The point of a factory is to hide the fact that you may get objects from slightly different classes. Therefore, if some of these objects need certain data and others don’t…

• Either they aren’t similar enough to be produced by the same factory. Then you need to have more than one factory and make the decision way earlier than you are currently doing.

• Or they are similar enough that the client isn’t supposed to care which one they get. Then you can’t expect the client to care about whether to send the extra data or not. It follows that they must always be passed, but the factory will sometimes ignore them without bothering the client with that implementation detail. Anything else would burden the client with the very distinction that the factory was supposed to conceal.

I handle this by using something like a Builder that contains various factories. The Builder is given hashes that contain the factories to make various types of objects, and it will look into the hash and retrieve the correct factory based on whatever criteria.

For example, I have multiple question types, and all of these can be made by some implementation of an IQuestionFactory. Most of the IQuestionFactory implementations are subclasses of BaseQuestionFactory, and they receive XML that represents one question and parse it according to the rules of the question, returning back some subclass of BaseQuestion.

Some questions need more information, such as if it is a form-filling question there might be a section of the XML that contains the questions that defines the fields that are used in the form. The Factories that make these questions implement IEnhancedQuestionFactory, and the builder will check the Factory that it retrieves from the QuestionFactoryRegistry and check to see if it implements this Interface. If it does, it passes the full XML of the file that contained the questions in addition to the individual XML that goes to the createQuestion method.

To show you how flexible this can get, some of the IQuestionFactory implementations are simply shells that contain multiple question factories, and they can look at the incoming XML and then call an internal factory and return whatever that makes. We use this when an XML file contains multiple question types.

However, it sounds like you need something more like the way we build Exercises (which are essentially Controllers wrapped around a collection of Questions), where the hash (ExerciseFactoryMap) is set up to return a default factory most of the time, but in cases where it has a specific Factory registered, it will return that. And we do have one specific Factory that will create an Exercise subclass that has an additional property, and it has the additionall capability to look at the XML and find the information that it needs to populate that property.

So, I’d agree with Killian Foth that the point of a factory is to hide the implementation details of how the object is being constructed, but there’s nothing that says you can’t combine multiple Factories in creative ways (for example, a Factory that contains other Factories that do the real work).

• The usual approach in cases like this (Command patternish) is passing additional parameters in the constructor. In your case, that means adding otherArgument to B during construction.

• Turning the logic around might open up some options. For instance moving doStuff to AParams:

  for (String arg : listOfArgs) {
      AParams p = AParams.create(arg, otherArgument));
      for (A a : listOfA) {
          p.doStuff(a); // Let AParams to the lifting depending on type of A
      }
  }
  

• Alternatively you can swallow your pride and check with instanceof in the loop and set otherArgument manually before calling doStuff.

If the above is not possible, then you have no choice except doing what you’re doing (broadening the parameter argument). Don’t overanalyze this stuff.