Meta-issue: REPL Improvements

[allanrenucci]

From @felixmulder comment in #3176's discussion:

• If a state is thrown away, and a new state used in its stead. That should be enough to reset the REPL. So, some mutability in ReplDriver init as well as Rendering should be moved into the session.
• Evaluation should be better encapsulated, it'd be so nice if we could get away without the classloading that sort of exists outside of our model.

[allanrenucci]

@felixmulder Can you develop the second point, I am not sure to fully understand.

[felixmulder]

Evaluation of code is a side-effect of rendering the rhs of values. For instance:

scala> println("foo")

creates an object, which is eventually roughly the equivalent of this java code:

final class rw$l1 {
  public static rw$l1 module = new this()
  public static scala.Unit res0;

  private this() {
    println("foo");
    res0 = new scala.Unit();
  }
}

In rendering#L60 we make sure not to render this because when we do the classloading on rendering#L55 the class (which is an object in scala terms) is initialized and the constructor does the println.

It would be great if we could capture this effect in our model of the REPL.

[OlivierBlanvillain]

I don't really understand these two points, what has to be done here?

[felixmulder]

It needs more types and such.

Seriously though, I don't remember exactly what was going on here. But it was quite hard to reason about what would be evaluated and where.

[OlivierBlanvillain]

Hmm OK, then I think I will close it for now...