Scala Compiler Plugins

Language
This doc page is specific to features shipped in Scala 2, which have either been removed in Scala 3 or replaced by an alternative. Unless otherwise stated, all the code examples in this page assume you are using Scala 2.

Lex Spoon (2008)
Seth Tisue (2018)

Introduction

A compiler plugin is a compiler component that lives in a separate JAR file from the main compiler. The compiler can then load that plugin and gain extra functionality.

This tutorial briefly walks you through writing a plugin for the Scala compiler. It does not go into depth on how to make your plugin actually do something useful, but just shows the basics needed to write a plugin and hook it into the Scala compiler.

You can read, but you can also watch TV

The contents of this guide overlaps substantially with Seth Tisue’s talk “Scala Compiler Plugins 101” (32 minute video). Although the talk is from April 2018, nearly all of the information in it still applies (as of November 2020).

When to write a plugin

Plugins let you modify the behavior of the Scala compiler without changing the main Scala distribution. If you write a compiler plugin that contains your compiler modification, then anyone you distribute the plugin to will be able to use your modification.

You should not actually need to modify the Scala compiler very frequently, because Scala’s light, flexible syntax will frequently allow you to provide a better solution using a clever library.

There are some cases, though, where a compiler modification is the best choice even for Scala. Popular compiler plugins (as of 2018) include:

Some tasks that required a compiler plugin in very early Scala versions can now be done using macros instead; see Macros.

How it works

A compiler plugin consists of:

  • Some code that implements an additional compiler phase.
  • Some code that uses the compiler plugin API to specify when exactly this new phase should run.
  • Additional code that specifies what options the plugin accepts.
  • An XML file containing metadata about the plugin

All of this is then packaged in a JAR file.

To use the plugin, a user adds the JAR file to their compile-time classpath and enables it by invoking scalac with -Xplugin:.... (Some build tools provide shortcuts for this; see below.)

All of this will be described in more detail below.

A simple plugin, beginning to end

This section walks through writing a simple plugin.

Suppose you want to write a plugin that detects division by zero in obvious cases. For example, suppose someone compiles a silly program like this:

object Test {
  val five = 5
  val amount = five / 0
  def main(args: Array[String]): Unit = {
    println(amount)
  }
}

Our plugin will generate an error like:

Test.scala:3: error: definitely division by zero
  val amount = five / 0
                    ^

There are several steps to making the plugin. First you need to write and compile the source of the plugin itself. Here is the source code for it:

package localhost

import scala.tools.nsc
import nsc.Global
import nsc.Phase
import nsc.plugins.Plugin
import nsc.plugins.PluginComponent

class DivByZero(val global: Global) extends Plugin {
  import global._

  val name = "divbyzero"
  val description = "checks for division by zero"
  val components = List[PluginComponent](Component)

  private object Component extends PluginComponent {
    val global: DivByZero.this.global.type = DivByZero.this.global
    val runsAfter = List[String]("refchecks")
    val phaseName = DivByZero.this.name
    def newPhase(_prev: Phase) = new DivByZeroPhase(_prev)
    class DivByZeroPhase(prev: Phase) extends StdPhase(prev) {
      override def name = DivByZero.this.name
      def apply(unit: CompilationUnit): Unit = {
        for ( tree @ Apply(Select(rcvr, nme.DIV), List(Literal(Constant(0)))) <- unit.body
             if rcvr.tpe <:< definitions.IntClass.tpe)
          {
            global.reporter.error(tree.pos, "definitely division by zero")
          }
      }
    }
  }
}

There is a lot going on even with this simple plugin. Here are a few aspects of note.

  • The plugin is described by a top-level class that inherits from Plugin, takes a Global as a constructor parameter, and exports that parameter as a val named global.
  • The plugin must define one or more component objects that inherits from PluginComponent. In this case the sole component is the nested Component object. The components of a plugin are listed in the components field.
  • Each component must define newPhase method that creates the component’s sole compiler phase. That phase will be inserted just after the specified compiler phase, in this case refchecks.
  • Each phase must define a method apply that does whatever you desire on the given compilation unit. Usually this involves examining the trees within the unit and doing some transformation on the tree.
  • The pattern match inside the body of apply shows one way of detecting certain tree shapes in user code. (Quasiquotes are another way.) Apply denotes a method call, and Select denotes the “selection” of a member, such as a.b. The details of tree processing are out of scope for this document, but see “Going further”, below, for links to further documentation.

The runsAfter method gives the plugin author control over when the phase is executed. As seen above, it is expected to return a list of phase names. This makes it possible to specify multiple phase names to precede the plugin. It is also possible, but optional, to specify a runsBefore constraint of phase names that this phase should precede. And it is also possible, but again optional, to specify a runsRightAfter constraint to run immediately after a specific phase.

More information on how phase ordering is controlled can be found in the Compiler Phase and Plug-in Initialization SID. (This document was last updated in 2009, so may be outdated in some details.)

The simplest way to specify an order is to implement runsRightAfter.

That’s the plugin itself. The next thing you need to do is write a plugin descriptor for it. A plugin descriptor is a small XML file giving the name and the entry point for the plugin. In this case it should look as follows:

<plugin>
  <name>divbyzero</name>
  <classname>localhost.DivByZero</classname>
</plugin>

The name of the plugin should match what is specified in your Plugin subclass, and the classname of the plugin is the name of the Plugin subclass. All other information about your plugin is in the Plugin subclass.

Put this XML in a file named scalac-plugin.xml and then create a jar with that file plus your compiled code:

mkdir classes
scalac -d classes ExPlugin.scala
cp scalac-plugin.xml classes
(cd classes; jar cf ../divbyzero.jar .)

That’s how it works with no build tool. If you are using sbt to build your plugin, then the XML file goes in src/main/resources.

Using a plugin with scalac

Now you can use your plugin with scalac by adding the -Xplugin: option:

$ scalac -Xplugin:divbyzero.jar Test.scala
Test.scala:3: error: definitely division by zero
  val amount = five / 0
                    ^
one error found

Publishing your plugin

When you are happy with how the plugin behaves, you may wish to publish the JAR to a Maven or Ivy repository where it can be resolved by a build tool. (For testing purposes, you can also publish it to your local machine only. In sbt, this is accomplished with publishLocal.)

In most respects, compiler plugins are ordinary Scala libraries, so publishing a plugin is like publishing any library. See the Library Author Guide and/or your build tool’s documentation on publishing.

Using a plugin from sbt

To make it convenient for end users to use your plugin once it has been published, sbt provides an addCompilerPlugin method you can call in your build definition, e.g.:

addCompilerPlugin("org.divbyzero" %% "divbyzero" % "1.0")

addCompilerPlugin performs multiple actions. It adds the JAR to the classpath (the compilation classpath only, not the runtime classpath) via libraryDependencies, and it also customizes scalacOptions to enable the plugin using -Xplugin.

For more details, see Compiler Plugin Support in the sbt manual.

Using your plugin in Mill

To use a scalac compiler plugin in your Mill project, you can override the scalacPluginIvyDeps target to add your plugins dependency coordinates.

Plugin options can be specified in scalacOptions.

Example:

// build.sc
import mill._, mill.scalalib._

object foo extends ScalaModule {
  // Add the compiler plugin divbyzero in version 1.0
  def scalacPluginIvyDeps = Agg(ivy"org.divbyzero:::divbyzero:1.0")
  // Enable the `verbose` option of the divbyzero plugin
  def scalacOptions = Seq("-P:divbyzero:verbose:true")
  // other settings
  // ...
}

Please notice, that compiler plugins are typically bound to the full version of the compiler, hence you have to use the ::: (instead of normal ::) between the organization and the artifact name, to declare your dependency.

For more information about plugin usage in Mill, please refer to the Mill documentation for Scala compiler plugins.

Developing compiler plugins with an IDE

Internally, the use of path-dependent types in the Scala compiler may confuse some IDEs such as IntelliJ. Correct plugin code may sometimes be highlighted as erroneous. The IDE is usually still useful under these circumstances, but remember to take its feedback with a grain of salt. If the error highlighting is distracting, the IDE may have a setting where you can disable it.

Useful compiler options

The previous section walked you through the basics of writing, using, and installing a compiler plugin. There are several compiler options related to plugins that you should know about.

  • -Xshow-phases—show a list of all compiler phases, including ones that come from plugins.
  • -Xplugin-list—show a list of all loaded plugins.
  • -Xplugin-disable:...—disable a plugin. Whenever the compiler encounters a plugin descriptor for the named plugin, it will skip over it and not even load the associated Plugin subclass.
  • -Xplugin-require:...—require that a plugin is loaded or else abort. This is mostly useful in build scripts.
  • -Xpluginsdir—specify the directory the compiler will scan to load plugins. Again, this is mostly useful for build scripts.

The following options are not specific to writing plugins, but are frequently used by plugin writers:

  • -Xprint:—print out the compiler trees immediately after the specified phase runs.
  • -Ybrowse:—like -Xprint:, but instead of printing the trees, opens a Swing-based GUI for browsing the trees.

Adding your own options

A compiler plugin can provide command-line options to the user. All such option start with -P: followed by the name of the plugin. For example, -P:foo:bar will pass option bar to plugin foo.

To add options to your own plugin, you must do two things. First, add a processOptions method to your Plugin subclass with the following type signature:

override def processOptions(
    options: List[String],
    error: String => Unit)

The compiler will invoke this method with all options the users specifies for your plugin. For convenience, the common prefix of -P: followed by your plugin name will already be stripped from all of the options passed in.

The second thing you should do is add a help message for your plugins options. All you need to do is override the val named optionsHelp. The string you specify will be printed out as part of the compiler’s -help output. By convention, each option is printed on one line. The option itself is printed starting in column 3, and the description of the option is printed starting in column 31. Type scalac -help to make sure you got the help string looking right.

Here is a complete plugin that has an option. This plugin has no behavior other than to print out its option.

package localhost

import scala.tools.nsc
import nsc.Global
import nsc.Phase
import nsc.plugins.Plugin
import nsc.plugins.PluginComponent

class Silly(val global: Global) extends Plugin {
  import global._

  val name = "silly"
  val description = "goose"
  val components = List[PluginComponent](Component)

  var level = 1000000

  override def processOptions(options: List[String], error: String => Unit): Unit = {
    for (option <- options) {
      if (option.startsWith("level:")) {
        level = option.substring("level:".length).toInt
      } else {
        error("Option not understood: "+option)
      }
    }
  }

  override val optionsHelp: Option[String] = Some(
    "  -P:silly:level:n             set the silliness to level n")

  private object Component extends PluginComponent {
    val global: Silly.this.global.type = Silly.this.global
    val runsAfter = List[String]("refchecks");
    val phaseName = Silly.this.name
    def newPhase(_prev: Phase) = new SillyPhase(_prev)

    class SillyPhase(prev: Phase) extends StdPhase(prev) {
      override def name = Silly.this.name
      def apply(unit: CompilationUnit): Unit = {
        println("Silliness level: " + level)
      }
    }
  }
}

Going further

For the details on how to make your plugin accomplish some task, you must consult other documentation on compiler internals. Relevant documents include:

  • Symbols, Trees, and Types is the single most important reference about the data structures used inside the compiler.
  • Quasiquotes are useful for pattern matching on ASTs.
    • The syntax summary in the quasiquotes guide is a useful concordance between user-level syntax and AST node types.

It’s also useful to look at other plugins and to study existing phases within the compiler source code.

Contributors to this page: