Scala conditional compilation - scala

Question

I'm writing a Scala program and I want it to work with two version of a big library.
This big library's version 2 changes the API very slightly (only one class constructor signature has an extra parameter).
// Lib v1
class APIClass(a: String, b:Integer){
...
}
// Lib v2
class APIClass(a: String, b: Integer, c: String){
...
}
// And my code extends APIClass.. And I have no #IFDEF
class MyClass() extends APIClass("x", 1){ // <-- would be APIClass("x", 1, "y") in library v2
...
}
I really don't want to branch my code. Because then I'd need to maintain two branches, and tomorrow 3,4,..branches for tiny API changes :(
Ideally we'd have a simple preprocessor in Scala, but the idea was rejected long ago by Scala community.
A thing I don't really couldn't grasp is: can Scalameta help simulating a preprocessor in this case? I.e. parsing two source files conditionally to - say - an environmental variable known at compile time?
If not, how would you approach this real life problem?

Answer 1

1. C++ preprocessors can be used with Java/Scala if you run cpp before javac or scalac (also there is Manifold).
2. If you really want to have conditional compilation in Scala you can use macro annotation (expanding at compile time)
macros/src/main/scala/extendsAPIClass.scala
import scala.annotation.{StaticAnnotation, compileTimeOnly}
import scala.language.experimental.macros
import scala.reflect.macros.blackbox
#compileTimeOnly("enable macro paradise")
class extendsAPIClass extends StaticAnnotation {
def macroTransform(annottees: Any*): Any = macro ExtendsAPIClassMacro.impl
}
object ExtendsAPIClassMacro {
def impl(c: blackbox.Context)(annottees: c.Tree*): c.Tree = {
import c.universe._
annottees match {
case q"$mods class $tpname[..$tparams] $ctorMods(...$paramss) extends { ..$earlydefns } with ..$parents { $self => ..$stats }" :: tail =>
def updateParents(parents: Seq[Tree], args: Seq[Tree]) =
q"""${tq"APIClass"}(..$args)""" +: parents.filter { case tq"scala.AnyRef" => false; case _ => true }
val parents1 = sys.env.get("LIB_VERSION") match {
case Some("1") => updateParents(parents, Seq(q""" "x" """, q"1"))
case Some("2") => updateParents(parents, Seq(q""" "x" """, q"1", q""" "y" """))
case None => parents
}
q"""
$mods class $tpname[..$tparams] $ctorMods(...$paramss) extends { ..$earlydefns } with ..$parents1 { $self => ..$stats }
..$tail
"""
}
}
}
core/src/main/scala/MyClass.scala (if LIB_VERSION=2)
#extendsAPIClass
class MyClass
//Warning:scalac: {
// class MyClass extends APIClass("x", 1, "y") {
// def <init>() = {
// super.<init>();
// ()
// }
// };
// ()
//}
build.sbt
ThisBuild / name := "macrosdemo"
lazy val commonSettings = Seq(
scalaVersion := "2.13.2",
organization := "com.example",
version := "1.0.0",
scalacOptions ++= Seq(
"-Ymacro-debug-lite",
"-Ymacro-annotations",
),
)
lazy val macros: Project = (project in file("macros")).settings(
commonSettings,
libraryDependencies ++= Seq(
scalaOrganization.value % "scala-reflect" % scalaVersion.value,
)
)
lazy val core: Project = (project in file("core")).aggregate(macros).dependsOn(macros).settings(
commonSettings,
)
)
3. Alternatively you can use Scalameta for code generation (at the time before compile time)
build.sbt
ThisBuild / name := "scalametacodegendemo"
lazy val commonSettings = Seq(
scalaVersion := "2.13.2",
organization := "com.example",
version := "1.0.0",
)
lazy val common = project
.settings(
commonSettings,
)
lazy val in = project
.dependsOn(common)
.settings(
commonSettings,
)
lazy val out = project
.dependsOn(common)
.settings(
sourceGenerators in Compile += Def.task {
Generator.gen(
inputDir = sourceDirectory.in(in, Compile).value,
outputDir = sourceManaged.in(Compile).value
)
}.taskValue,
commonSettings,
)
project/build.sbt
libraryDependencies += "org.scalameta" %% "scalameta" % "4.3.10"
project/Generator.scala
import sbt._
object Generator {
def gen(inputDir: File, outputDir: File): Seq[File] = {
val finder: PathFinder = inputDir ** "*.scala"
for(inputFile <- finder.get) yield {
val inputStr = IO.read(inputFile)
val outputFile = outputDir / inputFile.toURI.toString.stripPrefix(inputDir.toURI.toString)
val outputStr = Transformer.transform(inputStr)
IO.write(outputFile, outputStr)
outputFile
}
}
}
project/Transformer.scala
import scala.meta._
object Transformer {
def transform(input: String): String = {
val (v1on, v2on) = sys.env.get("LIB_VERSION") match {
case Some("1") => (true, false)
case Some("2") => (false, true)
case None => (false, false)
}
var v1 = false
var v2 = false
input.tokenize.get.filter(_.text match {
case "// Lib v1" =>
v1 = true
false
case "// End Lib v1" =>
v1 = false
false
case "// Lib v2" =>
v2 = true
false
case "// End Lib v2" =>
v2 = false
false
case _ => (v1on && v1) || (v2on && v2) || (!v1 && !v2)
}).mkString("")
}
}
common/src/main/scala/com/api/APIClass.scala
package com.api
class APIClass(a: String, b: Integer, c: String)
in/src/main/scala/com/example/MyClass.scala
package com.example
import com.api.APIClass
// Lib v1
class MyClass extends APIClass("x", 1)
// End Lib v1
// Lib v2
class MyClass extends APIClass("x", 1, "y")
// End Lib v2
out/target/scala-2.13/src_managed/main/scala/com/example/MyClass.scala
(after sbt out/compile if LIB_VERSION=2)
package com.example
import com.api.APIClass
class MyClass extends APIClass("x", 1, "y")
Macro annotation to override toString of Scala function
How to merge multiple imports in scala?

Answer 2

I see some options but none if them is "conditional compilation"
you can create 2 modules in your build - they would have a shared source directory and each of them you have a source directory for code specific to it. Then you would publish 2 versions of your whole library
create 3 modules - one with your library and an abstract class/trait that it would talk to/through and 2 other with version-specific implementation of the trait
The problem is - what if you build the code against v1 and user provided v2? Or the opposite? You emitted the bytecode but JVM expects something else and it all crashes.
Virtually every time you have such compatibility breaking changes, library either refuses to update or fork. Not because you wouldn't be able to generate 2 versions - you would. Problem is in the downstream - how would your users deal with this situation. If you are writing an application you can commit to one of these. If you are writing library and you don't want to lock users to your choices... you have to publish separate version for each choice.
Theoretically you could create one project, with 2 modules, which share the same code and use different branches like #ifdef macros in C++ using Scala macros or Scalameta - but that is a disaster if you want to use IDE or publish sourcecode that your users can use in IDE. No source to look at. No way to jump to the definition's source. Disassembled byte code at best.
So the solution that you simply have separate source directories for mismatching versions is much easier to read, write and maintain in a long run.