com.github.tonivade:purefun-transformer

Functional Programming Library for Java


Keywords
effects, experimental, free-monad, functional-programming, higher-kinded-types, immutable-collections, io-monad, java, memoization, monad, monad-transformers, monads, mtl, stream, tagless-final, tuples, typeclasses
License
MIT

Documentation

Purefun

 ____                  __             
|  _ \ _   _ _ __ ___ / _|_   _ _ __  
| |_) | | | | '__/ _ \ |_| | | | '_ \ 
|  __/| |_| | | |  __/  _| |_| | | | |
|_|    \__,_|_|  \___|_|  \__,_|_| |_|

Build Status Codacy Badge Join the chat at https://gitter.im/tonivade/purefun

This module was developed as the core of the zeromock project. It defines all the basic classes and interfaces used in the rest of the project.

Initially the module only held a few basic interfaces and it has grown to become an entire functional programming library (well, a humble one), and now is an independent library.

Working in this library helped me to learn and understand some important concepts of functional programming, and over time I implemented higher kinded types and type classes in Java. I don't know if this will be helpful for somebody, but the work is here to everyone want to use it.

Finally, I have to say thanks to vavr library author, this library is largely inspired in his work, and also to Scala standard library authors. I don't want to forget some of the projects I've used as reference: Arrow and cats for type classes implementation, fs2 for stream processing, and ZIO to implement my own version in Java. Their awesome work help me a lot.

Disclaimer

This project is not ready to be used in production, I use it to learn functional programming concepts by my self, but, if you want to use it, use it at your own risk. Anyway if you think is useful for you, go ahead, also any feedback and PR are very welcome.

Higher Kinded Types

In this project I have implemented some patterns of functional programming that need Higher Kinded Types. In Java there are not such thing, but it can be simulated using a especial codification of types.

In Scala we can define a higher kinded typed just like this Monad[F[_]] but in Java it can be codified like this Monad<F extends Witness>. Witness is a simple mark interface. Then we can define a type using a special codification like this:

interface SomeType<T> extends SomeTypeOf<T> { }

// Boilerplate
interface SomeTypeOf<T> implements Kind<SomeType_, T> {

  default Kind<SomeType_, T> kind() {
    return this;
  }

  // this is a safe cast
  static SomeType<T> narrowK(Kind<SomeType_, T> hkt) {
    return (SomeType<T>) hkt;
  }
  
  static Fixer<Kind<SomeType_, T>, SomeType<T>> toSomeType() {
    return SomeType::narrowK;
  }
}

final class SomeType_ extends Witness {
  private SomeType_() {}
}

It can be triky but, in the end is easy to work with. By the way, I tried to hide this details to the user of the library. Except with type classes because is the only way to implement them correctly.

So, there are interfaces to encode kinds of 1, 2 and 3 types. It can be defined types for 4, 5 or more types, but it wasn't necessary to implement the library.

Annotation Processor

In order to simplify working with higher kinded types, in the last version I've included an annotation processor to generate all this boilerplate code:

@HigherKind
interface SomeType<T> extends SomeTypeOf<T> { }

With this annotation, all the above code, is generated automatically.

Data types

Option

Is an alternative to Optional of Java standard library. It can contains two values, a some or a none

Option<String> some = Option.some("Hello world");

Option<String> none = Option.none();

Try

Is an implementation of scala Try in Java. It can contains two values, a success or a failure.

Try<String> success = Try.success("Hello world");

Try<String> failure = Try.failure(new RuntimeException("Error"));

Either

Is an implementation of scala Either in Java.

Either<Integer, String> right = Either.right("Hello world");

Either<Integer, String> left = Either.left(100);

Validation

This type represents two different states, valid or invalid, an also it allows to combine several validations using map2 to map5 methods.

Validation<String, String> name = Validation.valid("John Smith");
Validation<String, String> email = Validation.valid("john.smith@example.net");

// Person has a constructor with two String parameters, name and email.
Valdation<Sequence<String>, Person> person = Validation.map2(name, email, Person::new);

Const

A object with a phantom parameter:

Const<String, Integer> constInt = Const.of("Hello world!");

Const<String, Float> constFloat = constInt.retag();

assertEquals("Hello world!", constFloat.value());

Future

This is an experimental implementation of Future. Computations are executed in another thread inmediatelly.

Future<String> future = Future.success("Hello world!");

Future<String> result = future.flatMap(string -> Future.run(string::toUpperCase));

assertEquals(Try.success("HELLO WORLD!"), result.await());

Tuples

These classes allow to hold some values together, as tuples. There are tuples from 1 to 5.

Tuple1<String> tuple1 = Tuple.of("Hello world");

Tuple2<String, Integer> tuple2 = Tuple.of("John Smith", 100);

Data structures

Java doesn't define immutable collections, so I have implemented some of them.

Sequence

Is the equivalent to java Collection interface. It defines all the common methods.

ImmutableList

It represents a linked list. It has a head and a tail.

ImmutableSet

It represents a set of elements. This elements cannot be duplicated.

ImmutableArray

It represents an array. You can access to the elements by its position in the array.

ImmutableMap

This class represents a hash map.

ImmutableTree

This class represents a binary tree.

ImmutableTreeMap

This class represents a binary tree map.

Monads

Also I have implemented some Monads that allows to combine some operations.

State Monad

Is the traditional State Modad from FP languages, like Haskel or Scala. It allows to combine operations over a state. The state should be a immutable class. It recives an state and generates a tuple with the new state and an intermediate result.

State<ImmutableList<String>, Option<String>> read = State.state(list -> Tuple.of(list.tail(), list.head()));

Tuple<ImmutableList<String>, Option<String>> result = read.run(ImmutableList.of("a", "b", "c"));

assertEquals(Tuple.of(ImmutableList.of("b", "c"), Option.some("a")), result);

Reader Monad

This is an implementation of Reader Monad. It allows to combine operations over a common input. It can be used to inject dependencies.

Reader<ImmutableList<String>, String> read2 = Reader.reader(list -> list.tail().head().orElse(""));

String result = read2.eval(ImmutableList.of("a", "b", "c"));

assertEqual("b", result);

Writer Monad

It allow to combine operations over a common output.

Writer<ImmutableList<String>, Integer> writer = Writer.<String, Integer>listPure(5)
    .flatMap(value -> listWriter("add 5", value + 5))
    .flatMap(value -> listWriter("plus 2", value * 2));

assertAll(() -> assertEquals(Integer.valueOf(20), writer.getValue()),
          () -> assertEquals(listOf("add 5", "plus 2"), writer.getLog()));

IO Monad

This is a experimental implementation of IO Monad in java. Inspired in this work.

IO<Unit> echo = Console.print("write your name")
  .andThen(Console.read())
  .flatMap(name -> Console.print("Hello " + name))
  .andThen(Console.print("end"));

echo.unsafeRunSync();

Trampoline

Implements recursion using an iteration and is stack safe.

private Trampoline<Integer> fibLoop(Integer n) {
if (n < 2) {
  return Trampoline.done(n);
}
return Trampoline.more(() -> fibLoop(n - 1)).flatMap(x -> fibLoop(n - 2).map(y -> x + y));
}

Free

Free Monad

Finally, after hours of hard coding, I managed to implement a Free monad. This is a highly unstable implementation and I have implemented because it can be implemented. Inspired in this work.

Free<IOProgram_, Unit> echo =
  IOProgram.write("what's your name?")
    .andThen(IOProgram.read())
    .flatMap(text -> IOProgram.write("Hello " + text))
    .andThen(IOProgram.write("end"));

Kind<IO_, Unit> foldMap = echo.foldMap(IOInstances.monad(), new IOProgramInterperter());

foldMap.fix(toIO()).unsafeRunSync();

Free Applicative

Similar to Free monad, but allows static analysis without to run the program.

FreeAp<DSL_, Tuple5<Integer, Boolean, Double, String, Unit>> tuple =
    applicative.map5(
        DSL.readInt(2),
        DSL.readBoolean(false),
        DSL.readDouble(2.1),
        DSL.readString("hola mundo"),
        DSL.readUnit(),
        Tuple::of
    ).fix(toFreeAp());

Kind<Id_, Tuple5<Integer, Boolean, Double, String, Unit>> map =
    tuple.foldMap(idTransform(), IdInstances.applicative());

assertEquals(Id.of(Tuple.of(2, false, 2.1, "hola mundo", unit())), map.fix(toId()));

Monad Transformers

OptionT

Monad Transformer for Option type

OptionT<IO_, String> some = OptionT.some(IO.monad(), "abc");

OptionT<IO_, String> map = some.flatMap(value -> OptionT.some(IOInstances.monad(), value.toUpperCase()));

assertEquals("ABC", map.get().fix(toIO()).unsafeRunSync());

EitherT

Monad Transformer for Either type

EitherT<IO_, Nothing, String> right = EitherT.right(IO.monad(), "abc");

EitherT<IO_, Nothing, String> map = right.flatMap(value -> EitherT.right(IOInstances.monad(), value.toUpperCase()));

assertEquals("ABC", map.get().fix(toIO()).unsafeRunSync());

StateT

Monad Transformer for State type

StateT<IO_, ImmutableList<String>, Unit> state =
  pure("a").flatMap(append("b")).flatMap(append("c")).flatMap(end());

IO<Tuple2<ImmutableList<String>, Unit>> result = state.run(ImmutableList.empty()).fix(toIO());

assertEquals(Tuple.of(listOf("a", "b", "c"), unit()), result.unsafeRunSync());

WriterT

Monad Transformer for Writer type

WriterT<Id_, Sequence<String>, Integer> writer =
    WriterT.<Id_, Sequence<String>, Integer>pure(monoid, monad, 5)
    .flatMap(value -> lift(monoid, monad, Tuple.of(listOf("add 5"), value + 5)))
    .flatMap(value -> lift(monoid, monad, Tuple.of(listOf("plus 2"), value * 2)));

assertAll(() -> assertEquals(Id.of(Integer.valueOf(20)), writer.getValue()),
          () -> assertEquals(Id.of(listOf("add 5", "plus 2")), writer.getLog()));

Kleisli

Also I implemented the Kleisli composition for functions that returns monadic values like Option, Try or Either.

Kleisli<Try_, String, Integer> toInt = Kleisli.lift(Try.monad(), Integer::parseInt);
Kleisli<Try_, Integer, Double> half = Kleisli.lift(Try.monad(), i -> i / 2.);

Kind<Try_, Double> result = toInt.compose(half).run("123");

assertEquals(Try.success(61.5), result);

Stream

An experimental version of a Stream like scala fs2 project.

StreamOf<IO_> streamOfIO = Stream.ofIO();

IO<String> readFile = streamOfIO.eval(IO.of(() -> reader(file)))
  .flatMap(reader -> streamOfIO.iterate(() -> Option.of(() -> readLine(reader))))
  .takeWhile(Option::isPresent)
  .map(Option::get)
  .foldLeft("", (a, b) -> a + "\n" + b)
  .fix(toIO())
  .recoverWith(UncheckedIOException.class, cons("--- file not found ---"));

String content = readFile.unsafeRunSync();

Effects

An experimental version of PureIO similar to ZIO.

PureIO<Console, Throwable, Unit> echoProgram =
  Console.println("what's your name?")
      .andThen(Console.readln())
      .flatMap(name -> Console.println("Hello " + name));

interface Console {

  <R extends Console> Console.Service<R> console();

  static PureIO<Console, Throwable, String> readln() {
    return PureIO.accessM(env -> env.console().readln());
  }

  static PureIO<Console, Throwable, Unit> println(String text) {
    return PureIO.accessM(env -> env.console().println(text));
  }

  interface Service<R extends Console> {
    PureIO<R, Throwable, String> readln();

    PureIO<R, Throwable, Unit> println(String text);
  }
}

Additionally, there are aliases for some PureIO special cases:

UIO<T>      =>  PureIO<?, Nothing, T>
EIO<E, T>   =>  PureIO<?, E, T>
Task<T>     =>  PureIO<?, Throwable, T>
RIO<R, T>   =>  PureIO<R, Throwable, T>
URIO<T>     =>  PureIO<R, Nothing, T>

Algebraic Effects

Also, I have implemented a version of delimited control monad based in this project. With this monad, you can implement algebraic effects. Example:

  // the effect
  interface Amb {
    Control<Boolean> flip();
  }

  // the program, if true then 2 else 3
  Control<Integer> program(Amb amb) {
    return amb.flip().map(x -> x ? 2 : 3);
  }

  @Test
  void test() {
    Control<ImmutableList<Integer>> handled = ambList(this::program);

    assertEquals(listOf(2, 3), handled.run());
  }

  <R> Control<ImmutableList<R>> ambList(Function1<Amb, Control<R>> program) {
    return new AmbList<R>().apply(amb -> program.apply(amb).map(ImmutableList::of));
  }

  final class AmbList<R> implements Handler<ImmutableList<R>, Amb>, Amb {

    @Override
    public Amb effect() { return this; }

    @Override
    public Control<Boolean> flip() {
      return use(resume ->
          resume.apply(true) // first flip return true
            .flatMap(ts -> resume.apply(false) // second flip return false
              .map(ts::appendAll)));
    }
  }

Type Classes

With higher kinded types simulation we can implement typeclases.

                       Invariant -- Contravariant
                             \
       SemigroupK           Functor -- Comonad
           |               /       \
         MonoidK   _ Applicative   Traverse -- Foldable
           |      /      |      \
       Alternative    Selective  ApplicativeError
                         |        |
  MonadWriter          Monad      |
        \________________|        |
        /          /      \      / 
  MonadState  MonadReader  MonadError_____
                               \          \
                            MonadThrow  Bracket
                                  \      /
                        Defer -- MonadDefer -- Timer
                                     |
                                   Async
                                     |
                                 Concurrent

Functor

public interface Functor<F extends Witness> extends Invariant<F> {

  <T, R> Kind<F, R> map(Kind<F, T> value, Function1<T, R> map);
}

Applicative

public interface Applicative<F extends Witness> extends Functor<F> {

  <T> Kind<F, T> pure(T value);

  <T, R> Kind<F, R> ap(Kind<F, T> value, Kind<F, Function1<T, R>> apply);

  @Override
  default <T, R> Kind<F, R> map(Kind<F, T> value, Function1<T, R> map) {
    return ap(value, pure(map));
  }
}

Selective

public interface Selective<F extends Witness> extends Applicative<F> {

  <A, B> Kind<F, B> select(Kind<F, Either<A, B>> value, Kind<F, Function1<A, B>> apply);

  default <A, B, C> Kind<F, C> branch(Kind<F, Either<A, B>> value,
                                      Kind<F, Function1<A, C>> applyA,
                                      Kind<F, Function1<B, C>> applyB) {
    Kind<F, Either<A, Either<B, C>>> abc = map(value, either -> either.map(Either::left));
    Kind<F, Function1<A, Either<B, C>>> fabc = map(applyA, fb -> fb.andThen(Either::right));
    return select(select(abc, fabc), applyB);
  }
}

Monad

public interface Monad<F extends Witness> extends Selective<F> {

  <T, R> Kind<F, R> flatMap(Kind<F, T> value, Function1<T, ? extends Kind<F, R>> map);

  @Override
  default <T, R> Kind<F, R> map(Kind<F, T> value, Function1<T, R> map) {
    return flatMap(value, map.andThen(this::pure));
  }

  @Override
  default <T, R> Kind<F, R> ap(Kind<F, T> value, Kind<F, Function1<T, R>> apply) {
    return flatMap(apply, map -> map(value, map));
  }

  @Override
  default <A, B> Kind<F, B> select(Kind<F, Either<A, B>> value, Kind<F, Function1<A, B>> apply) {
    return flatMap(value, either -> either.fold(a -> map(apply, map -> map.apply(a)), this::<B>pure));
  }
}

Semigroup

It represents a binary operation over a type.

@FunctionalInterface
public interface Semigroup<T> {
  T combine(T t1, T t2);
}

There are instances for strings and integers.

Monoid

Extends Semigroup adding a zero operation that represent an identity.

public interface Monoid<T> extends Semigroup<T> {
  T zero();
}

There are instances for strings and integers.

SemigroupK

It represents a Semigroup but defined for a kind, like a List, so it extends a regular Semigroup.

MonoidK

The same like SemigroupK but for a Monoid.

Invariant

public interface Invariant<F extends Witness> {
  <A, B> Kind<F, B> imap(Kind<F, A> value, Function1<A, B> map, Function1<B, A> comap);
}

Contravariant

public interface Contravariant<F extends Witness> extends Invariant<F> {
  <A, B> Kind<F, B> contramap(Kind<F, A> value, Function1<B, A> map);
}

Bifunctor

public interface Bifunctor<F extends Witness> {
  <A, B, C, D> Kind<Kind<F, C>, D> bimap(Kind<Kind<F, A>, B> value, Function1<A, C> leftMap, Function1<B, D> rightMap);
}

Profunctor

public interface Profunctor<F extends Witness> {
  <A, B, C, D> Kind<Kind<F, C>, D> dimap(Kind<Kind<F, A>, B> value, Function1<C, A> contramap, Function1<B, D> map);
}

Applicative Error

public interface ApplicativeError<F extends Witness, E> extends Applicative<F> {

  <A> Kind<F, A> raiseError(E error);

  <A> Kind<F, A> handleErrorWith(Kind<F, A> value, Function1<E, ? extends Kind<F, A>> handler);
}

Monad Error

public interface MonadError<F extends Witness, E> extends ApplicativeError<F, E>, Monad<F> {

  default <A> Kind<F, A> ensure(Kind<F, A> value, Producer<E> error, Matcher1<A> matcher) {
    return flatMap(value, a -> matcher.match(a) ? pure(a) : raiseError(error.get()));
  }
}

Monad Throw

public interface MonadThrow<F extends Witness> extends MonadError<F, Throwable> {

}

MonadReader

public interface MonadReader<F extends Witness, R> extends Monad<F> {

  Kind<F, R> ask();

  default <A> Kind<F, A> reader(Function1<R, A> mapper) {
    return map(ask(), mapper);
  }
}

MonadState

public interface MonadState<F extends Witness, S> extends Monad<F> {
  Kind<F, S> get();
  Kind<F, Unit> set(S state);

  default Kind<F, Unit> modify(Operator1<S> mapper) {
    return flatMap(get(), s -> set(mapper.apply(s)));
  }

  default <A> Kind<F, A> inspect(Function1<S, A> mapper) {
    return map(get(), mapper);
  }

  default <A> Kind<F, A> state(Function1<S, Tuple2<S, A>> mapper) {
    return flatMap(get(), s -> mapper.apply(s).applyTo((s1, a) -> map(set(s1), x -> a)));
  }
}

MonadWriter

public interface MonadWriter<F extends Witness, W> extends Monad<F> {

  <A> Kind<F, A> writer(Tuple2<W, A> value);
  <A> Kind<F, Tuple2<W, A>> listen(Kind<F, A> value);
  <A> Kind<F, A> pass(Kind<F, Tuple2<Operator1<W>, A>> value);

  default Kind<F, Unit> tell(W writer) {
    return writer(Tuple.of(writer, unit()));
  }
}

Comonad

public interface Comonad<F extends Witness> extends Functor<F> {

  <A, B> Kind<F, B> coflatMap(Kind<F, A> value, Function1<Kind<F, A>, B> map);

  <A> A extract(Kind<F, A> value);

  default <A> Kind<F, Kind<F, A>> coflatten(Kind<F, A> value) {
    return coflatMap(value, identity());
  }
}

Foldable

public interface Foldable<F extends Witness> {

  <A, B> B foldLeft(Kind<F, A> value, B initial, Function2<B, A, B> mapper);

  <A, B> Eval<B> foldRight(Kind<F, A> value, Eval<B> initial, Function2<A, Eval<B>, Eval<B>> mapper);
}

Traverse

public interface Traverse<F extends Witness> extends Functor<F>, Foldable<F> {

  <G extends Witness, T, R> Kind<G, Kind<F, R>> traverse(Applicative<G> applicative, Kind<F, T> value,
      Function1<T, ? extends Kind<G, R>> mapper);
}

Semigroupal

public interface Semigroupal<F extends Witness> {

  <A, B> Kind<F, Tuple2<A, B>> product(Kind<F, A> fa, Kind<F, B> fb);
}

Defer

public interface Defer<F extends Witness> {

  <A> Kind<F, A> defer(Producer<Kind<F, A>> defer);
}

Bracket

public interface Bracket<F extends Witness, E> extends MonadError<F, E> {

  <A, B> Kind<F, B> bracket(Kind<F, A> acquire, Function1<A, ? extends Kind<F, B>> use, Consumer1<A> release);
}

MonadDefer

public interface MonadDefer<F extends Witness> extends MonadThrow<F>, Bracket<F, Throwable>, Defer<F>, Timer<F> {

  default <A> Kind<F, A> later(Producer<A> later) {
    return defer(() -> Try.of(later::get).fold(this::raiseError, this::pure));
  }
}

Async

public interface Async<F extends Witness> extends MonadDefer<F> {
  
  <A> Kind<F, A> async(Consumer1<Consumer1<Try<A>>> consumer);

}

Timer

public interface Timer<F extends Witness> {

  Kind<F, Unit> sleep(Duration duration);
}

FunctionK

It represents a natural transformation between two different kinds.

public interface FunctionK<F extends Witness, G extends Witness> {
  <T> Kind<G, T> apply(Kind<F, T> from);
}

Optics

         __Iso__
        /       \
    Lens        Prism
        \       /
         Optional

Iso

An Iso is an optic which converts elements of some type into elements of other type without loss. In other words, it's isomorphic.

Point point = new Point(1, 2);

Iso<Point, Tuple2<Integer, Integer>> pointToTuple = 
  Iso.of(p -> Tuple.of(p.x, p.y), 
         t -> new Point(t.get1(), t.get2()));

assertEquals(point, pointToTuple.set(pointToTuple.get(point)));

Lens

A Lens is an optic used to zoom inside a structure. In other words, it's an abstraction of a setter and a getter but with immutable objects.

Lens<Employee, String> nameLens = Lens.of(Employee::getName, Employee::withName);

Employee pepe = new Employee("pepe");

assertEquals("pepe", nameLens.get(pepe));
assertEquals("paco", nameLens.get(nameLens.set(pepe, "paco")));

We can compose Lenses to get deeper inside. For example, if we add an attribute of type Address into Employee. We can create a lens to access the city name of the address of the employee.

Lens<Employee, Address> addressLens = Lens.of(Employee::getAddress, Employee::withAddress);
Lens<Address, String> cityLens = Lens.of(Address::getCity, Address::withCity);
Lens<Employee, String> cityAddressLens = addressLens.compose(cityLens);

Employee pepe = new Employee("pepe", new Address("Madrid"));

assertEquals("Madrid", cityAddressLens.get(pepe));

Prism

A Prism is a lossless invertible optic that can see into a structure and optionally find a value.

Function1<String, Option<Integer>> parseInt = ...; // is a method that only returns a value when the string can be parsed

Prism<String, Integer> stringToInteger = Prism.of(parseInt, String::valueOf);

assertEquals(Option.some(5), stringToInteger.getOption("5"));
assertEquals(Option.none(), stringToInteger.getOption("a"));
assertEquals("5", stringToInteger.reverseGet(5));

Optional

An Optional is an optic that allows to see into a structure and getting, setting like a Lens an optional find a value like a Prism.

Optional<Employee, Address> addressOptional = Optional.of(
  Employee::withAddress, employee -> Option.of(employee::getAddress)
);

Address madrid = new Address("Madrid");
Employee pepe = new Employee("pepe", null);

assertEquals(Option.none(), addressOptional.getOption(pepe));
assertEquals(Option.some(madrid), addressOptional.getOption(addressOptional.set(pepe, madrid)));

Composition

Optional Prism Lens Iso
Optional Optional Optional Optional Optional
Prism Optional Prism Optional Prism
Lens Optional Optional Lens Lens
Iso Optional Prism Lens Iso

Equal

This class helps to create readable equals methods. An example:

@Override
public boolean equals(Object obj) {
  return Equal.<Data>of()
    .comparing(Data::getId)
    .comparing(Data::getValue)
    .applyTo(this, obj);
}

Stargazers over time

Stargazers over time

License

purefun is released under MIT license