I have a method intended to do some simple statistics for data IO, as shown below.
def ioStatSink[T <: { def length: Int }](): Sink[T, Future[(Long, Long)]] = Sink.fold((0L, 0L))((acc, bytes) => (acc._1 + 1L, acc._2 + bytes.length))
As I want it to be able to handle different data types that have a { def length: Int }
method, I make it generic.
The problem is, this method invocation uses reflection.
As this method is called millions fo times, I don't want it to have performance issue.
I know class instantiation with reflection has performance penalty, but how about this method invocation?
(another issue about the method is, it cannot adapt to types with a method of { def length: Long }
, any suggestion to deal with this?)
Since you said you were interested in the typeclass alternative.
Here is a simple demo.
import scala.language.higherKinds
trait Sizeable[T] {
def size(t: T): Long
}
object Sizeable {
private final val _IterableOnceSizable: Sizeable[IterableOnce[_]] =
new Sizeable[IterableOnce[_]] {
override final def size(iter: IterableOnce[_]): Long = {
val knownSize = iter.knownSize
if (knownSize == -1) iter.iterator.size.toLong
else knownSize.toLong
}
}
implicit final def CollectionSizeable[C[_], T](implicit ev: C[T] <:< IterableOnce[T]): Sizeable[C[T]] =
_IterableOnceSizable.asInstanceOf[Sizeable[C[T]]]
}
object syntax {
object sizeable {
implicit class SizeableOps[T](private val sizable: T) extends AnyVal {
@inline def size(implicit ev: Sizeable[T]): Long =
ev.size(sizable)
@inline def length(implicit ev: Sizeable[T]): Long =
ev.size(sizable)
}
}
}
import syntax.sizeable._
def ioStatSink[T : Sizeable](): Sink[T, Future[(Long, Long)]] =
Sink.fold((0L, 0L))((acc, bytes) => (acc._1 + 1L, acc._2 + bytes.length))