Question

Given a trait containing an object with a large-ish number of instances (100 files, each 100 lines) stored in non-lazy val

trait Repository {
  object dao {
    val a = new A
    val b = new B
    val c = new C
    ...
  }
}

what is the initialization cost? Assume that there are 100 Repository consumers, one per DAO instance (i.e. “object X extends Repository” * 100). Will object X, Y, Z consumers each incur the overhead of 100 DAO instances?

Given that Repository consumers will only ever need to reference a subset of DAO instances, I am thinking of taking the lazy initialization approach instead, but trying to see what the trade off is here between cost of lazy vs. non-lazy initialization.

If, on JVM container startup, all 100 DAO instances are initialized, then a non-lazy approach would avoid the unnecessary overhead of lazy initialization. However, if 100 DAO instances are created for each Repository consumer, better to go lazy and reduce memory usage.

Answer 1

As far as I know, objects are lazily initialised. For instance with

trait Repository {
  object dao {
    val a = { println( "aqui" ); 33 }
  }
}

Observe:

val x = new Repository {}  // pasa nada

Only now:

x.dao // pasa algo