mergeMap/flatMap and Java 8 type inference

[benjchristensen]

@adelnizamutdinov I don't see a new issue created, so wanted to pick it up here.

You said this:

mergeMap is almost unusable with Java 8 type inference
Always have to extend Observable::just and Observable::empty to anonymous classes

Can you elaborate?

[benjchristensen]

        Observable.from(1, 2, 3, 4).mergeMap(i -> {
            if (i < 3) {
                return Observable.just("value_" + i);
            } else {
                return Observable.empty();
            }
        }).forEach(System.out::println);

[benjchristensen]

It also correctly infers it is a String:

        Observable<String> mergeMap = Observable.from(1, 2, 3, 4).mergeMap(i -> {
            if (i < 3) {
                return Observable.just("value_" + i);
            } else {
                return Observable.empty();
            }
        });

[meoyawn]

@benjchristensen
here's the example that won't compile because of the variance/contravariance and java 8 lambda's type inference:

static class User {
    String name;
    int age;

    User(String name, int age) {
        this.name = name;
        this.age = age;
    }
}

public static void main(String[] args) {
    savedUser()
            .mergeMap(Observable::just, e -> fetchUser(), Observable::empty)
            .forEach(user -> System.out.println(user.name));
}

static Observable<User> savedUser() {
    return Observable.error(new IllegalStateException("there are no saved users"));
}

static Observable<User> fetchUser() {
    return Observable.just(new User("Adel", 13));
}

[benjchristensen]

This code compiles and runs for me:

import rx.Observable;


public class Issue1550 {

    static class User {
        String name;
        int age;

        User(String name, int age) {
            this.name = name;
            this.age = age;
        }
    }

    public static void main(String[] args) {
        savedUser()
                .mergeMap(Observable::just, e -> fetchUser(), Observable::empty)
                .forEach(user -> System.out.println(user.name));
    }

    static Observable<User> savedUser() {
        return Observable.error(new IllegalStateException("there are no saved users"));
    }

    static Observable<User> fetchUser() {
        return Observable.just(new User("Adel", 13));
    }

}

It outputs:

Adel

[benjchristensen]

Ah interesting, it works in Eclipse, but not from command line.

lgml-bechristensen:libs benjchristensen$ /Library/Java/JavaVirtualMachines/jdk1.8.0_0.jdk/Contents/Home/bin/javac -cp rxjava-core-0.20.0-RC4-SNAPSHOT.jar Issue1550.java 
Issue1550.java:18: error: incompatible types: no instance(s) of type variable(s) T#1 exist so that Observable<T#1> conforms to ? extends Observable<? extends R>
                .mergeMap(Observable::just, e -> fetchUser(), Observable::empty)
                         ^
  where T#1,R,T#2 are type-variables:
    T#1 extends Object declared in method <T#1>empty()
    R extends Object declared in method <R>mergeMap(Func1<? super T#2,? extends Observable<? extends R>>,Func1<? super Throwable,? extends Observable<? extends R>>,Func0<? extends Observable<? extends R>>)
    T#2 extends Object declared in class Observable
1 error

[benjchristensen]

I believe this is now fixed as of #1558.

If you can confirm against the branch, please do. Reopen if there are still issues.

[meoyawn]

@benjchristensen I don't have the rights to reopen the issues
but I cloned the master, built the jar and tried to compile:

$JAVA8_HOME/bin/javac Test.java -cp rxjava-core-0.20.0-RC5-SNAPSHOT.jar
Test.java:19: error: incompatible types: no instance(s) of type variable(s) T#1 exist so that Observable<T#1> conforms to ? extends Observable<? extends R>
                .mergeMap(Observable::just, e -> fetchUser(), Observable::empty)
                         ^
  where T#1,R,T#2 are type-variables:
    T#1 extends Object declared in method <T#1>empty()
    R extends Object declared in method <R>mergeMap(Func1<? super T#2,? extends Observable<? extends R>>,Func1<? super Throwable,? extends Observable<? extends R>>,Func0<? extends Observable<? extends R>>)
    T#2 extends Object declared in class Observable

I don't really think this is an issue, because expanding the method ref as a new Func1... works just fine, but is ugly in java8
so I thought maybe it's me who's doing something wrong
I am very bad with co- and contravariance so I can't even figure out how to change mergeMap's signature so that the thing compiles with method references

[meoyawn]

@benjchristensen I'm using Oracle JDK on Ubuntu 14.04

$JAVA8_HOME/bin/javac -version
javac 1.8.0_11

[benjchristensen]

@zsxwing You're good at figuring this kind of stuff out, so is there anything we should be doing differently here, or is this just a weakness of Java 8?

[zsxwing]

@benjchristensen actually, I know little about Java type inference. I try the following codes and it works:

import rx.Observable;
import rx.functions.*;

public class Issue1550 {

    static class User {
        String name;
        int age;

        User(String name, int age) {
            this.name = name;
            this.age = age;
        }
    }

    public static void main(String[] args) {
        Func0<Observable<User>> empty = () -> Observable.empty();
        Func1<User, Observable<User>> just = (User user) -> Observable.just(user);
        savedUser().flatMap(
                just,
                e -> fetchUser(),
                empty)
        .forEach(user -> System.out.println(user.name));
    }

    static Observable<User> savedUser() {
        return Observable.error(new IllegalStateException("there are no saved users"));
    }

    static Observable<User> fetchUser() {
        return Observable.just(new User("Adel", 13));
    }

}

So I guess it's a weakness of Java type inference. Looks we have to write a temp variable to tell the compiler type information.

[zsxwing]

The following codes work, too:

import rx.Observable;
import rx.functions.*;

public class Issue1550 {

    static class User {
        String name;
        int age;

        User(String name, int age) {
            this.name = name;
            this.age = age;
        }
    }

    public static void main(String[] args) {
        Func0<Observable<User>> empty = Observable::empty;
        Func1<User, Observable<User>> just = Observable::just;
        savedUser().flatMap(
                just,
                e -> fetchUser(),
                empty)
        .forEach(user -> System.out.println(user.name));
    }

    static Observable<User> savedUser() {
        return Observable.error(new IllegalStateException("there are no saved users"));
    }

    static Observable<User> fetchUser() {
        return Observable.just(new User("Adel", 13));
    }

}

[akarnokd]

Java 8 type inference has some weaknesses, and javac won't compile code that will compile just fine in Eclipse 4.4. Based on some comments in lambda-dev, those javac shortcommings won't be fixed until Java 9.

[benjchristensen]

so @akarnokd is it your assessment that there is not anything better we can do with the RxJava APIs and it's just up to the language/compiler itself?

[akarnokd]

Java 6 and 7 developers would need the full covariance in the parameter types and only Java 8 non-Eclipse developers have to do workarounds such as above. Besides, the Stream.flatMap has full covariance in its signature:

<R> Stream<R> flatMap(Function<? super T,? extends Stream<? extends R>> mapper)

So it is either lambda workarounds for Java 8 devs or unchecked casting for Java 6/7 to bypass the lack of variance.

[benjchristensen]

I'm considering this "as good as it can get" then since our flatMap matches the Stream interface definition:

<R> Observable<R> flatMap(Func1<? super T, ? extends Observable<? extends R>> func)

Thanks @akarnokd for the insight.