 |
| Macro shot (Photo credit: Wikipedia) |
Scala 2.10 RC2 was released this Friday. Considering 2.9 had 3 RC releases, 2.10 final is probably quite near. And it brings some awesome features. One of them are macros
Macros
So what are macros basically? Code executed at compile time. And that's about it.
So what is so great about that? Well you can do AST modifications and stuff that gets quite into compiler-plugin territory in your regular code. That means you can do pretty advanced stuff and do abstraction with performance. Oh yeah, you can also do type checking and emit compile-time errors. Safety first kids!
Usages
SLICK uses macros(in the experimental API) to transform scala expressions into SQL. At compile time! ScalaMock uses it to provide more natural API for testing. As said, you can use it for code generation or validation at compile time. Good library design will be able to minimize boilerplate code even further now. And some people will argue that macros make scala even harder language.
Type Macros
This is the best part for me. But unfortuntely it's not implemented yet. Or at least not dcumented. There are some methods with suspisious names in the API but no useful documentation. In all presentations this is referred to as "future work" but I still have my fingers crossed it makes it into final release.
So what's the fuss? Automatically generated types. Large scale code-gen.
As in ability to programatically create types at compile time. As a consequence you can create whole classes with bunch of methods. And I already have a use case of my own. I want to make a typesafe ORM for Android that's super fast. I did YodaLib ORM while back. It uses reflection(although it's fast enough usually) and provides a Cursor that lazily wraps rows into classes. And you need to make sure by hand that your class coresponds to columns of your result set. Not very safe. I had an idea to make static-typed safe inferface for the database when I first heard about HList. You would do projection as a HList and result rows would be lazily wrapped into HLists. But using them for wrapping every row(possibly filling data in with reflection) would be a performance penalty. Not to mention a mess to implement. Now consider using a macro to generate code for wrapping. It would be no slower than accessing columns by hand. And a type macro would automatically create a case class for given projection. Heavens. I'm just waiting for official documentation on macros...this is a tempting project.
(oh yeah, it would manage your scema too, so you don't need to worry about consistency between your code and your schema)
Documentation
Here's scalamacros.org which gives some information. Also some quite useful slides. I hope now that 2.10 is in RC things stabilize, because in the milestone releases api was changing constantly. Nightly API scaladoc.... Proper documentation is apparently coming soon,
Le Code
A use case for macros, loop unrolling.
Below is a trivial sample of repetitive code.
class Manual{
def show(n: Int){
println("number "+n)
}
def code(){
show(1)
show(2)
show(3)
show(4)
show(5)
}
}
We can deal with repetition writing a loop. (Higher order function really)for( i <- 1 to 5 ) show(i)But this doesnt generate the same AST(and bytecode)!
Protip: use
scalac -Xprint:parser -Yshow-trees Manual.scalato see AST after parsing.
Sometimes(rarely!) you want to unroll the loop to produce same byte code as typing all the iterations by hand.
Macros.unroll(1,5,1)(show)With a proper unroll macro defined. I spent an hour to come up with this implementation...and then scalac started crashing on me... Is there something terrible in my code?
I gave up and went on to do useful stuff...But macros hear me! I'll be back.
import reflect.macros.{Context}
import scala.language.experimental.macros
object Macros {
def unroll(start: Int, end: Int, step: Int)(body: Int => Unit) =
macro unrollImpl
def unrollImpl(c: Context)(start: c.Expr[Int] ,end: c.Expr[Int], step: c.Expr[Int])(body: c.Expr[Int => Unit]): c.Expr[Any] = {
import c.universe._
val Literal(Constant(start_value: Int)) = start.tree
val Literal(Constant(end_value: Int)) = end.tree
val Literal(Constant(step_value: Int)) = step.tree
val invocations = Range(start_value, end_value, step_value) map { n =>
val n_exp = c.Expr(Literal(Constant(n)))
reify{
((body.splice)(n_exp.splice))
}.tree
}
c.Expr(Block(invocations:_*))
}
}
