention in making the stream API something comparable to LINQ, which may well be true, but still LINQ offers a good example of an API that makes extensive use of high-order programming to query collections, and even if the stream API is not yet intended to query things like XML or SQL (at least initially/yet), I feel that it is still a good benchmark to compare the power of expressiveness in this new Java API. For me, this gives me an idea of how evolved this API is today when compared to the power of expressiveness of this another state-of-art technology. As it is customary in the internet world, others may freely disagree with me and present their personal perspectives as I did here.

    For the cases that I could implement, the examples are based on the use of Stream and IntStream classes, and the static methods available on the Streams and Collectors helper classes. In order to keep it simple and readable I am not showing in the code examples below all the imports and static imports that  I used. For that matter you may want to take a look at the following Gist where I have published most of these examples and where you will be able to see full imports needed to make the code run properly.

    Restriction Operators

    Challenge 1: Filtering

    Say we have a List of names and we would like to find all those names where "am" occurs:


    LINQ

    string[] names = { "Sam", "Pamela", "Dave", "Pascal", "Erik" };
    List<string> filteredNames = names.Where(c => c.Contains("am"))
                                      .ToList();
    

    Java Streams

    String[] names = {"Sam","Pamela", "Dave", "Pascal", "Erik"};
    List<String> filteredNames = stream(names)
    			     .filter(c -> c.contains("am"))
    			     .collect(toList());
    

    Challenge 2: Indexed Filtering

    Find all the names in the array "names" where the length of the name is less than or equal to the index of the element + 1.


    LINQ

    string[] names = { "Sam", "Pamela", "Dave", "Pascal", "Erik" };
    var nameList = names.Where((c, index) => c.Length <= index + 1).ToList();
    

    Java Streams

    Now this one was particularly tricky in Java because, as of today, the API for streams does not contain any methods that indicate the index of an element within the stream. So, I was forced to generate an indexed stream out of the original array:

    String[] names = {"Sam","Pamela", "Dave", "Pascal", "Erik"};
    
    List<String> nameList;
    Stream<Integer> indices = intRange(1, names.length).boxed();
    nameList = zip(indices, stream(names), SimpleEntry::new)
    			.filter(e -> e.getValue().length() <= e.getKey())
    			.map(Entry::getValue)
    			.collect(toList());
    

    Now, the lack of indices in the stream made the algorithm more verbose, but it was also interesting to notice the incompatibilities between primitive-type streams, like IntStream and reference type streams like Stream<Integer>. In this case, I was forced to transform the IntStream returned by intRange into a Stream<Integer> in order to make the argument compatible with the types expected by zip as you can see in the line #4.


    Projection Operators

    Challenge 3: Selecting/Mapping

    Say we have a list of names and we would like to print “Hello” in front of all the names:


     LINQ

    List<string> nameList1 = new List(){ "Anders", "David", "James",
                                         "Jeff", "Joe", "Erik" };
    nameList1.Select(c => "Hello! " + c).ToList()
             .ForEach(c => Console.WriteLine(c));
    

    Java Streams

    List<String> nameList1 = asList("Anders", "David", "James",
                                    "Jeff", "Joe", "Erik");
    nameList1.stream()
    	 .map(c -> "Hello! " + c)
    	 .forEach(System.out::println);
    

    Challenge 4: Selecting Many/Flattening

    Suppose, we have a dictionary such that each key has a list of values attached to them. Now, we want to project all the elements in a single collection:


    LINQ

    Dictionary<string, List<string>> map = new Dictionary<string,List<string>>();
    map.Add("UK", new List<string>() {"Bermingham", "Bradford", "Liverpool"});
    map.Add("USA", new List<string>() {"NYC", "New Jersey", "Boston", "Buffalo"});
    var cities = map.SelectMany(c => c.Value).ToList();
    

    Java Streams

    Once more, the Java syntax, as of today, is a bit more verbose. First, we must transform the map to an entry set, and from there, we can generate a stream that we can later flatten to a list of cities, as follows:

    Map<String, List<String>> map = new LinkedHashMap<>();
    map.put("UK", asList("Bermingham","Bradford","Liverpool"));
    map.put("USA", asList("NYC","New Jersey","Boston","Buffalo"));
    
    FlatMapper<Entry<String, List<String>>,String> flattener;
    flattener = (entry,consumer) -> { entry.getValue().forEach(consumer); };
    
    List<String> cities = map.entrySet()
    			 .stream()
    			 .flatMap( flattener )
    			 .collect(toList());
    

    Ideally the lines 5, 6 should be defined as inline arguments of flatMap, but the compiler cannot properly infer the types of the expression precisely for the overloading problems that I described above. Given the amount of type information that must be provided, I felt that it was best to define it in another line as I did in #5 and #6. It looks awful,  I know. Clearly, we are in desperate need of an API that offers operations to deal with maps or tuples.


    Partitioning Operators

    Challenge 5: Taking an Arbitrary Number of Items

    In this case we are interested in evaluating only the first n elements of a collection. For instance, using a finite list of numbers, obtain the first 4 of them.


    LINQ

    int[] numbers = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13 };
    var first4 = numbers.Take(4).ToList();
    

    Java Streams

    Once more, in Java, the absence of value types forced me to make a conversion from an IntStream into a Stream<Integer> to make this work (see line #5), basically because there is not value type collector. An alternative would have been to consume the stream itself (i.e. forEach) or to collect items into an array using Stream.toArray() method.

    int[] numbers = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12,13 };
    
    List<Integer> firstFour;
    firstFour = stream(numbers).limit(4)
                               .boxed()
                               .collect(toList());
    

    Challenge 6: Taking Items Based on Predicate

    In this case we are interested in taking items out of a collection as long as they satisfy a predicate. Once we find an item that does not satisfy the predicate we stop there.


    LINQ

    string[] moreNames = { "Sam", "Samuel", "Dave", "Pascal", "Erik",  "Sid" };
    var sNames = moreNames.TakeWhile(c => c.StartsWith("S"));
    

    Java Streams

    As of today, there is no way to implement this idiom in Java using the streams API. There is an alternative way to do this, but it is not the same thing. The beauty of takeWhile is that is should be short-circuited, that is, it should stop the evaluation in the moment that one item does not satisfies the predicate. The following version in Java does not have this property:

    String[] names  = { "Sam","Samuel","Dave","Pascal","Erik","Sid" };
    
    List<String> found;
    found = stream(names).collect(partitioningBy( c -> c.startsWith("S")))
                         .get(true);
    

    The collector created by partitioningBy (line #4) forces the evaluation of the whole stream, placing items into a boolean map, where all items that satisfy the predicate are bound to true and all others to false. So clearly, it is not the same thing. I hope that as the Oracle expert group works on the API they fix this omission.


    Challenge 7: Skipping an Arbitrary Number of Items

    In this case we are interested in skipping items in a collection up to certain arbitrary number, then we keep the rest of the items.


    LINQ

    string[] vipNames = { "Sam", "Samuel", "Samu", "Remo", "Arnold","Terry" };
    var skippedList = vipNames.Skip(3).ToList();//Leaving the first 3.
    

    Java Streams

    In Java, the solution consists in creating a new stream where the first n elements have been discarded. As follows:

    String[] vipNames = { "Sam", "Samuel", "Samu", "Remo", "Arnold","Terry" };
    
    List<String> skippedList;
    skippedList = stream(vipNames).substream(3).collect(toList());
    

    Challenge 8: Skipping Items Based on Predicate

    In this case we are interested in skipping items out of a collection as long as they satisfy a predicate. Once we find an item that does not satisfy the predicate we take the rest of the items from there.


    LINQ

    int[] numbers = { 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 20 };
    var skippedList = numbers.SkipWhile(c => c < 10);
    

    Java Streams

    With current streams API I found no way to implement this idiom.

    Ordering Operators

    Challenge 9: Ordering/Sorting Elements

    Order the elements of a collection alphabetically:


    LINQ

    string[] friends = { "Sam", "Pamela", "Dave", "Anders", "Erik" };
    friends = friends.OrderBy(c => c).ToArray();
    

    Java Streams

    In the case of Java, we use the sorted method to produce the same result. The sorted method can also accept a Comparator to determine the sorting criteria.

    String[] friends = { "Sam", "Pamela", "Dave", "Anders", "Erik" };
    friends = stream(friends).sorted().toArray(String[]::new);
    

    Challenge 10: Ordering/Sorting Elements by Specific Criterium

    Order the elements of a collection strings according to the length of the string:


    LINQ

    string[] friends = { "Sam", "Pamela", "Dave", "Anders", "Erik" };
    friends = friends.OrderBy(c => c.Length).ToArray();
    

    Java Streams

    In this case we pass a Comparator to the sort method. And once again, we hit against the problem of overloaded methods defined in the API to deal with the lack of value types in Java. Here we are forced to provide an explicit casting (line #3) to help the compiler:

    String[] friends = { "Sam", "Pamela", "Dave", "Anders", "Erik" };
    friends = stream(friends)
               .sorted(comparing((ToIntFunction<String>)String::length))
               .toArray(String[]::new);
    

    An alternative way could be to provide an implementation of the Comparator (line #3), in this case, by means of a binary lambda expression. This is a little cleaner, but more verbose.

    String[] friends = { "Sam", "Pamela", "Dave", "Anders", "Erik" };
    friends = stream(friends)
                .sorted( (s1,s2) -> Integer.compare(s1.length(), s2.length()))
                .toArray(String[]::new);
    

    Challenge 11: Ordering/Sorting Elements by Multiple Criteria

    Order the elements of a collection of strings according to several sorting criteria:


    LINQ

    string[] fruits = {"grape", "passionfruit", "banana",
                       "apple", "orange", "raspberry",
                       "mango", "blueberry" };
    
    //Sort the strings first by their length and then alphabetically.
    //preserving the first order.
    var sortedFruits = fruits.OrderBy(fruit =>fruit.Length)
                             .ThenBy(fruit => fruit);
    

    Java Streams

    Originally I had thought it was not possible to implement this idiom with the latest release of the API, but one of our readers had a good suggestion. Even so I was not able to get rid of the castings in lines #5 and #7. Once again the interface pollution causes the need for castings to clarify which of the overloaded methods are the ones being implemented in Comparators and Comparator here.

    String[] fruits = {"grape", "passionfruit", "banana","apple",
                       "orange", "raspberry","mango", "blueberry" };
    
    Comparator<String> comparator;
    comparator = comparing((Function<String,Integer>)String::length,
                           Integer::compare)
                .thenComparing((Comparator<String>)String::compareTo);
    
    fruits = stream(fruits) .sorted(comparator)
                            .toArray(String[]::new);
    

    Grouping Operators

    Challenge 12: Grouping by a Criterium

    Group the elements of a collection of strings by their length.


    LINQ

    string[] names = {"Sam", "Samuel", "Samu", "Ravi", "Ratna",  "Barsha"};
    var groups = names.GroupBy(c => c.Length);
    

    Java Streams

    String[] names = {"Sam", "Samuel", "Samu", "Ravi", "Ratna",  "Barsha"};
    
    Map<Integer,List<String>> groups;
    groups = stream(names).collect(groupingBy(String::length));
    

    Set Operators

    The current implementation of streams is way behind LINQ in this area. From all possible set operations, the only ones currently implemented are “distinct” and “concat”, although “concat” is not a set operation because it would accept duplicates, the correct would be to have a “union” operation, but this does not exist in the stream API yet.

    Challenge 13: Filter Distinct Elements

    Obtain all the distinct elements from a collection.


    LINQ

    string[] songIds = {"Song#1", "Song#2", "Song#2", "Song#2", "Song#3", "Song#1"};
    //This will work as strings implement IComparable
    var uniqueSongIds = songIds.Distinct();
    

    Java Streams

    String[] songIds = {"Song#1", "Song#2", "Song#2", "Song#2", "Song#3", "Song#1"};
    //according to Object.equals
    stream(songIds).distinct();
    

    Challenge 14: Union of Two Sets

    Join together two sets of items.


    LINQ

    List<string> friends1 = new List<string>() {"Anders", "David","James",
                                                "Jeff", "Joe", "Erik"};
    List<string> friends2 = new List<string>() { "Erik", "David", "Derik" };
    var allMyFriends = friends1.Union(friends2);
    

    Java Streams

    In Java we have to concatenate the two streams and then obtain the distinct elements.

    List<String> friends1 = asList("Anders","David","James","Jeff","Joe","Erik");
    List<String> friends2 = asList("Erik","David","Derik");
    Stream<String> allMyFriends = concat(friends1.stream(),
                                         friends2.stream()).distinct();
    

    Element Operatos

    Challenge 15: First Element

    Obtain the first element of a collection.


    LINQ

    string[] otherFriends = {"Sam", "Danny", "Jeff", "Erik", "Anders","Derik"};
    string firstName = otherFriends.First();
    string firstNameConditional = otherFriends.First(c => c.Length == 5);
    

    Java Streams

    In Java we use the findFirst method which returns an Option object. The object may contain something or nothing, to validate that one must invoke the isPresent method on the returned object.

    String[] otherFriends = {"Sam", "Danny", "Jeff", "Erik", "Anders","Derik"};
    Optional<String> found = stream(otherFriends).findFirst();
    
    Optional<String> maybe = stream(otherFriends).filter(c -> c.length() == 5)
                                                 .findFirst();
    if(maybe.isPresent()) {
       //do something with found data
    }
    

    Range Operators

    Challenge 16: Generate a Range of Numbers

    Generate a range of numbers that are multiples of 11.


    LINQ

    var multiplesOfEleven = Enumerable.Range(1, 100).Where(c => c % 11 == 0);
    

    Java Streams

    IntStream multiplesOfEleven = intRange(1,100).filter(n -> n % 11 == 0);
    

    Quantifier Operators

    Challenge 17: All

    Do all elements in a collection satisfy a predicate?


    LINQ

    string[] persons = {"Sam", "Danny", "Jeff", "Erik", "Anders","Derik"};
    bool x = persons.All(c => c.Length == 5);
    

    Java Streams

    String[] persons = {"Sam", "Danny", "Jeff", "Erik", "Anders","Derik"};
    boolean x = stream(persons).allMatch(c -> c.length() == 5);
    

    Challenge 18: Any

    Do any elements in a collection satisfy a predicate?


    LINQ

    string[] persons = {"Sam", "Danny", "Jeff", "Erik", "Anders","Derik"};
    bool x = persons.Any(c => c.Length == 5);
    

    Java Streams

    String[] persons = {"Sam", "Danny", "Jeff", "Erik", "Anders","Derik"};
    boolean x = stream(persons).anyMatch(c -> c.length() == 5);
    

    Merging Operators

    Challenge 19: Zip

    Combine two collections into a single collection.


    LINQ

    string[] salutations = {"Mr.", "Mrs.", "Ms", "Master"};
    string[] firstNames = {"Samuel", "Jenny", "Joyace", "Sam"};
    string lastName = "McEnzie";
    
    salutations.Zip(firstNames, (sal, first) => sal + " " + first)
               .ToList()
               .ForEach(c => Console.WriteLine(c + " " + lastName));
    

    Java Streams

    String[] salutations = {"Mr.", "Mrs.", "Ms", "Master"};
    String[] firstNames = {"Samuel", "Jenny", "Joyace", "Sam"};
    String lastName = "McEnzie";
    
    zip(
        stream(salutations),
        stream(firstNames),
        (sal,first) -> sal + " " +first)
    .forEach(c -> { System.out.println(c + " " + lastName); });
    

    In general, things are still not looking good, but I really hope that will change in the coming months. There are challenges that will be difficult to overcome, like the issues with value types, but in my opinion we could  live with those problems as long as we are provided with an API sufficiently expressive.

    I am currently working on another set of examples in which I work on a small data model and see what I can do with it to run all kinds of queries using the Stream API, but I will leave that for my next post.

    Related Posts

    Further Reading