And along comes a visitor

Every time I look at a visitor pattern, I wonder, why.
What is the use of this thing?

Especially, when I consider modularity, I just don't understand.

Let's say I have different objects:

Door

Engine

Wheel

Blinker

Trunk

Chair

(all of them are CarParts, of course)

When there's a cross-cutting concern which needs to be handled, the discussion arises: do you modify all the objects (say, they need a getCustomerValue, getBuildValue, getProductionTime, getSupplier, getRequiredComponents, for example, since we're using these parts in a factory and we need those), or maybe do you want all Supplier code to be in the same spot (since else knowledge of all suppliers gets scattered around, and you'd want it in a single spot)

For the last, the code would get pretty messy.
it would look something like

Suppliers.getSupplierOf(Carpart part) {

if(part instanceOf Door) {

  return "supplierOne";
}

if( part instanceOf Engine) {

   if(((Engine) part)).getCar().getModel() == "mark2")

  return "supplierTwo";

   else { ... }

}

... etc.

}

That has quite a lot of if statements (since it's a matcher-like code). It's a bit of a drag to read since you don't know the order, and determining the code for, say the Trunk is not immediately obvious.

Wouldn't it be nice if we were able to have a 

String supplier = "not found";

visit(Engine engine) {

  if(engine.getCar().getModel() == "markt2") {

    supplier = "supplierTwo";

  } else { ... }

}

The Visitor pattern allows that, but it has some drawbacks.

The visitor is a double dispatch pattern, in which you define a visitor to actually have all those methods, and the implementing classes can then do a callback on them.

This would lead to code to something like this:

Carpart part = ...

SupplierVisitor supplier = new SupplierVisitor();

part.accept(supplier);

String result = supplier.getName();

With this construct, the Carpart has no knowledge of the entire supply chain, and the supplychain knowledge is kept in one location, namely, the supplierVisitor.

Disadvantage:

There is a cost to this, however. It requires the visitor to have complete knowledge of all the carparts.

This means that it is impossible to add carparts to our model without having to modify the visitor.

And thus, this might break the desired modularity of the application: The Visitor knows the domain model, and the domain model knows the visitor.

Advantage

If these are mere interfaces, it might be a fine solution. It then allows a piece of code to work on all the carparts, without the carparts having (actual) knowledge of the differing concern.

Strategy pattern

A different approach is the strategy + repository pattern.

This works when there may be carparts added at compile time and there is no code which knows about all car parts.

For example, a module 'taillight' is added.

When using a visitor, the visitor now fails since it gets unknown objects. 

But now assume that there is a SupplierStrategy, and the taillight supplies a TaillightSupplierStrategy

Our code would then be 

Carpart part = ...

SupplierStrategy strategy = supplierStrategyRepository.find(part);

String result = strategy.getSupplier();  

The disadvantage of this is that there is no guarantee that our repository contains a strategy for each and every part!

For example, what happens when we forget to write the TaillightSupplierStrategy (or add it to the repository?)

Then our find action will fail at runtime. So we get runtime errors. When using a visitor, it will fail at compile time in a similar way.... I think?


Also:
https://dzone.com/articles/visitor-pattern-re-visited