Java: What is the best way to filter a Collection?

Question

I want to filter a java.util.Collection based on a predicate.

Answer 1

org.apache.commons.collections.CollectionUtils#filter(Collection,Predicate)

Answer 2

Consider Google Collections for an updated Collections framework that supports generics.

Answer 3

Are you sure you want to filter the Collection itself, rather than an iterator?

see org.apache.commons.collections.iterators.FilterIterator

Answer 4

The setup:

public interface Predicate<T> {
  public boolean filter(T t);
}

void filterCollection(Collection<T> col, Predicate<T> predicate) {
  for (Iterator i = col.iterator(); i.hasNext();) {
    T obj = i.next();
    if (predicate.filter(obj)) {
      i.remove();
    }
  }
}

The usage:

List<MyObject> myList = ...;
filterCollection(myList, new Predicate<MyObject>() {
  public boolean filter(MyObject obj) {
    return obj.shouldFilter();
  }
});

Answer 5

"Best" way is too wide a request. Is it "shortest"? "Fastest"? "Readable"? Filter in place or into another collection?

Simplest (but not most readable) way is to iterate it and use Iterator.remove() method:

Iterator<Foo> it = col.iterator();
while( it.hasNext() ) {
  Foo foo = it.next();
  if( !condition(foo) ) it.remove();
}

Now, to make it more readable, you can wrap it into a utility method. Then invent a IPredicate interface, create an anonymous implementation of that interface and do something like:

CollectionUtils.filterInPlace(col,
  new IPredicate<Foo>(){
    public boolean keepIt(Foo foo) {
      return foo.isBar();
    }
  });

where filterInPlace() iterate the collection and calls Predicate.keepIt() to learn if the instance to be kept in the collection.

I don't really see a justification for bringing in a third-party library just for this task.

Answer 6

Assuming that you are using Java 1.5, and that you cannot add Google Collections, I would do something very similar to what the Google guys did. This is a slight variation on Jon's comments.

First add this interface to your codebase.

public interface Predicate<T> { boolean apply(T type); }

Its implementors can answer when a certain predicate is true of a certain type. E.g. If T were User and AuthorizedUserPredicate<User> implements Predicate<T>, then AuthorizedUserPredicate#apply returns whether the passed in User is authorized.

Then in some utility class, you could say

public static <T> Collection<T> filter(Collection<T> target, Predicate<T> predicate) {
    Collection<T> result = new ArrayList<T>();
    for (T element: target) {
        if (predicate.apply(element)) {
            result.add(element);
        }
    }
    return result;
}

So, assuming that you have the use of the above might be

Predicate<User> isAuthorized = new Predicate<User>() {
    public boolean apply(User user) {
        // binds a boolean method in User to a reference
        return user.isAuthorized();
    }
};
// allUsers is a Collection<User>
Collection<User> authorizedUsers = filter(allUsers, isAuthorized);

If performance on the linear check is of concern, then I might want to have a domain object that has the target collection. The domain object that has the target collection would have filtering logic for the methods that initialize, add and set the target collection.

Answer 7

com.google.common.collect.Collections2#filter(Collection,Predicate)

in Google Collections

Answer 8

Use the jbfilter framework : http://code.google.com/p/jbfilter/

Answer 9

With the ForEach DSL you may write

import static ch.akuhn.util.query.Query.select;
import static ch.akuhn.util.query.Query.$result;
import ch.akuhn.util.query.Select;

Collection<String> collection = ...

for (Select<String> each : select(collection)) {
    each.yield = each.value.length() > 3;
}

Collection<String> result = $result();

Given a collection of [The, quick, brown, fox, jumps, over, the, lazy, dog] this results in [quick, brown, jumps, over, lazy], ie all strings longer than three characters.

All iteration styles supported by the ForEach DSL are

• AllSatisfy
• AnySatisfy
• Collect
• Counnt
• CutPieces
• Detect
• GroupedBy
• IndexOf
• InjectInto
• Reject
• Select

For more details, please refer to https://www.iam.unibe.ch/scg/svn_repos/Sources/ForEach

Answer 10

lambdaj allows to filter collections without writing loops or inner classes as in the following example:

List<Person> beerDrinkers = select(persons, having(on(Person.class).getAge(),
    greaterThan(16)));

Can you imagine something more readable? You can find it here:

http://code.google.com/p/lambdaj/

Answer 11

This, combined with the lack of real closures, is my biggest gripe for Java. Honestly, most of the methods mentioned above are pretty easy to read and REALLY efficient; however, after spending time with .Net, Erlang, etc... list comprehension integrated at the language level makes everything so much cleaner. Without additions at the language level, Java just cant be as clean as many other languages in this area.

If performance is a huge concern, Google collections is the way to go (or write your own simple predicate utility). Lambdaj syntax is more readable for some people, but it is not quite as efficient.

And then there is a library I wrote. I will ignore any questions in regard to its efficiency (yea, its that bad)...... Yes, i know its clearly reflection based, and no I don't actually use it, but it does work:

LinkedList<Person> list = ......
Iterable<Person> filtered = 
                        Query.from(list).where("x=> x.Age >= 21 && x.Age <= 50");

Answer 12

I wrote an extended Iterable class that support applying functional algorithms without copying the collection content.

Usage:

List<Integer> myList = new ArrayList<Integer>(){ 1, 2, 3, 4, 5 }

Iterable<Integer> filtered = Iterable.wrap(myList).select(new Predicate1<Integer>()
{
    public Boolean call(Integer n) throws FunctionalException
    {
        return n % 2 == 0;
    }
})

for( int n : filtered )
{
    System.out.println(n);
}

The code above will actually execute

for( int n : myList )
{
    if( n % 2 == 0 ) 
    {
        System.out.println(n);
    }
}