[java] When do you use map vs flatMap in RxJava?

When do you use map vs flatMap in RxJava?

Say, for example, we want to map Files containing JSON into Strings that contain the JSON--

Using map, we have to deal with the Exception somehow. But how?:

Observable.from(jsonFile).map(new Func1<File, String>() {
    @Override public String call(File file) {
        try {
            return new Gson().toJson(new FileReader(file), Object.class);
        } catch (FileNotFoundException e) {
            // So Exception. What to do ?
        }
        return null; // Not good :(
    }
});

Using flatMap, it's much more verbose, but we can forward the problem down the chain of Observables and handle the error if we choose somewhere else and even retry:

Observable.from(jsonFile).flatMap(new Func1<File, Observable<String>>() {
    @Override public Observable<String> call(final File file) {
        return Observable.create(new Observable.OnSubscribe<String>() {
            @Override public void call(Subscriber<? super String> subscriber) {
                try {
                    String json = new Gson().toJson(new FileReader(file), Object.class);

                    subscriber.onNext(json);
                    subscriber.onCompleted();
                } catch (FileNotFoundException e) {
                    subscriber.onError(e);
                }
            }
        });
    }
});

I like the simplicity of the map, but the error handling of flatmap (not the verbosity). I haven't seen any best practices on this floating around and I'm curious how this is being used in practice.

In that scenario use map, you don't need a new Observable for it.

you should use Exceptions.propagate, which is a wrapper so you can send those checked exceptions to the rx mechanism

Observable<String> obs = Observable.from(jsonFile).map(new Func1<File, String>() { 
    @Override public String call(File file) {
        try { 
            return new Gson().toJson(new FileReader(file), Object.class);
        } catch (FileNotFoundException e) {
            throw Exceptions.propagate(t); /will propagate it as error
        } 
    } 
});

You then should handle this error in the subscriber

obs.subscribe(new Subscriber<String>() {
    @Override 
    public void onNext(String s) { //valid result }

    @Override 
    public void onCompleted() { } 

    @Override 
    public void onError(Throwable e) { //e might be the FileNotFoundException you got }
};); 

There is an excellent post for it: http://blog.danlew.net/2015/12/08/error-handling-in-rxjava/

In some cases you might end up having chain of observables, wherein your observable would return another observable. 'flatmap' kind of unwraps the second observable which is buried in the first one and let you directly access the data second observable is spitting out while subscribing.

I just wanted to add that with flatMap, you don't really need to use your own custom Observable inside the function and you can rely on standard factory methods/operators:

Observable.from(jsonFile).flatMap(new Func1<File, Observable<String>>() {
    @Override public Observable<String> call(final File file) {
        try {
            String json = new Gson().toJson(new FileReader(file), Object.class);
            return Observable.just(json);
        } catch (FileNotFoundException ex) {
            return Observable.<String>error(ex);
        }
    }
});

Generally, you should avoid throwing (Runtime-) exceptions from onXXX methods and callbacks if possible, even though we placed as many safeguards as we could in RxJava.

Flatmap maps observables to observables. Map maps items to items.

Flatmap is more flexible but Map is more lightweight and direct, so it kind of depends on your usecase.

If you are doing ANYTHING async (including switching threads), you should be using Flatmap, as Map will not check if the consumer is disposed (part of the lightweight-ness)

The way I think about it is that you use flatMap when the function you wanted to put inside of map() returns an Observable. In which case you might still try to use map() but it would be unpractical. Let me try to explain why.

If in such case you decided to stick with map, you would get an Observable<Observable<Something>>. For example in your case, if we used an imaginary RxGson library, that returned an Observable<String> from it's toJson() method (instead of simply returning a String) it would look like this:

Observable.from(jsonFile).map(new Func1<File, Observable<String>>() {
    @Override public Observable<String>> call(File file) {
        return new RxGson().toJson(new FileReader(file), Object.class);
    }
}); // you get Observable<Observable<String>> here

At this point it would be pretty tricky to subscribe() to such an observable. Inside of it you would get an Observable<String> to which you would again need to subscribe() to get the value. Which is not practical or nice to look at.

So to make it useful one idea is to "flatten" this observable of observables (you might start to see where the name _flat_Map comes from). RxJava provides a few ways to flatten observables and for sake of simplicity lets assume merge is what we want. Merge basically takes a bunch of observables and emits whenever any of them emits. (Lots of people would argue switch would be a better default. But if you're emitting just one value, it doesn't matter anyway.)

So amending our previous snippet we would get:

Observable.from(jsonFile).map(new Func1<File, Observable<String>>() {
    @Override public Observable<String>> call(File file) {
        return new RxGson().toJson(new FileReader(file), Object.class);
    }
}).merge(); // you get Observable<String> here

This is a lot more useful, because subscribing to that (or mapping, or filtering, or...) you just get the String value. (Also, mind you, such variant of merge() does not exist in RxJava, but if you understand the idea of merge then I hope you also understand how that would work.)

So basically because such merge() should probably only ever be useful when it succeeds a map() returning an observable and so you don't have to type this over and over again, flatMap() was created as a shorthand. It applies the mapping function just as a normal map() would, but later instead of emitting the returned values it also "flattens" (or merges) them.

That's the general use case. It is most useful in a codebase that uses Rx allover the place and you've got many methods returning observables, which you want to chain with other methods returning observables.

In your use case it happens to be useful as well, because map() can only transform one value emitted in onNext() into another value emitted in onNext(). But it cannot transform it into multiple values, no value at all or an error. And as akarnokd wrote in his answer (and mind you he's much smarter than me, probably in general, but at least when it comes to RxJava) you shouldn't throw exceptions from your map(). So instead you can use flatMap() and

return Observable.just(value);

when all goes well, but

return Observable.error(exception);

when something fails.
See his answer for a complete snippet: https://stackoverflow.com/a/30330772/1402641

FlatMap behaves very much like map, the difference is that the function it applies returns an observable itself, so it's perfectly suited to map over asynchronous operations.

In the practical sense, the function Map applies just makes a transformation over the chained response (not returning an Observable); while the function FlatMap applies returns an Observable<T>, that is why FlatMap is recommended if you plan to make an asynchronous call inside the method.

Summary:

• Map returns an object of type T
• FlatMap returns an Observable.

A clear example can be seen here: http://blog.couchbase.com/why-couchbase-chose-rxjava-new-java-sdk .

Couchbase Java 2.X Client uses Rx to provide asynchronous calls in a convenient way. Since it uses Rx, it has the methods map and FlatMap, the explanation in their documentation might be helpful to understand the general concept.

To handle errors, override onError on your susbcriber.

Subscriber<String> mySubscriber = new Subscriber<String>() {
    @Override
    public void onNext(String s) { System.out.println(s); }

    @Override
    public void onCompleted() { }

    @Override
    public void onError(Throwable e) { }
};

It might help to look at this document: http://blog.danlew.net/2014/09/15/grokking-rxjava-part-1/

A good source about how to manage errors with RX can be found at: https://gist.github.com/daschl/db9fcc9d2b932115b679

In your case you need map, since there is only 1 input and 1 output.

map - supplied function simply accepts an item and returns an item which will be emitted further (only once) down.

flatMap - supplied function accepts an item then returns an "Observable", meaning each item of the new "Observable" will be emitted separately further down.

May be code will clear things up for you:

Observable.just("item1").map( str -> {
    System.out.println("inside the map " + str);
    return str;
}).subscribe(System.out::println);

Observable.just("item2").flatMap( str -> {
    System.out.println("inside the flatMap " + str);
    return Observable.just(str + "+", str + "++" , str + "+++");
}).subscribe(System.out::println);

Output:

inside the map item1
item1
inside the flatMap item2
item2+
item2++
item2+++

The question is When do you use map vs flatMap in RxJava?. And I think a simple demo is more specific.

When you want to convert item emitted to another type , in your case converting file to String, map and flatMap can both work. But I prefer map operator because it's more clearly.

However in some place, flatMap can do magic work but map can't. For example, I want to get a user's info but I have to first get his id when user login in. Obviously I need two requests and they are in order.

Let's begin.

Observable<LoginResponse> login(String email, String password);

Observable<UserInfo> fetchUserInfo(String userId);

Here are two methods, one for login returned Response, and another for fetching user info.

login(email, password)
        .flatMap(response ->
                fetchUserInfo(response.id))
        .subscribe(userInfo -> {
            // get user info and you update ui now
        });

As you see, in function flatMap applies, at first I get user id from Response then fetch user info. When two requests are finished, we can do our job such as updating UI or save data into database.

However if you use map you can't write such nice code. In a word, flatMap can help us serialize requests.

Here is a simple thumb-rule that I use help me decide as when to use flatMap() over map() in Rx's Observable.

Once you come to a decision that you're going to employ a map transformation, you'd write your transformation code to return some Object right?

If what you're returning as end result of your transformation is:

• a non-observable object then you'd use just map(). And map() wraps that object in an Observable and emits it.

• an Observable object, then you'd use flatMap(). And flatMap() unwraps the Observable, picks the returned object, wraps it with its own Observable and emits it.

Say for example we've a method titleCase(String inputParam) that returns Titled Cased String object of the input param. The return type of this method can be String or Observable<String>.

• If the return type of titleCase(..) were to be mere String, then you'd use map(s -> titleCase(s))

• If the return type of titleCase(..) were to be Observable<String>, then you'd use flatMap(s -> titleCase(s))

Hope that clarifies.