I’ve realized that Factories are actually kind of fine, in particular when contexualized as being the equivalent of partials from the world of functionals.
IMO factory functions are totally fine – I hesitate to even give them a special name b/c functions that can return an object are not special.
However I think good use cases for Factory classes (and long-lived stateful instances of) are scarce, often being better served using other constructs.
I have never seen them used well. I expect there IS some use case out there where it makes sense but I haven’t seen it yet. So many times I’ve seen factories that can only return one type. So why did you use a factory? And a factory that returns more than one type is 50/50 to be scary.
Yeah, I went through the whole shape examples thing in school. The OOP I was taught in school was bullshit.
Make it simpler. Organizing things into classes is absolutely fine. Seven layers of abstraction is typically not fine.
Consider the following: You have a class A that has a few dependencies it needs. The dependencies B and C never change, but D will generally be different for each time the class needs to be used. You also happen to be using dependency injection in this case. You could either:
Inject the dependencies B and C for any call site where you need an instance of A and have a given D, or
Create an AFactory, which depends on B and C, having a method create with a parameter D returning A, and then inject that for all call sites where you have a given D.
This is a stripped example, but one I personally have both seen and productively used frequently at work.
In this case the AFactory could practically be renamed PartialA and be functionally the same thing.
You could also imagine a factory that returns different implementations of a given interface based on either static (B and C in the previous example) or dynamic dependencies (D in the previous example).
Now B and C cannot be replaced for the purposes of testing the component in isolation, though. The hardcoded dependency just increased the testing complexity by a factor of B * C.
OOP is great, and can be much simpler than what you’ve seen.
It’s Java culture that’s terrible. The word “Factory” is a code smell all on its own.
Just like any tool, don’t overuse it. Don’t force extra layers of abstraction.
Why is the word factory a code smell?
It means you’re compensating for the lack of optional/named parameters in your language.
I’ve realized that Factories are actually kind of fine, in particular when contexualized as being the equivalent of partials from the world of functionals.
IMO factory functions are totally fine – I hesitate to even give them a special name b/c functions that can return an object are not special.
However I think good use cases for Factory classes (and long-lived stateful instances of) are scarce, often being better served using other constructs.
I have never seen them used well. I expect there IS some use case out there where it makes sense but I haven’t seen it yet. So many times I’ve seen factories that can only return one type. So why did you use a factory? And a factory that returns more than one type is 50/50 to be scary.
Yeah, I went through the whole shape examples thing in school. The OOP I was taught in school was bullshit.
Make it simpler. Organizing things into classes is absolutely fine. Seven layers of abstraction is typically not fine.
Consider the following: You have a class A that has a few dependencies it needs. The dependencies B and C never change, but D will generally be different for each time the class needs to be used. You also happen to be using dependency injection in this case. You could either:
This is a stripped example, but one I personally have both seen and productively used frequently at work.
In this case the AFactory could practically be renamed PartialA and be functionally the same thing.
You could also imagine a factory that returns different implementations of a given interface based on either static (B and C in the previous example) or dynamic dependencies (D in the previous example).
Sounds easy to simplify:
Use one of: constructor
A(d), functiona(d), or methodd.a()to construct A’s.B and C never change, so I invoke YAGNI and hardcode them in this one and only place, abstracting them away entirely.
No factories, no dependency injection frameworks.
Now B and C cannot be replaced for the purposes of testing the component in isolation, though. The hardcoded dependency just increased the testing complexity by a factor of B * C.
That’s changing the goal posts to “not static”