STOP IT WITH YOUR STATE

Tuesday, December 7, 2010 :: Tagged under: pablolife culture rants essay. ⏰ 4 minutes.


Hey! Thanks for reading! Just a reminder that I wrote this some years ago, and may have much more complicated feelings about this topic than I did when I wrote it. Happy to elaborate, feel free to reach out to me! 😄


Alright forks, time for some smugness. I know, on this blog?!? But bear with me. I'm just completely frakking pissed at a number of bugs I've been chasing down, at work and at home, and after looking for the more general problem, it's all about state. ALL OF IT.

Okay, not all of it. Most of it has to do with poor abstractions at the programming language level. But the primary one is state, and mostly, I think it's about time just done gone outgrown that shit.

Java was such a rousing success in part because it was so completely idiot-proof. When you remove memory management and pointers from C++, and sandpaper a few of its edges (no header files or need to declare methods, for example), virtually everyone can program in it and not make too big a mess. While I think Java is a completely neutered language (not to mention a liability issue), I think it really didn't go far enough. No, forks, let's learn from our friends in the 1970's and 80's and just flat out remove mutable state already, since most programs don't need it and most programmers can't handle the responsibility of it. The same goes for pointers, but you already knew that.

Despite it being un-rantey, let's qualify this for a second: I'm not saying all programs should be written statelessly, or that nobody should be using pointers. And Lord knows all programmers would ideally know more than just the basics on these topics. We'll always have a need for embedded systems, operating systems, very fast media processing, and scripts we'll use only once and not want to think too hard about.

The problem comes with all the rest of the programs, that don't need instruction-level efficiency; many are still written with mutable state as the default rather than a last resort. This is like solving your problems with beatings before you try reason. Most programmers use it willy nilly, thinking they're holding firecrackers when they're really holding grenades.

Here's ultimately the reason: in today's day and age, knowing what we know and writing the programs we want to write on the hardware we have, state is an abstraction that is too easily broken for the power it gives you, and it's about time we outgrew it. Let's look at the popular programming paradigms from the abstractions they rested on, and you notice a trend:

[]4

Structured programming: this was at first met with resistance: why would you eliminate gotos? The answer was the best noun-made-command I know: goto hell. As programs got larger and larger, there came a point where there was just too much complexity associated with being able to arbitrarily jump to any lexical location in your code. Since this was often in languages with file inclusion, this meant you could jump anywhere in any file you were including from anywhere else!

What was the solution? Was it moar gotos? IDE's to help with goto-related programming? No forks, they just cut that shit out and started structured programming. This led to a great bloom in software, where we got all our delicious *nix dinosaurs we know and love even today.

But what about GIANT programs, like Word? What made that possible? Because again, at a certain scale, this paradigm got unmanageable. Namely, keeping track of all the data associated with a function got to be way too complex. Structs of structs of structs would lead to 3-star programmers that made code a vile mess to work with and debug. It was only when we decided to encapsulate data with its related function in objects that we got to Object-Oriented programming.

And that led to its own bloom of awesomeness. But really forks, have we solved software? Is it too unreasonable to think that maybe objects* rely too much on state to be as scalable as the programs we want to write these days?

(Never mind the other, non-state objections to OOP, such as modeling programs that _do_ things entirely with _nouns_, or the avoidable lame hoop- jumping called Design Patterns that objects make us use).

It sounds extreme, but you don't know the shackles you're programming with until you break free of them. Once the outputs of a function depend entirely on its inputs, it actually becomes reusable. You only have to test the outputs for every set of inputs, not the outputs for every set of inputs at every possible state the whole rest of your program can be in!

Further, once you eliminate the null value or null pointer, a function call must return "Whatever I promised," not "Whatever I promised or NULL lol go debug that."

And the sooner we do this, the sooner we don't have to sit at our debuggers stepping through reentrant code looking at register values for the exact moment the memory's head a splode.

Most of all, I'd have to find another way to be inflammatory, rantey, and use too much formatting markup.

* = Note that my beef isn't entirely with objects, per se. there are object-oriented languages that don't rely on state. Scala is the most popular that encourages functional programming, and Reia has immutable objects.

(for those curious, I'm mostly up to proverbial here with my last bug on a major ScrabbleCheat feature. I veered away from Erlang to write the UI in Ruby with an ncurses wrapper, and now that I've dabbled in a stateful solution I'm fighting memory corruption in a language without explicit memory management. Bleh.

Meanwhile, I'm hating on C++ because a) I use it at work, b) in the playtimes I've used it, I've only ever regretted it, and c) it's C++, what kind of language snob would I be if I treated it fairly?)

Thanks for the read! Disagreed? Violent agreement!? Feel free to drop me a line at , or leave a comment below! I'd love to hear from you 😄