A right-biased implementation of Haskell's Either a b for Java, using Java 8 for mapping/folding and type inference.
An Either<A,B> is used to "condense" two distinct options down to a single choice.
"Right-biased" means that, like Haskell's Either a b, when an Either<A, B> has both options available, it prefers
the one on the "right" (that is, B). The mnemonic often used in Haskell to remember/explain this "bias" is that an
Either (usually used for error-checking), gives you either the right answer, or else whatever's left.
Yes, polymorphism is great in Java. Animal.speak(), Dog.speak(), Cat.speak(), and so on.
But sometimes you want to be REALLY EXPLICIT about the possible types of objects you're dealing with, either (ha ha) because they don't logically make sense in the same inheritance structure or because you want to limit your possible cases to exactly two classes.
For example, let's say I'm dealing with my code and some third-party library that still uses java.util.Date (a class
that has been largely deprecated since JDK 1.1, and is almost always a strictly-worse choice than using either the newer
java.time apis, or Joda Time) .
Terrible, I know, but there's always horrible legacy code out there in the world.
I don't want to further propagate this old usage of a terrible deprecated class, so I, a Modern Java Developer, prefer
to use java.time.LocalDateTime in my code.
Except great, now I've gotta deal with grabbing a List<LocalDateTime> from these and a List<Date> from those and
then do a bunch of gnarly conversion everywhere.
Using Either.java, I can instead easily build a List<Either<Date,LocalDatetime>>. This tells me (and the compiler)
that I'm dealing with a bunch of things that are either a Date (yuck) or a LocalDateTime.
I can even nicely convert them all into LocalDateTimes:
List<Either<Date, LocalDateTime>> birthdays = allThePeople.stream()
.map(person -> Either.<Date,LocalDateTime>either(this::getDeprecatedDOBFromPerson, this::getNiceNewDOBFromPersonIfAvailable)
.collect(Collectors.toList());
List<LocalDateTime> convertedBirthdays = birthdays.stream()
.map(eitherDateOrLocalDateTime.fold(
(Date deprecatedDOB) -> myFunctionToConvertDateToLocalDateTime(deprecatedDOB),
(LocalDatetime newDOB) -> newDOB
)).collect(Collectors.toList());Boom. Now you have a list of LocalDateTimes, and it's explicit from the code that some of the people had deprecated Dates, while others had nice, shiny new LocalDateTimes. Even the compiler can tell!
Other common use-cases for an Either include capturing and handling errors gracefully using
Either<SomeKindOfException, SuccessfulResultClass>. As mentioned earlier, the convention in the Haskell world (from
which I totally "borrowed" the Either) is that an Either gives you "either the Right answer or whatever's Left" -- that
is, errors on the left, expected output on the right.
For this reason, this Either is right-biased; if you give it Either.either(()->42, ()->"Hello, World!");, you'll get a
Right containing "Hello, World!", not a Left containing 42. I swear, it's not a political thing; there just
needs to be a predictable rule-of-thumb for how to handle it when the Either gets both a left value and a right
value (after all, it's called an Either, not a Both).
Because this errors-to-the-left, results-to-the-right idiom is so common, this library also includes a Result<T> class,
which instead of being Left<L,R> or Right<L,R> is Err<T> or Ok<T>. Instead of Either.either(() -> "left", () -> 42),
the constructor method you'll want is Result.attempt(() -> someMethodThatMightThrowAnException()). You can even chain
a series of possibly-failing functions using map:
Result<C> = Result.attempt(() -> someOperationThatMightFailOrReturnA())
.map(a -> someOtherOperationThatMightFailOReturnB(a))
.map(b -> someThirdOperationThatMightFailOrReturnC(b));If you're working with collections of Eithers (for example, you're performing a series of validations, all of which can
either succeed or return a validation error message), you'll probably be interested in EitherCollectors, with its
.toLeftBiased() and .toRightBiased() methods for use with Java Streams:
// Let's say you have this variable already defined from your validations:
// List<Either<ValidationError, ValidationSuccess>> validationResults
// And you want to fail the user's request if there's *any* error, and return *all* the errors if so. EitherCollectors can help!
Either<List<ValidationError>, List<ValidationSuccess>> errorsOrSuccesses = validationResults.stream()
.collect(EitherCollectors.toLeftBiased());
// If there's even *one* left-hand ValidationError in our list, then we'll get a Left containing *all* the ValiidationErrors.
// If not, we'll get a Right containing any ValidationSuccess results.
// So we can do this:
return errorsOrSuccesses.fold(
(List<ValidationError> errors) -> new UserFacingErrorsResponse(errors),
(List<ValidationSuccess successes) -> new SuccessResponse(successes)
);
Wanna see more? Check out the unit tests for a run-down of how Eithers behave. If those tests aren't descriptive enough, or you think they should behave differently, open a Github issue! I built this because it didn't look like anyone else had built it for Java yet, and I may have lost something in the translation. I'm totally open for feedback.
Simple! It's in Maven Central, so just add this to your pomfile (or the equivalent for gradle/sbt/lein/whatever):
<dependency>
<groupId>com.spencerwi</groupId>
<artifactId>Either.java</artifactId>
<version>2.9.0</version>
</dependency>