I want to design a Scala program that accepts Scala files as parameters which can customize the execution of the program. In particular, I want to supply at runtime files that contain implementations of methods that will be invoked by the program. How can I properly depend on external files and invoke their methods dynamically at runtime? Ideally, I would also like those files to be able to depend on methods and classes in my program.
Example Scenario: I have a function that contains the line val p: Plant = Greenhouse.getPlant()
, and the Greenhouse
class with the getPlant
method is defined in one of the files that will be supplied at runtime. In that file, the method getPlant
returns a Rose
, where Rose <: Plant
and Plant
is defined in the original program. How do I achieve (or approximate) this interdependency, assuming the files are only joined at runtime and not at compile-time?
Here's how to do it using only standard Scala. The non-obvious stuff is all in GreenhouseFactory
:
package customizable
abstract class Plant
case class Rose() extends Plant
abstract class Greenhouse {
def getPlant(): Plant
}
case class GreenhouseFactory(implFilename: String) {
import reflect.runtime.currentMirror
import tools.reflect.ToolBox
val toolbox = currentMirror.mkToolBox()
import toolbox.u._
import io.Source
val fileContents = Source.fromFile(implFilename).getLines.mkString("\n")
val tree = toolbox.parse("import customizable._; " + fileContents)
val compiledCode = toolbox.compile(tree)
def make(): Greenhouse = compiledCode().asInstanceOf[Greenhouse]
}
object Main {
def main(args: Array[String]) {
val greenhouseFactory = GreenhouseFactory("external.scala")
val greenhouse = greenhouseFactory.make()
val p = greenhouse.getPlant()
println(p)
}
}
Put your override expression in external.scala
:
new Greenhouse {
override def getPlant() = new Rose()
}
The output is:
Rose()
The only tricky thing is that GreenhouseFactory
needs to prepend that import
statement to provide access to all the types and symbols needed by the external files. To make that easy, make a single package with all those things.
The compiler ToolBox
is sort of documented here. The only thing you really need to know, other than the weird imports, is that toolbox.parse
converts a string (Scala source code) into an abstract syntax tree, and toolbox.compile
converts the abstract syntax tree into a function with signature () => Any
. Since this is dynamically compiled code, you have to live with casting the Any
to the type that you expect.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With