Reputation: 189
Cannot compile even if a function that handles function arguments is passed to macro. A sample is shown below.
trait Generated[Z] {
def deserialize[A](t: A): Z
}
def from[A, Z](apl: A => Z): Generated[Z] = macro GeneratorMacro.from[A, Z]
class GeneratorMacro(val c: blackbox.Context) {
import c.universe._
def from[A: c.WeakTypeTag, Z: c.WeakTypeTag](apl: c.Expr[A => Z]): c.Expr[Generated[Z]] = {
reify {
new Generated[Z] {
def deserialize[A](t: A): Z = {
apl.splice.apply(t)
}
}
}
}
}
object Generation {
def apply(input: String): Generated[Int] = {
Generator.from[Option[String], Int] {
case Some(i) => input.toInt + i.toInt // compilation failed
case None => 0
}
}
}
An error occurs at this time.
Error: scalac: Error while emitting Generation.scala
value input
Isn't the class recompiled with the macro expanded? If recompiled with inline expansion, no compilation error should occur.
object Generation {
def apply(input: String): Generated[Int] = {
new Generated[Int] {
def deserialize(t: String): Int = {
{
case Some(i) => input.toInt + i.toInt // input should be visible
case None => 0
}.apply(t)
}
}
}
}
What is going on and how to avoid it.
Upvotes: 0
Views: 712
Reputation: 51648
The first error I have is
Warning:scalac: {
final class $anon extends App.this.Generated[Int] {
def <init>() = {
super.<init>();
()
};
def deserialize(t: Option[String]): Int = ((x0$1: Option[String]) => x0$1 match {
case (value: String)Some[String]((i @ _)) => scala.Predef.augmentString(input).toInt.+(scala.Predef.augmentString(i).toInt)
case scala.None => 0
}).apply(t)
};
new $anon()
}
Error:(6, 43) object creation impossible. Missing implementation for:
def deserialize[A](t: A): Int // inherited from trait Generated
Generator.from[Option[String], Int] {
Obviously this is because you define
reify {
new Generated[Z] {
def deserialize(t: A): Z = {
...
instead of def deserialize[A](t: A): Z
(@OlegPyzhcov pointed this out in the comments).
Regarding your error Error while emitting ...
the thing is that
{
case Some(i: String) => input.toInt + i.toInt
case None => 0
}
has type not ... => ...
i.e. Function1[..., ...]
but actually PartialFunction[..., ...]
.
Try either
object Generator {
def from[Z](apl: PartialFunction[Any, Z]): Generated[Z] = macro GeneratorMacro.from[Z]
}
class GeneratorMacro(val c: blackbox.Context) {
import c.universe._
def from[Z: c.WeakTypeTag](apl: c.Expr[Any => Z]): c.Expr[Generated[Z]] = {
reify {
new Generated[Z] {
def deserialize[A](t: A): Z = {
apl.splice.apply(t)
}
}
}
}
}
Generator.from[Int]({
case Some(i: String) => input.toInt + i.toInt
case None => 0
})
or
object Generator {
def from[Z](apl: Any => Z): Generated[Z] = macro GeneratorMacro.from[Z]
}
class GeneratorMacro(val c: blackbox.Context) {
import c.universe._
def from[Z: c.WeakTypeTag](apl: c.Expr[Any => Z]): c.Expr[Generated[Z]] = {
reify {
new Generated[Z] {
def deserialize[A](t: A): Z = {
apl.splice.apply(t)
}
}
}
}
}
Generator.from[Int](new (Any => Int) {
override def apply(x: Any): Int = x match {
case Some(i: String) => input.toInt + i.toInt
case None => 0
}
})
Upvotes: 1
Reputation: 4608
This seems to be impossible, AFAICS. The compiler creates an anonymous class for your code, but it will not capture the lexical context of the macro call inside it.
The code looks a bit like this after the lambdalift phase:
def apply(input: String): Generated[Int] = ({
new <$anon: Generated>()
}: Generated);
...
final class $anon extends Object with Generated {
def deserialize(t: Option): Int = {
... // <- your code is here !!
};
}
Of course, the code has no access to the input
variable at this place.
This might be a bug in the Scala compiler...
Upvotes: 1