Understand Error Handling in Scala

What's Exception

Exception are objects that defined in Java to represent error scenarios. Unhandled exceptions could terminate the program or show default message from http server.

Excpetion hirarchy

Since Scala is designed to reuse Java standard or thirdparth library without much effort. It still worth to know the exception hirarachy originated in Java. It will help us deal with error scenarios.

• Throwable is the root of exception hirarchy.
• Exception subclasses represent errors that the program can recover from.
• Error subclasses represent errors that a program generally shouldn't expect to catch and recover from, like OutOfMemoryError.
• RuntimeException is subclass of Exception, they represent programming errors that a program shouldn't generally expect to occur, but could recover from, like NullPointerException.
• Subclasses of Exception, other than RuntimeException, represent errors that a program could generally meet, like network connection error, file read/write error.

Checked vs unchecked exceptions

Checked excpetions need to be declared in the method that throws it and must be handled by the callers, unless it will fail the compiler.

public FileInputStream(File file) throws FileNotFoundException

try {
  FileInputStream fin = new FileInputStream(file);
} catch (FileNotFoundException e) {
  e.getMessage()
}

In Java we can also keep throwing the exception without handling it.

The Errors and RuntimeExceptions (and their subclasses) are called unchecked exceptions, means that:

• they can be thrown “at any time”
• methods don't need to declare that they can throw such an exception
• callers don't have to handle them explicitly.

Exceptions in Scala

All exceptions in Scala are unchecked which means the Scala compiler doesn't require you to declare and handle the exception explicitly.

As a programming language aimed for functional programming, we should never throw exception in Scala code. It breaks the referential transparency. We have following two functions foo1 and foo2.

def foo1() = if(false) throw new Exception else 2

def foo2() = {
  val a = throw new Exception
  if (false) a  else 2
}

foo1() // res0: Int = 2
foo2() // java.lang.Exception

Calling foo1 will terminate normally, calling foo2 will throw an exception.

We should always use a FP way to handle exception in Scala. I have listed a few options below.

How to handle exception

try catch

We should avoid using try catch except calling some unsafe or Java oriented API that provided by thirdparty library. Instead of using try catch, it's recommended to use Try which will be explained in another section.

def divide(m: Int, n: Int): Int =
  m / n

val result = try {
  divide(10, 0)
} catch {
  case e: Exception => e.getMessage
}

// result is string: / by zero

Option

When you don't care about the error information, it's ok to use None along with Option monad to represent the uncared error, and use Some(value) to represent the normal calculation result of a function like following example.

def divide(m: Int, n: Int): Option[Int] =
  if (n == 0) None
  else Some(m / n)

val result = for {
  a <- divide(10, 3)
  b <- divide(7, 0)
} yield a + b

// result is None

Either

Either monad is one of the most frequently used error handling methods. Use Left(error) to represent the error include type inforamtion, Right(value) to represent the normal result.

def divide(m: Int, n: Int): Either[String, Double] =
  if (n == 0) Left("Divident can't be zero")
  else Right(m / n)

val result = for {
  a <- divide(10, 3)
  b <- divide(7, 0)
} yield a + b

// result is Left("Divident can't be zero")

Try

Try monad is another commonly used way to do error handling in Scala. It's used to dealing with some code which will throw exception out of your control like thirdparty library oriented in Java. Better than try catch block, you can get the monad control flow as a bonus. Refer to this blog to get more knowledge about monad control flow. With Try monad, you can use Failure(error) for error and Success(value) for nomal value.

def divide(m: Int, n: Int): Try[Int] =
  Try{
    m / n
  }

val result = for {
  a <- divide(10, 0)
  b <- divide(10, 3)
} yield a + b

// result is Failure(java.lang.ArithmeticException: / by zero)

Validated

Validated is the data type provided by cats. It's widely used in scenarios that need accumulating error information, like:

• input form check
• configuration check

With Validated, user can see all the error information at once, don't need to wait to see next error until fixed previous one. Use Invalid(NonEmptyList(error)) when error happens, use Valid(value) when calculation ends properly. NonEmptyList is a semigroup defined in cats to fulfill data accumulation.

def divide(m: Int, n: Int): Validated[NonEmptyList[String], Int] =
  if (n == 0) Invalid(NonEmptyList.of("one error"))
  else Valid(m / n)

val result = (divide4(5, 1), divide4(10, 1)).mapN((_, _))

// result is Invalid(NonEmptyList(one error, one error))

Monad Error

MonadError is a monad that abstracts over error handling monad like Option / Either / Try, which means we can writing business code without concerning about the specific error handling monad. In this way, our code becomes more generic and compact.

• pure: create a value with a successful result wrapped by the error handling monad.
• raiseError – create a value which represents an error.
• fromEither / fromTry / fromOption / fromValidated – convert a specific error handling monad into the related MonadError
• handleError / handleErrorWith / recover / recoverWith – handle the error scenario and return a successful value.

One example looks like following code:

trait CustomError[A] {
  def errorFromString(str: String): A
  def errorFromThrowable(e: Throwable): A
}

object CustomizeError {
  implicit val ThrowableCustomError = new CustomError[Throwable] {
    override def errorFromString(str: String): Throwable =
      new Exception(str)
    override def errorFromThrowable(e: Throwable): Throwable =
      e
  }
}

def divide[F[_], E](m: Int, n: Int)(implicit monadError: MonadError[F, E], error: CustomError[E]): F[Int] =
  if (n == 0) monadError.raiseError(error.errorFromString("Divide by 0"))
  else monadError.pure(m / n)

import cats.implicits._
val divideTry = divide[Try, Throwable](10, 0)

Error Handling with IO Monad

IO monad is commonly used while dealing with side effects provided by cats-effect. There is an instance of MonadError[IO, Throwable] in cats-effect, all the error handling is done throught it. The convenient function to handle error scenario.

• raiseError: constructs an IO which wraps up a Throwable.
• attemp: sequence the convertion between exception in IO and Either.

case class BusinessException(msg: String) extends Exception(msg)

val boom = IO.raiseError(new BusinessException("error happens"))

boom.unsafeRunSync()

boom.attempt.map{
  case Left(BusinessException(msg)) => "business error"
  case _ => "other error"
}.unsafeRunSync()

val result = IO.delay(10 / 0).flatMap(n => IO(println(n)))

Summary

Error handling is one of the most important field in programming. FP provides a more graceful way to do it comparing with OO. Review this post and contact me if you got any thoughts.

Reference

• Scala best practice: do not throw exception
• Cats data type: Validated
• The exception hierarchy in Java
• Handling exceptions in Scala
• Monad Error for the rest of us
• Cats Effect: IO