Releasing spores 0.4.3

Jorge Vicente Cantero

BLOG

Today we at the Scala Center are happy to announce a new release of spores, now designed to work with Java Serialization. In the August Advisory Board meeting, IBM proposed that the Scala Center work on improving the status quo of serialization in Scala. The basic idea of the proposal was to pick up work on spores to prevent users from dealing with serialization runtime errors.

Spores

Spores, a project started in 2014 by Heather Miller and Philipp Haller, took a first step towards solving this problem.

Spores provide a way to control what a closures captures, and thus hopes to make it easier to operate with them in concurrent and distributed enviornments. The key idea is to capture more type information than a standard closure, and to use this type information to have more fine-grained control over the closures’s environment.

One creates a spore like this:

import scala.spores._
val greeting = Hello
val exampleSpore = spore {
  val capturedGreeting = greeting
  (name: String) =>
    println(s"$capturedGreeting, $name!")
}

Note that anything which the function in the spore refers to is explicitly declared after spore. Variables that are not explicitly listed here result in a compilation error. This makes it clear to developers what a function captures in its environment, and allows the spores framework to check for other properties for all of the values declared in the environment, such as that each value is able to be serialized.

Though, wait a second. Java Serialization does all of its checks at runtime. If spores use type information to figure out whether or not something can be serialized at compile time, how can this work with Java Serialization?

To achieve this, previously, spores required Scala Pickling to transitively check the entire enviornment was able to be serialized, statically, at compile-time. However, we now wish to perform this check instead for Java Serialization, as Java Serialization is most often used by developers.

The transitive checker

In order to ensure that captured types only extend Serializable, and whose members are also Serializable, spores version 0.4.1 comes with a compiler plugin that complements the spores macro library. Its goal is to transitively check that every captured variable, and all of its members, are extend Serializable and are thus serializable at compile time.

Let’s see it with an example:

class NotSerializable(val o: Object) extends Serializable

Capturing an instance of NotSerializable inside a spore that is going to be serialized crashes our program with a runtime exception. Although the class NotSerializable extends indeed java.io.Serializable, the member o of type Object does not and the JVM does not know how to serialize it.

The transitive checker makes sure that examples like the above are detected at compile-time rather than runtime.

Getting started

To add spores to your project, add the following keys to your SBT build file:

libraryDependencies += "ch.epfl.scala" %% "spores" % "0.4.3"
addCompilerPlugin("ch.epfl.scala" %% "spores-serialization" % "0.4.3")

For further information on how to use it and a specification of spores, check the docs.

Next steps

This stable release is the first step towards the spores and Spark integration. In the upcoming weeks, we will work with IBM to improve the developer experience of using spores in Spark and solve reported issues.