Hacker News Reader

Comment by monero-xmr | Hacker News Reader

The hill I’ll die on is to avoid classes, everywhere. You don’t need classes. Just use functions and structs (or plain objects, etc.). OOP was a terrible mistake and mixing state with logic was a major catastrophe that poisoned a few generations of engineers. Burn the Gang of Four design patterns and scatter the ashes into the ocean.

View on HN

Yikes. That kind of all-or-nothing thinking is a red flag.

Structs and functions are great if you see yourself adding more new functions than data types in the future (the existing functions don't need to change!). OOP is great if you see yourself adding more data types (the existing classes don't need to change!).

> OOP was a terrible mistake and mixing state with logic was a major catastrophe

You're gonna have to justify that. Can't just take your word for it. Where do you store your state, if not in objects?

The struct. OOP tries to combine the struct (data) and functionality (methods) into one and it is a bad idea.

And that’s the essence of OOP. Once you grasp why that is such a colossal blunder, it’s obvious what a mistake the entire endeavor truly is.

That's how the real world is. The state and the functionality are in one thing. OOP models the real world. Are you suggesting that data should be stored in global variables instead?

You don’t lose scoping by avoiding classes. I’m not really sure what you are asking. You store all data in primitives or structs, and pass it along to functions, which mutate the data. Much simpler to reason about, comprehend, debug, test, and so on.

> You store all data in primitives or structs, and pass it along to functions

OK, so in global variables, like in C language. And you think this is better than OOP. Glad that's working for you!

Uh no. You would typically retrieve the data in a function and assign it to a variable scoped to the function. The struct or array then gets passed through a pipeline of functions to do something possibly writing the data back to disk or the db.

Even if there is some data I want to keep in memory there is no need for a global usually. In go I may just keep it in a var in a goroutine that doesn’t exit until the process does.

In go you mix state with logic (example below), which apparently is a "major catastrophe" according to some on this thread.

   type Engine struct {
       HorsePower int
   }

   func (e Engine) Start() {
       fmt.Println("Engine is starting with", e.HorsePower, "horsepower.")
   }

> OK, so in global variables, like in C language.

Not at all!

This is the main reason OO has dominated for so long. OO has somehow taken on the meaning "Anything which isn't C".

And it's not even an accurate representation of C, it's a straw man version of C. C programmers know the perils of global variables.

Here is some code:

  int myState;
  void myFunc() {

  }

If I take the above code and make it look like C:

  #ifndef MYSTUFF_H
  #define MYSTUFF_H
  int myState;
  void myFunc() {
    ...
  }
  #endif

Yuck! Global variable! disgusting!

Now I make it look like Java:

  class MyStuff {
    int myState;
    void myFunc() {
      ...
    }
  }

This is somehow "encapsulated". But start calling myFunc() at different times from different threads and see if myState behaves more like a stack variable or a global variable.

OO made it much more socially acceptable to pollute your code with this kind of "encapsulation". And OO devs decided that even this was too much of a straightjacket - some people liked C#'s take on properties, where you didn't have to work so hard writing getters and setters (further breaking encapsulation).

Real C Programmers™ actually encapsulate:

    my_state_t myFunc(my_state_t in) {..}

In your last line of code the my_state_t is really the "this" pointer. You're simulating OOP.

Yes!

Once you take away inheritance ("prefer composition"), polymorphism (better offerings elsewhere), encapsulation (not encapsulation!), message-passing (better offerings elsewhere), you're left with a small syntactic trick that really wasn't that hard to do in C.

Even Lua gives you that [1]:

> This use of a self parameter is a central point in any object-oriented language. Most OO languages have this mechanism partly hidden from the programmer, so that she does not have to declare this parameter (although she still can use the name self or this inside a method). Lua can also hide this parameter, using the colon operator.

    function Account.withdraw (self, v)
      self.balance = self.balance - v
    end

    function Account:withdraw (v)
      self.balance = self.balance - v
    end

[1] https://www.lua.org/pil/16.html

This is miniature OOP and that puts you in "major catastrophe" territory according to some on this thread.

> Are you suggesting that data should be stored in global variables instead?

According to FP, data shouldnt be stored where possible and yes, if you can, store global or top level immutable blobs and pass by reference.

> That's how the real world is.

Well, maybe not. I dont rememer where it came to me that OOP is like classical physics, where everything has to get passed around with the speed of light and FP is like quantum physics, stateless until you do the computation. I dont know how accurate this analogy is but it still fascinates me since modelling reality is enlarge what programming is.

I will join you on that hill, just make a pureish function/method that does the thing its supposed to do and job done. Although in 10% of cases OOP and inheritance is useful

OOP can be used without inheritance, using composition.

Actually all four ways are possible: with either, neither, or both.

The GoF book says to prefer composition over inheritance much of the time, in an early page of the book.

From 15 years industry experience my biggest takeaway is to ignore advice from anyone who claims their point of view applies "everywhere".

Well I have over 20 years experience so maybe you need 5 more years or so

Reaching 20 years mark here, and I kinda agree and disagree with your initial take.

Hear me out, I don't think classes should be used to mutate state, or perform actions on their own. But they are a nice way to namespace and think about components properly. It can be achieved with just structs and functions (strongly typed, for the love of all deities) but giving a container like a class has its advantages to namespace some categories of functions belonging to their own structs.

I agree that many other uses of classes (and fucking design patterns piled on top) were, and still are, a major source of pain in software development, I'm very glad to see that a lot of folks around the 15-25 years of experience realised the footguns are not worth it and adapted to write more readable "dumb" code.

> namespace

Precisely. I’m a DBRE, not SWE, but I use Python a lot, and have written some internal tooling. Classes as namespacing is wonderful.

Just namespace with folder and file structure. Done

I share your take, and I’ll even say that classes are an underused tool when applying functional principles in mixed paradigm languages. If you use a class as a struct (with value semantics), you get:

- A type

- A name for it

- A canonical way to build a value of that type

- A consistent way to reference its semantic meaning at runtime and in any static/build time/documentation format

- A singular, easily discoverable place to change it

Most "non-oop" languages like Go and Rust have all of this. They also generally have interfaces. What they don't have is inheritance, which I think is what really differentiates OOP from structs with interfaces.

Even inheritance of classes-as-structs is probably under-appreciated. It’s not quite as flexible as more general composition. But it’s not nearly as problematic as OOP inheritance, because it’s ultimately only a hierarchical composition of taxonomy and value construction semantics. It can become problematic if you apply the wrong taxonomy, but that would be true without inheritance as well.

Amen. Everything in software is about tradeoffs. The only sensible answers to software questions start with "it depends". Dogmatic shittalking of OOP, a paradigm that quite a few developers are happily using to produce successful software, only highlights a lack of perspective and an unjustified arrogance.

SOLID point. Pun intended.

Structs with functions and interfaces retains most of the good parts of OOP. I agree that inheritance seems to be more trouble than it's worth.

People create a bajillion non-deterministic little state machines in their code and act like it's no big deal.

I generally agree with you. The only time I find classes make sense is when you need a stateful imperative handle. Those cases are rare.

Correct, and even then it may be possible to just return some functions, capturing the state via closures.

State via closures can be harder to think about in my opinion

Reply on news.ycombinator.com