It's just another way to say the set of all languages is distinct from the empty set. Every language has it's flaws and so there will be complaints about every language. Even if there was a perfect language people would still have different opinions on it. So basically what you posted is a true statement with no use.
that is because Brian Will understands OOP. to criticize something you first have to understand it. and if you believe he know OOP better than most of us and also procedural, why is there so much disbelieve in his statement that procedural is better? the way Java does OOP is just a special case of procedural programming. it forbids a lot and what it does not is called OOP in Java.
Yes, and for yet another level of irony, this was at least as boring and confusing as are the tutorial lessons and especially those advocat-preaching speeches of OOP. Well, literally couldn't agree more - only the limits of understanding make barrier for agreeing. This proves that to the limit of understanding and beyond, OOP sucks!
No worry, you have nothing worthy to understand from this video. Better read Gang of 4, Martin Fowler, Robert Martin and other references. His whole argumentation, especially about encapsulation (eg from: 23:04) , is extremely biased and perverse. His arguments are based on object graph examples which are badly designed in first place. If you respect reasonably SOLID and general OOP principles, you should NEVER get that kind of spaghetti object graph in first place. Then the code would be encapsulated already (SRP, DIP, ISP, LSP, demeter law, and so on...). Then you wouldn't need to encapsulate non existing spaghettis such as he's attempting to, to make his point about how "OOP is bad". And about his statement: "Abstraction = simplified complextity; abstract = hard to understand". Well why abstraction simplifies complexity is because we do NOT have to understand what is abstracted if it's abstracted, so again a perverse argument... Furthermore, abstraction is not more related to OOP than to procedural or any other paradigm. "Mistery why industry tends by far toward OOP"... Did you try developping then maintaining complex enterprise applications (changing often) in procedural, compared to OO? That wouldn't be a mistery anymore for you then... "Procedural languages are more polymorphic than OO languages", lol, special mention for that collector one... And the same goes for a lot of his arguments... That guy is a clickbait sophist misleading least experimented people...
@Clément Cazaud Have you seen the guy's video where he breaks down four real OOP examples from proponents of OOP and how they could be rewritten in procedural form? I think one of the examples in that video was from Uncle Bob (Robert Martin). To be clear, I really respect Uncle Bob and I've bought and read his Clean Code book (his chapters about comments, naming things, and code rot really convinced me to kick some bad habits while programming), but I think the guy's video brings up an interesting point on how one example of Uncle Bob's code being convoluted because it was written in OOP fashion. I'm still relatively new to programming, so my opinion probably isn't really worth much at this point.
given that requirements are always changing, a free form approach is more appropriate. If you start out with a rigid plan/structure you are in fact doing waterfall.
The biggest takeaway I get from this, is that this man really, really hates jumping around. He wants to read the function in one place. I can respect it.
Well and I don't want to read the function at all when using it in a smaller context somewhere else, what I don't have to if it is on e well named, well abstracted and I know it's tested.
@@googleuser2016 This exactly. The author of the video would have you believe that readability is just about aesthetics, while in reality it's arguably one of the most important things and we spend a lot more time reading existing code than writing it.
@@s0mbres Could you elaborate on that? I work with (old-ish C-based) rail control systems with dozens of processes and as you can consequently imagine, hundreds of source files and thousands of functions all around. Nothing makes me feel more thankful than coming across a function that doesn't try to do everything. Core dumps are easier to analyze if the stack isn't completely flat, it's easier to set breakpoints, etc.
@@rndszrvaltas It probably depends on how the code is done. Good code would be lots of small functions that do their own thing, they work, and you can ignore the ones that work. A true nightmare would be lots of small functions that are all interlinked in some bizarre way where X relies on Y, Y relies on X, both of them call Z, Z sometimes calls X or Y, etc. It's lots of possible issues and you're never really sure which one is fucking up, so you end up jumping back and forth just to figure out wtf is going on and how it's supposed to work.
right? like there's no silver bullet, there's a lot of these "oop bad functional good" or vice versa bullshit on every forums and it always strikes me as biased. there's soo many different kinds of weird real world problems that one coding style isn't enough.
Seems like too many developers have lost sight of this. The end user doesn't give two shyts about whether you used functional or OOP programming in your coding. They only care that it works...and as long as it does what difference does it make? Code it the way you are comfortable coding it instead of constantly chasing after what's in vogue right now.
Define “the job”. If it’s anything other than “make it work this week and then delete it”, these kinds of discussions are vital to the long term flexibility and maintainability of the code you write.
Transparent “working code” may be working, but it still falls on various spectrums of maintainability, flexibility to change, complexity, performance, etc. I’m tired of people using “different tools for different jobs” as a hand-wavy excuse to use shitty tools just because they feel familiar. Sure, don’t go chasing fads, but it’s worth considering whether programming really IS that difficult, or whether we just make it that way out of ignorance and inertia.
At company after company I have dealt with confusing messes of deep inheritance hierarchies, sometimes 10-15 levels deep, where each subclass adds just a couple of lines of code. The prime consideration was keeping the classification pure in a philosophical way, like we were inventorying the animal kingdom. A good portion of the developer efforts were targeted towards dealing with the structure of these classifications and not getting any actual features done. Factories that make factories. Singletons with pages of boilerplate to do one simple thing, no object ever necessary in the first place.
Spot on. Anyone who writes lots of JavaScript, Python, Ruby or Go and then dips into the C# or Java world knows exactly what this nonsense feels like. It's developers scratching their own itches of systems level thinking, trying to create the ultimate "system" to solve code complexity. Like a mini-game where they're trying to create an encyclopedia for the world. But by trying to systemamitize away complexity they create even more of it. Now you need your devs to understand dozens of different "design patterns" that supposedly decouple your program and make the codebase easier to work with. Why then are they so complex, require constant abstract thinking and make the codebase even more complicated to work with? Why does every design pattern come with the disclaimer that it doesn't actually fully decouple anything and that it's just a different form of coupling? The coupling comes from needing the cross cutting code. There is nothing we can do about that problem other than make it easier to understand for our developers. Creating all sorts of wacky service locators, inversion of control containers, factories, dependency injectors doesn't decouple anything.
Generic interfaces with only one implementation, factories that just pass the value back to the constructor, subjects and observers that only ever get called once, Massive mutators receiving several different strategies do to the same thing an if/else already does. In general web dev, a lot of "problems" we solve are so simple, and well supported by the base features of the language, that any attempt to show off how smart you are by "using patterns" is guaranteed to generate bloat.
I've known some very narcissistic programmers who write this kind of code for huge organizations like banks. Usually they have zero experience outside their language of choice from 15 years ago and think "standards" and "good practices" are gospel. They usually have zero creativity and think things like "you should use DateTime to store a year value because it's made for it" instead of using a bleeping int. That guy built a framework for every single little thing, and thought Microsoft invented marshalling. He literally could not define the term outside the context of the .Net Framework. And big companies paid him well to write overengineered crap. Of course, he constantly bragged about all this amazing stuff he'd learned 'at the top.' 😂
I come back to this video a lot. Here's some timestamps: 4:38 Definition of Terms (Procedural, Imperative, Functional) 8:00 Why does OOP dominate the industry? (Java) 15:50 What is the appeal of OOP? 17:18 The One True Way to do OOP (Bandaids) 18:08 What's wrong with OOP (Encapsulation) 20:05 Shared State (Not too different than a global variable) 21:10 Encapsulation requires direct hierarchy (Problems.) 25:41 Premature erected wall building = cool-aide man solutions (OOoooH YeaaaaH!) 26:14 When starting bad structure is worse than an absence of structure 26:44 The mind games of OOP (Unnatural data types, kingdom of nouns, Manager classes) 28:54 Stupid questions you have to ask yourself (Analysis paralysis) 29:52 Abstractions hide complexity (The princess is in another castle) 31:43 Spreading your code out unhelpfully (Increases the surface area of code) 33:19 Solution! Good procedural code: 34:24 What to do about shared state? 34:46 Parameterize! Try not to use globals. 35:15 Bundle globals you do use into a single datatype 35:48 Prefer pure functions 36:19 Use namespaces / packages / modules 37:15 Long functions are fine! Logic in sequence = code in sequence. Use "section comments" 38:55 Use nested functions. (Functions inside a function, so you know it only gets used multiple times there.) 39:50 Constrain scope of local variables (Anonymous functions, use blocks, Jai programming language) 43:32 Conclusion - liberate yourself. For those who enjoyed this, I also recommend talks from Casey Muratori, Jonathan Blow, and David Acton. Thanks for creating this Brian!
@@drygordspellweaver8761 Nope, employeE. In each of my 20yr past cases, applicants (including myself) were talking to their potential teammates about some tech details. Guess most employeRs actually don't even have a plan what the heck their devs are doing in detail... 😁
Your coding style is marxism. Communistic paradigm be like... Every variable, every method public. No private APIs. No borders between namespaces. No inheritance - every variable is common! Get rid of bourgeoisie class. Lines of the code, unite!
@@a0flj0 I find some things from programming outside IT: My favourite example are ads. Ads are nothing other than some unreliable message from some host to us clients. Normal people see products. I see that some outside API tries to change my internal state in suspicious way, so I need to project my thinking, so I will be able to handle this spam. (Sorry for my eng) PS Minutes before writing my previous comment I had watched Zizek xD
As much as I love object-oriented programming, I have to admit that you strike some good points that I haven't considered. After backing up my original code for a particular project, I deleted the whole thing (because it was a mess) and started over from scratch. I will try to implement these principles in that project in hopes of making my code more manageable.
My advice is to write procedural code inside classes and only when it really makes sense use inheritance, interfaces, design patterns and other OO stuff.
@@HumanBeingSpawn Beware, OOP fanboys always gonna try to gaslight you into thinking you are the problem after your code turns into a mess by following their advice.
@@tongobong1 I have started doing this without realizing what I was doing. My inheritance seldom goes much deeper than 2 levels and even then usually only for datatypes
@@MartinSparkes-BadDragon Objects are great for modeling business logic. When you have a domain expert telling you what he wants software to do it is great to represent his knowledge with objects.
I am a mother one of those old timers - started programming back in 1980. That was with an HP calculator with 15 lines of code and 8 memory registers. There were also the ‘optical cards’. You scribble your program onto cards by selecting various numbers, send them off to the mainframe, then a few days later back comes a printout. No keyboards. No monitors, and the printout just as likely to say ‘syntax error on card 3’ as it is to provide any meaningful result. Imagine, A code, run, test cycle measured in days! I have been through almost every ‘revolution’ there has been. Several times. These new fangled technologies come and go out of fashion on roughly a 10 year cycle, just with new TLAs. I have also been responsible for pushing a fair few of them myself as well as resigning myself to ‘here we go again’. One thing to keep in mind is the Turing Machine. Turing proved that all computers, and all languages, are equivalent. Pretty much, once they can do basic logic, then all languages have equal expressive power. Anything you can do in one, you can do in another. So arguments about procedural vs OO vs Functional are moot. I can write an OO compiler using a procedural language and vice-versa. Not efficient, perhaps, but doable. So what it comes down to, as mentioned in this video, are the practicalities. In practice, in a real team, with real people with real business problems and challenges. The efficiency of writing and maintaining the code. The efficiency of new hires getting up to speed and the risks of losing people with ‘the knowledge’. Building a system that is easy to adapt and extend that Ames the users/customers happy. This video singles out ‘excessive OO’ or ‘extreme OO’ as a bad thing - in particular encapsulation. Quell surprise. Excessive anything is a bad thing! For the comp sci students out there, just remember all these technologies are tools. To be successful in a programming career you will need to master a fair few of them. No real world problem, worth solving, can be done well with a single tool. As a wise boss said to me once, ‘this is technology. With the right tool it will go 100 times faster’. This was when I was writing my own ORM layer - not realising there was already a library to do that. Similarly, I once had 85 lines of procedural code replaced with a single line of (damn clever) SQL. Being a programmer is a bit like being a doctor. You can’t solve every problem with a scalpel. You can’t solve every problem with antibiotics. Every person is a bit different. You need years of training and a wide variety of diagnostic as well as preventative as well as curative tools, medicines, machines and hands on experience - book learning alone will not cut it. The main factor for deciding what tools to use is the Problem Domain. One thing I have noticed, time and time again, is that the ‘best’ computer code accurately reflects the Domain it is working in. I guess it is called DDD these days. From that perspective, and getting back to the video, your encapsulation level should reflect the natural encapsulations of the domain you are working in. Though this is more of a heuristic than an absolute rule. Let’s call it ‘Domain Oriented Heuristic’ programming or DOH programming for short. 👍
Shirom Makkad mmm. Well, Domain Driven Design is the thing to search for. Most custom written (let’s say in-house) systems, built to support a specific business, end up with libraries, objects and data models that reflect the nouns and verbs and datasets that the business users use themselves. Encapsulation is also reflected in the various departments and functions of the business. Where this gets really interesting, is dealing with mergers/takeovers and internal re-organisations. Over time, what happens is that the system moves towards an ‘industry standard’ model - particularly as engineers and users are hired and fired between companies but within that industry.
Well said old timer. Just left my own comment on this 45 minute rant. I beat you though. Had to carry my cards to the computer building in a wheelbarrow. Those were the days - not. You still programming?
graham287 yeah - spent quite a few years ‘managing’ and ‘enterprise architecture’ but back to programming now. Mainly Python - which I think is now my favourite language. Mainly because it ‘defaults’ the right way, most of the time and automatically deals with edge cases. So less code.
create state, create controllers, design gui. besides that you need a state that allows function for your site/app the controls to manipulate it towards the users objective and then the GUI for an end user to manipulate the controllers they have access to.
Jerome Potts I mean, this is one of the ONLY videos on UA-cam which explicitly calls out OOP as a bad paradigm. If his assertion is more true than false, then I would say this video is indeed one of the most important programming-related videos on UA-cam right now. If it’s not, then whatever, maybe it’s not that important, and then the pitchforks wouldn’t be necessary... just a thought.
This is so good. My first C++ job, 20 years ago, was writing code in corporate codebase where a genius code architect had de-crufted a horrible architecture of earlier encapsulation. Nine layers of OOP encapsulation were collapsed into a single layer in an epic refactoring. That happened shortly before I joined the company and I supported both the "old" product and the "new" product. I had started this job fresh out of procedural coding and the excesses of unsupportable "isa" / "hasa" confusions were freshly there on day one. I still like C++, I guess. But this is such a well thought out, battle scarred view. Thank you for this!
The beauty of C++ is that you can practice OOP (classes, encapsulation) etc. where it seems to fit the problem, but not where it doesn't. The problem with Java is *everything* has to be a goddamned object, whether that concept fits the problem or not. Not everything "is-a" object or "has-a" object or "is-a-kind-of" object. Some things just "is".
@@UncaAlbyGmail Except java was, is, and forever will be a million times better than C++. At least in java you can actually code a functional, logical code, whereas in C++ the best you can do is some fibonacci
@@UncaAlbyGmail I still like C++, and having OOP in places saves time. I am not sure I'd like to inherit somebody's codebase who goes through all the mistakes of OOP. I also never warmed to Java.
A calculator is a very decent program to compare differences between languages, you observed right. You don't need complexity to know the rightness, only how logically sound and consistent it is. A calculator program fills these requirements very well.
@@equinox2584 But you have to limit the input to a certain set of characters, otherwise you can end up with a code-injection vulnerability in your program.
@@badunius_code my first calculator was a graphical calculator made with tkinter, mostly to avoid messing around with eval and hardening against code injection lol
In reality though the "combination of objects and functional programming" is already the real world way people use OOP and i might argue the standard. The problems with OOP are really just strawman uses of OOP. In the real world, encapsulation and abstraction are only used if it is beneficial, like reducing complexity and improving readability. In school, it's probably the case that everything is over-engineered, but that has educational benefits too. You get to experience the friction that comes with objectifying everything
One hundred percent agree. For example, he introduces a theoretical OOP practice(not sending object references), completely admits that almost nobody follows this practice, and then spends minutes following that explaining why this practice that nobody follows is bad. Well yeah, no shit, thats why nobody does it.
Agreed. Even when in java, a lot of big projects end up having one or several "Tool" static classes which are used as a repository of procedural functions, used with parameters and all that. Because it makes sense, even if it is not ideologically perfect.
If I'm not mistaken, he acknowledged that people don't really program that way, didn't he? Anyways, I don't really care about that point. The thing you said that bothered me was the "school fail upwards by teaching you to do things that way", kind of part. I agree, but I just don't like that reality. Could schools be explicit about this and tell you that full OOP doesn't work? Shouldn't students learn something useful instead of that? If anything, I think we should push back against the literature that pretends that full OOP works.
@@MjolnirFeaw yes, those repositories suck ass, because they are usually strewn all over the code base, and the next guy is just gonna write his own "Helper" function rather than see if something already does the job. They result from not actually working on the architecture.
This is how I saw scientists program intuitively. I was an experimental physicist before changing to the industry and most programs, I have seen back then, written by physicist followed the functional programming idea. Maybe, because all of us were comming from a mathematical background. Some programs were a mess though and comments were in Portugese, German and English on different layers of code. ... Good memories. And good video.
i knoww XD Procedural coding has its name literally because its how most would learn to code if self guided. i think. usually accurate but not probably not correct.
Well, the quirky thing is that we're doing OOP when deep "in the beast of the machine" it's literally running the code in a sequence vis a vis procedural.
You must understand that every programming paradigm solves those problems, that other paradigms can't solve. And different tasks require different paradigms
I've made enough OOP mistakes that much of this resonated with me and wasn't really surprising. But the bigger take-away is much of these same problems apply to microservices. Each microservice couples it's own data storage with the service and they don't share databases. This causes all sorts of problems managing the indirectly-shared state. It is a similar problem, by trying to force a small encapsulation, problems get spread around but not made any easier.
@@Ian-eb2io That is a fundamental concept in microservices. Each service has it's own data store containing all the data it needs to use. The point is to reduce how chatty your system is, rather than call another service for the information, read it from your own data store. This prevents other issues such as deadlocks or infinite recursion. It breaks DRY, but is an intentional duplication.
@@username7763 Are you telling me each services having it's own data store is so that it CAN have access to the data? Otherwise it has to call other services to get the data for it?
I recently ported a PHP web api client into my project without thinking . It has 1300 lines of code, 8 classes 4 files. Pure unadulterated OOP straight out of College... The amount of lines performing the actual function of the application? 8 lines. Which was what was left after i pilfered it for my own use (i had a duh day not knowing php uses CURL library for webapi.)
My goodness, as a student learning OOP I found the criticism here so relatable. I spent DAYS thinking about how my project could be conceptualized into classes and which methods belong to which type. I decided it was my own lack of OO design experience but I'm so glad to learn I'm not alone and it's never possible to make perfect object-behavior encapsulation in the real world. The "matchmaking game" is absolutely real.
Fellow student here, I see inheritance useful if the main code changes a lot and the clases that inherit do not change. Otherwise, composition is useful. I find it rarely in my programming that I used OOP succesfully, but the idea that I want emphasize is that if this style of codijg is hard to implement, maintain and change for any programmer at any level, it's probably not good. And I haven't even talked about scalability...
I've been in the industry for 20+ years and thought I would share my experience. These are just tools. Imagine 2 handymen arguing that the hammer is THE tool and another claiming a screwdriver is THE tool. They go back and forth pointing out the inadequacies of the other while showing the advantages, elegance, power and more than anything personal aesthetic preference for their preferred tool. Meanwhile the thousand other handymen in town use both tools alongside each other as well as a dozen other tools in their toolbox. Out of the 2 arguing handymen one is certainly more right than the other, though no one is likely to know which. All the other handymen in town know how and when to use specialized tools and they are getting paid the same if not more than the 2 arguing. Learn what you can, try it out, use it when it applies, get paid. School is just one of the tools along the way. There are many problems with the educational system, but that's too much to cover here. What the education systems effectively do is certify you can solve problems and you can see things through. You will learn more about effective design in your first year working than you did in the 4 years to get your undergrad. The learning never stops. Your intuition will continue to develop through your entire career. Listen and extract from these hot takes but please don't take them as gospel. Mostly don't take anything as gospel, just figure out how things are useful. Enjoy your journey.
The anonymous functions denoted by the “use” keyword exist in C++ lambdas, right down to the explicit list of named function variables that can be shared with the anonymous function.
Is that what a lambda function is? I have been trying to understand lambdas for so long that it's embarrassing. The "use" idea described at 41:40 makes perfect sense to me. Is this all a lambda is?
@@1u8taheb6 Yes that is exactly what a C++ lambda is, it allows you to capture any or all variables from the enclosing scope either as references or values, and also allows you to pass in regular variables like you would to any function, you can combine these options in any way you like, they are very handy.
@@koacado Python Lambdas allow only one expression. Python lambdas are also closured which means they capture the variables of the surrounding scope. The proposed "use"-block should be like an immediately invoked function that is NOT a closure.
I return to this video periodically, every year or so, as I improve my knowledge of programming concepts, each time I understand it a little bit better :)
My first programming language was Fortran IV in 1979. For over 30 years I got paid to write COBOL code. During the time I’ve seen languages and theories come and go. I learned PL/SQL in 1997. As a procedural programmer, switching to OOP drove me nuts for many of the reasons you mentioned. All those factories and patterns made it nearly impossible to find the business logic. I retired and came back to do pl/sql again.
COBOL and Fortran are optimized for a specific area of problems - and inside their area they are fine because they are very limited - it is difficult to write code in COBOL or Fortran which is hard to understand. But imagine you have to implement a graphical user interface having a dynamic number windows containing complex dynamic layouts of widgets of different type (Checkboxes, Radiobuttons, Textedits) using COBOL or Fortran - it would be a nightmare...
reminds me of a time I was on a Java project, one part was an incomprehensible kludge and the developer responsible answered every question with "oh this is the (whizbang) design pattern, refer to (book) if you want to understand it." When I finally dove into the code, I found the synchronisation was completely broken, and that he didn't have a clue about multithreading but always managed to blame the resulting crashes on other things.
That is the main problem the OOP programmers just use 80% time to refactor and only 20% (or less) of the time to make the program how can they not take 5 times more?
I disagree, because cognitive complexity is a real thing. Our human brains can only hold so much in memory at once, so it is our goal to design a software architecture that optimizes for that. I can say definitively from my own experience that giant procedural code I wrote in my early years is far more difficult to read and comprehend. Once you study it for a while it becomes understandable, but only after a while. Code where I broke up into smaller pieces is much faster to read and comprehend, and I make fewer mistakes with code written like this.
Agree. I think more importantly, breaking code up into smaller pieces, sets things up for unit testing later on. Rather, what I got from this video is to try really hard to avoid needlessly increasing surface area thereby displacing complexity. But in procedural programming's defense, procedural programming often helps me build momentum until i need to break the code up to make it easier to unit test.
Also do not forget that, going into more of a domain driven design paradigm, eventing allows for decoupling when things get too far apart from a domain standpoint. Think about how much UI systems rely on eventing, but often still use OO concepts as well.
Yes, cognitive complexity is a real thing. But OOP does not help with it. All it does is throw the performance aspect out the window so that the problem you are trying to solve is a simpler one. Dumbing things down means you aren't solving the same problem anymore. For a given problem, if you aren't disregarding the codes performance, the problem is necessarily as complex as the problem is. You can't just willy-nilly group things up without affecting how the code runs. When you are trying to solve the more complicated version of the problem (which is not done for nothing, it is useful), thinking in OOP terms will not do anything useful.
It drives me nuts that people act like Common Lisp never existed, exclaiming "Oh, why can't I have " when it was available for the past 30+ years. Or worse -- crapping on all that Lisp provided and then embracing it as soon as someone adds curly braces instead of parentheses.
As someone relatively new to programming, I appreciate how thoroughly you managed to explain your argument in this video despite all the jargon that was involved. It also taught me more about how and when to use both procedural and OOP, so thank you for that.
Thank you so much for this. As a programmer from the days of Fortran IV through Pascal, C, Java, JavaScript etc I always felt a bit of a fraud when not writing "proper" object oriented code. I much preferred writing in a procedural way. Now I know why!!! And yes teaching OOP is a real hassle as I'm always thinking, but really I don't need to do this... 😀
Minimizing state is something I think a lot of people gradually learn over time. Function side effects make your code difficult to use if your project grows enough, function side effects can just cause an overwhelming amount of bugs.
I really disagree with people saying OOP is bad, but that's mostly because if it weren't for OOP, modding minecraft would be fuckin *impossible*, like holy hell it would suck.
@@hongkyang7107 Moddable because it's small. For bigger size ones like Minecraft it's practically impossible to do with just procedurals and even if you could, you'll need a degree on Minecraft to mod it.
While I agree with most of your arguments, what you've mostly described is just bad programming practice. Procedural programming has its own pitfalls, and an inexperienced coder will turn it into crap just the same.
Gavin Schuette neither is procedural programming. They're just two different ways of abstracting the logic so you don't have to code on the bare metal. Use whichever floats your boat.
That is a sound premise. I've always felt that C++ is one of the better languages out there because it's a Swiss Army Knife of tools. You can write pure OOP or pure Procedural or mix and match. We even have his "use a, b {...}" example in C++ with lambdas. I think it comes down to picking the right tool for the job. Don't make everything a class, and don't make everything a 64 parameter function. Use what fits the situation best, and remember bad programmers will always choose the wrong tool.
@@djpeterson7479 why couldn't they define classes for c++ too and other functionality of other languages so we could have one single language and either use it as oop or make our own universe. Why is that so hard if c++ is as flexible as you describe? Im just asking i dont know nothing about these but whoever gonna respond to this comment just dont be a dumb
procedural is not an essay ! COBOL is so damn verbose, is it still ADD X TO VARIABLE or MOVE BlahBlah TO thisOtherVariable ? Everyone else found that Variable += X or variable =variable + x is as informative without screaming in prose.
@@TricoliciSerghei I cut my teeth on SNOBOL. It's not really SNOBOL I wouldn't want to go back to (although frankly I can't remember much that far back) it's the primitive Dec MicroVax hardware I'd rather not go back to. "Hey guys, is the tape deck free?" or "is anyone running anything at the moment?" Glad we're not heading back there!
Just found this video, found it really informative and I'm glad someone pointed out that OOP isn't some magical methodology you learn and becomes default. From what I found with OOP is that most cases it's either a) Trying to be an overengineered struct, or b) Trying to be an overengineered function (or group of functions). It seems most people use them because it's normally the most advanced thing they learnt and they assume "it's what the real programmers do" (Like a kid mimicking adults). Unless you specifically need inheritance or instancing (e.g., gamedev), FP makes way more sense than OOP
There's been a shift in gamedev away from OOP because objects are slow and hard to multithread in practice and class hierarchies are simply too restrictive to express complex game logic nicely. These days many big titles are built using entity component systems which in many ways mimic relational databases, with every entity being an ID with components like position and health associated with it.
There is a other way of seeing it, and it's fairly simple : OOP is good to learn to learn how to suffer in it and know not to touch it later on, while still learning about UML stuff (which can be transposed in FP)...
This year, I took the plunge and decided to write an entire project using just functions and interfaces, in TypeScript. I had to stop and regroup a few times along the way, but the exciting thing is, the stuff I've had to learn along the way really seems more generally applicable - whereas a lot of the OO patterns and principles often seems like (as he says) a workaround or a bandaid. What you learn by committing yourself to this is much more general - simple techniques that work every time, not just for one specific problem or scenario. I have 23 years of developer experience and pushing myself to go all in on this feels like leveling up - for the first time in a long while. I was on the fence about this video in 2016. I knew there was something true to it, but I had no idea where to start. It does have something to do with the language you're using - I was using mainly PHP, where functions and interfaces aren't really practical or useful the way they are in TypeScript. I don't agree with absolutely everything in this video. Part of me wonders if the author hasn't changed his mind about some of the finer points in this video over the years - we all grow and learn, right? I kind of wish he would post an updated version of this presentation. Anyhow, I gave it my like today, 5 years later. 😄
I also don't agree with everything stated, particularly about functional programming. Computers just don't work that way so will never be efficient enough. I was a true believer of OOP as a C++ game programmer for many years. One day a friend suggested I try writing a game engine in C instead of C++. I was skeptical as I loved my fancy C++ systems and templates etc, but in the end I found my code to be much simpler, smaller and more elegant. Unfortunately I still have to work on OOP game engines and nobody will believe me that procedural is better.
I've started coding in a more procedural style i think, but i like to slap most stuff into a class anyway ... just with more emphasis on making many methods pure functions that don't touch the object's properties. And i reduce the object dependencies. I find when my code is hard to test, it's hard to think about, and it's hard to maintain. I also make most stuff public because idk when i might need to mess with things. Idk. I used to put a lot of weight into designing things perfectly for extensibility or whatever & talks like this (and moreso molly rocket) have lead me more toward just ... get it done, clean it up a little bit, and don't worry so much about design. I feel like I'm explaining myself poorly though
"Kernels of good ideas have been taken to holistic extremes." That's a very succinct way of pointing out a bad tendency people exhibit in many areas not just coding. I'm adding that to my "great quotes" list and attributing it to Brian Will.
Poo-pooing autocomplete as "groping your way through" is a nonsense argument. Programming isn't about memorizing rote syntax, and having that belief is a recipe for disaster in a codebase of any significant size.
Imo a good programmer is the one that comes out with te best and most efficient solution, not the one that memorized all syntax. Autocomplete is a really good tool that improves workflow exponentialy
Absolutely. Especially if you're modifying someone else's code or code you wrote more than a year ago, autocomplete is an absolute lifesaver. This is why it takes me 3 times as long to write JS as Java.... I have to constantly look for function names in other files that were written years ago by other people.
it seems like these "oop critics" have only ever woked on single person small scale applications where they could reinvent the wheel without it taking years - its also proably a very shitty wheel if they just came up with it themselves.
It really depends on the problem. I programmed the first ever constructed OOP-language in 1970 which was SIMULA and has since then used various OOP-languages including Object Pascal, Java, Javascript, C++ and Ruby. Some problems are really difficult to tackle without objects and for some others objects just distract from the task what you actually want to achieve. Nowadays I mostly use procedural programming but occasionally add in objects for special tasks. It just boils down to common sense when to use it. I agree to that OOP is overused and often just adds confusion when one wants to understand what the code of others actually do.
Agreed. In some domains there are natural relations between data and methods that are never going to change, so objects work well here. In others there are cross-cutting concerns, so standalone data structures and functions work better. It's not really so difficult if you take a pragmatic, undogmatic approach. All my apps end up with modules that contain objects, standalone data structures, and libraries of functions. Works for me. Seems to me that it's not objects that are the issue - they have their legitimate uses. It's dogmatic OO thinking that attempts to force every domain into the OO paradigm.
+The Foun Neither do I and according to the video it would be a waste of time figuring it out. O well, I don't even program yet. I ordered a book on structured BASIC today. when I am confident in making 3D Games in Basic, I plan to learn C. When I am able to make AI for Arduino in my sleep in C, I plan to learn Varilog. When I am able to synthesis instincts for robotics in Varilog, I plan to learn Assembly. When I can wright a kernel in Assembly, Then I will consider myself a programmer. It all begins with Structured Basic and game programming.
Most programmers are perfectly capable to make a mess of any project without oop, good programmers can make a gem with any style of programming. Must admit that oop can rely boost the clusterfuckery of a project.
I like the name spacing iOOP provides but I don't think you should be hiding internals of an object unless you actually need to. Other developers just hide things by default and cringe when they are forced to add accessors to their private variables. :/ I go the other way around untill I see abuse I'll keep accessors to private variables and remove them if they start being abused.
@@carlosgarza31 when you expose variables or methods they become part of your public API. removing them in big projects later when you discover that people abuse them (probably because you have exposed something you never intended to be used by someone else than you), then you have to mark those variables and methods as deprecated, release a new version and give time (depending on the size of the project or the user base that process might take years), and then finally you can mark them private to hide them again. information hiding is one of the key principles of oop and a powerful tool in the hands of the right programmer.
@Minori Housaki Not necessarily. It's the difference between `queue_push(q, elem)` and `q.push(elem)`, or even `queue.push(q.elem)` if you put your queue handling functions in a module.
As a Software Developer with over 20 years in, having developed in many different languages using multiple methodologies and patterns, this was an interesting presentation, Thank You Brian! The pitfalls you point out and complexities are absolutely real. While I don't think that falling back to the 90's is the answer, as you seem to insinuate, your argument points to something very important, which is too keep architectural and design concerns at the forefront as you write and modify code. "Bolting on" without understanding the design of a program is a fast way to create spaghetti, not matter functional or OO. I find that good program layering and being familiar with Design Patterns is crucial, and helps avoid some if not all of the pitfalls. If I am a doctor doing surgery, I better know how the organs (objects) are laid out and connected (patterns). Just knowing how everything functions is not sufficient in a complex system. As far as Agile goes, that is not a coding "thing", it is a shift in the way we think about building and maintaining things, but should also involve many non-coders in order to correctly set expectations about what gets done when. I will be watching some more of your content. There are some great titles, and you obviously have a lot of real world experience to draw from.
You're right to bet in design and architecture, but it doesn't change the fact that once in implementation you'll face the dilemma for full/partial encapsulation as he mentions around 25:00 - full encapsulation results in lots of additional classes created to help you deal with state (design patterns). Those classes are not related to the problem being solved. They increase complexity because it doesn't make the code any easier to read and leads to poor performance - partial encapsulation results in spaghetti code.
@@visitante-pc5zc you are certainly right. I deal with this in two ways. 1. I'm aware of most standard pattern classes and their functions. They are just part of the world I live in 2. A LOT of those classes are relegated to the framework and work magically.
1) OO was proposed as a solution to problems inherent in functional, top-down design. It was first implemented in languages like Simula (1960s) / Smalltalk (1972), Python (1991). Java wasn’t created until 1996, OO was well established by then. OO was adopted after struggling with many problems in industry, having a long theoretical debate, then doing some experimentation. 2) Brian fails to mention the problems with functional / procedural design. Top level interfaces become very fragile without encapsulation, Iterative development often leads to repeated refactoring of the higher level interfaces. Reuse within programs is common, having every consumer maintain its state for an interface adds complexity and will become unwieldy at scale. It's also simply unnecessary. And no, making the shared state global is not a good solution. 3) Test driven development is not consistent with OO. Test driven development was pretty much a fad imposed by managers. It is just design by interface applied to testing with some other rituals built in, it has more in common with functional design than OO. It suffers from the same shortcomings. 4) Brians biggest mistake is he pitches functional and OO as opposite one another. OO is not about data modeling, or data driven design, or modeling real life objects. The early theoreticians overemphasized the data modeling aspect of OO because they were contrasting it with the prevailing design style at the time, functional design. Had data modeling been popular instead, OO would have been described in terms of designing for behaviors. OO is simply the bundling together of functions and data, with some support for minimizing duplication and access. That ideal balance of functional and data-oriented styles you keep alluding to - that is a good use of OO. OO is also totally compatible with FRP, infact FRP is used alongside objects in most code implementations of FRP. It actually compliments OO very well as it solves most of the state management problems. And FRP is an entirely separate paradigm from functional / procedural programming that dominated in the 60s - 70s. 5) Your object graph is just an undirected mess. You claim the only alternative is a hierarchical structure, this is simply not correct. You can maintain a cyclical object graph and trigger update cycles... you can have mid-level objects that manage lifecycles, and then have an object-graph of them at scale, or a hierarchical structure of them. You can come up with bad designs for anything, this is not a criticism of the paradigm. 6) refactoring code into smaller components is a functional technique, I'm not sure where you got the association with OO. Many functional programmers advocate super-small functions that confuse people ... I've never seen OO designers advocate using tons of super small classes. That's just bad design, the fault of that is on the programmer, not the paradigm. The programmer has to choose the right granularity. 7) Inheritance is used all over the place, there's just a good and bad way to use it. There is not some uniform consensus it should be avoided. When you use inheritance you just have to make sure that the base class won't be torn in different directions. Usually this means using it on smaller components without complex responsibilities. 8) OO is meant for large projects where architectural issues matter. Java was designed specifically for this purpose. If your application is 3000 lines, you might not need OO. You might be able to get away with using global variables at that scale (though if your application grows much larger you're pushing it). In an enterprise scale codebase where you have 10 million lines, no architect would ever design for reliance on global state, this would only result in complete chaos. You exaggerate the flaws of OO without providing a real alternative, and don't address the scalability problems in procedural programs at all, hardly.
@@ChristopherGray00 I actually just like listening to people talk about how something sucks which I know nothing about. rn I do know what OOP is about and its usecases and use it myself when needed
I think the video could be more digestible if you didn't sensationalize it with "This is the most important programming video you will ever watch". Not that I am one to talk though.
Hey, just want to thank you for this video. This explained something really well that I’ve noticed from trying to grapple with existing codebases at work and have intuited as a kind of encumbrance. Chasing classes down gives me a headache - when I want to understand what code does ideally I want it as sequential as is sensible so that I can follow the changes of values and potentially interrogate by pulling values out at different stages. Having classes built on classes built on classes all the way down means that often I get to the functionality I want but then can’t find weird the values get into that piece of functionality in the first place as whilst I’ve followed the hierarchy down one leg I now need to follow it up another! I’m a Data Engineer but have come into it without a degree in any computing field, and many of the things you’ve mentioned in this video I have come to on my own. Such as passing global variables into functions explicitly not just calling them for the sake of it. I totally agree that there is a place for objects/classes for data structures makes sense but not making all functionality Objects. DataFrames are excellent for my work and what they do makes sense, but that doesn’t mean that style of structure makes sense for everything.
Here’s a fun project: write your own Flattenizer. Scan through a code base and inline all the scattered logic into one location as if it were procedural.
Your description of modelling real world things with OOP makes me think about a large corporation and all the pointless layers of middle management. Its like we found a way to replicate *that* relationship in our code.
@@yousefabdelmonem3788 Well yeah. Furthermore, those necessary evils are a fundamental problems arising from scale. As the problems become bigger and harder to solve by one person, you need to make bigger and bigger compromises. It's also useful to remember when criticising governments as "less optimal" than corporations. They only have more clutter because they have harder problems to solve. Same goes for big code.
@@ekki1993 No. Thats not a fair comparison with the governments. Some governments dont even use version control for their engineers when they make their software. People dont say governments are less optimal because of their messy solutions, people say governments are less optimal because bad managers are even harder to fire, and bad employees are the only ones that flex their worker rights. Its just inherently bad.
I'd say MOST good java code is actually written in the style above. The things which are encapsulated in classes tend to be a) various functions which are logically grouped together and b) configuration (which is a bit like global variables as it pointed out in the video ). The reason for this is that most server side codes needs to be entirely re-entrant, so storing state (other than configuration ), is a no no. State ends up in databases (of various forms ). The only exception to this is various caching type functionality.
I had second hand experience with Java web developers back in my early day as a software engineer. I saw way too many static class methods to the point I thought to myself: Are they even doing OOP? Well I was still an OOP fanatic at that time -- my foolish younger self.
Functional programming has also its clear limits when you need mutable variables/objects/functionality because that is available only in run time: for instance user interface devices and dynamic inputs/outputs. In this case it is better and easier to code the business with dynamic objects than with monads.
@@user26912 Clojure/FP makes things much slower and trashes memory so you have to ask yourself why are you using multithreading parallelism at all. It's done only for performance and if you waste this on clean code i don't understand. Then write just zero shared data multitasking.
@TheJacrespo Uhhh- no? If your buffer is 256 bytes long and the user inputs 300 bytes, just increase the size of the buffer at runtime. This is how it is ALWAYS done- the difference is that OOP incurs a heavy performance cost by doing that stuff under the hood to account for all general cases. The garbage collector for example…
In the end, it all comes down to error handling. Monads do this at the expense of performance and cognitive overhead. I've always preferred simpler pragmatic approaches, and they seem to get the job done.
Object oriented programming has some good ideas, which applied intelligently can help produce good code. But when those ideas are reduced to formulas and applied blindly, you can get some very bad code. And the latter seems to be the norm for much of the industry.
@@justgame5508 That's one of the reasons why I like C++, it's the whole toolbox of paradigms. Even if there are a couple weird dangerous tools in there that can prick your fingers (i.e. placement new), it's still nice to have them.
The biggest problem I have with this video is that it completely forgoes one crucial aspect of development - testing. Not having modular components and having "God functions" that do everything in one place makes your software really hard to test, you can't really test individual bits of logic and it will become a nightmare to debug your functions if you need to make even tiny changes. Besides that this obviously introduces a very high barrier of entry for understanding your code, there is so much that a human brain can process and keep track of while reading and grasping the full picture will take much longer than if the same logic was split into separate components, and in many cases the ability for your colleagues to quickly understand what the code does without delving too deep into implementation details is really important, they in general would have other shit to do and spending an hour to understand what a function does is not a good use of their time
Working on large, highly complex projects, I've found the opposite to be true. Human brains struggle with interlocking complexity, not cardinality. We use the functional core imperative shell apporach -- even a non-technical person can open the "shell" procedure, read it top to bottom and understand what it does, hell they might even be able to make minor edits. Uncle Bob-style OOP does the opposite -- it fragments business logic into as many small, individually meaningless units as possible, making it much harder to build out a mental model in your head. Our code practically has a UML sequence diagram built-in, and our engineers love it.
@@dukiwave Its funny because I've also worked on large and complex projects, and modular code with broken down function points is absolutely crucial if you ever want to extend or maintain your code. Having smaller function blocks allows fast isolation to problematic code. Not being able to form a mental model is more of a skill issue and can be overcome with time, code comments and documentations.
@fetherfulbiped No. Nononononono no. You have it COMPLETELY backwards. It is infinitely easier to debug procedural code and reason about it when it is right there inlined for you. And figuring out a codebase is made HARDER by fragmentation it into 10,000 little pieces scattered everywhere. Your coworker can just as easily use a FUNCTION over some class with a built in method.
41:55 C++ has a solution for this: lambda functions, so instead of doing "use x, y { ... }", you do "[x, y] { ... }". It won't have access to any other variable than x and y. Edit: and it does exactly what you want, it makes clones of those variables, and if you really want to change them, you can pass them as reference for ex, with &x. It's really like having a function but without having to name one, and it can be placed anywhere.
@@jukit3906 SO, it emulate a function with args passed by Value.. great note : capture can also do references. Still the same functionality as a named function, but the named function can be shared more , re-used more and syntax is cryptic ..
@@moestietabarnak nope, it is more useful: it is dynamic (that is, a function can hold lambdas with different captures and function bodies) Only virtual member functions are comparables.
With C you can shoot yourself in the foot. With C++ you can reuse the bullet. With Java you would have a Bullet factory. With C# you would have a bullet that would query the subject to harm itself.
I appreciate your insight, this has allowed me to view my code from another perspective. Thank you for taking your time to express and share these ideas.
I propose the idea that, in programming, there is no good to be found through inconsistency. Great job, Brian. I was in the business for over 35 years, and when OOP was first announced and I got my first taste of it, my thinking swirled around the realization that in the attempt to add clarity, OOP would add layers of complexity that have nothing to do with code functionality. I'm grateful to hear someone else take OOP apart on this basis. Thank you for this presentation.
Domain Driven Design (DDD) addresses--and in my opinion eradicates--all the issues described. Now DDD is language agnostic, although Evans himself points out that OO languages are usually better suited for it. I think the issue is that most people confuse the paradigm of OOP (which has matured over the years) with Java's implementation of it. Now first I want to say I think this video makes excellent points about some of the shortcomings of old OOP paradigms, although I think the title of this video should be "Strict Encapsulation is Bad." There are, however, some points I think need to be addressed. (I've addressed test driven design in another comment). 1.) The analogy of the over-architected house is (pretty much by definition) describing a poorly planned application. He mentions walls that were prematurely established, but this is a planning flaw and not directly related to OOP. I could argue that procedural code is like a winding country road, it just randomly weaves and has no coherent sense of an actual destination. This is not a flaw with procedural programming, this is just poorly planned code. 2.) He states early on that he will ignore inheritance--rightly so--because even OO programmers warn against it. He then later references complex hierarchy graphs, which are due to inheritance. The modern approach to OOP is much more focused on dependency injection and interfaces. As far as complexity goes, the relations between objects in well designed software are only as complex as the domain requires. Any other complexity is almost certainly due to poor design choices and not the paradigm itself. 3.) "In procedural code there is only the call graph." I think this is very misleading because, in a complex domain, the call graph will only be decipherable by the software designers/engineers themselves. It is very unlikely that an "outsider" would understand the call graph outside the context of the software itself. This is why OOP is such a great tool for complex business domains. I worked on a healthcare application and one of the most valuable aspects of our design was that we could use the same graphs and flowcharts our developers were using to discuss the application with doctors (the domain experts). There was no intermediate translation needed to describe the business logic (interactions between objects) to the people who were actually in charge of implementing it (the developers). This is the point behind DDD's ubiquitous language. 4.) There definitely is a need for functionality that lives outside of objects. Anonymous functions now exist in most languages, as mentioned in many other comments. I would only like to add that creating a "container" for these functions is very much like a namespace and can actually be helpful in describing the context of certain functions and grouping them together in ways that are self-documenting. 5.) Abstraction is a term that has a specific meaning in computer science: it refers to the level at which the actual inner-workings of the computer itself are hidden from the client/programmer. For example, even a simple expression like "int y = x + 5;" is an abstraction that hides some of the implementation details (allocate a memory slot for an integer, add the value in the x memory slot with 5 and store it in the allocated memory). The abstraction of OOP is described properly in this video as "simplified interface over complex inner workings." But this is a GREAT thing! This means once an algorithm (business logic) is perfected (using test driven design) it can then be abstracted to an easier to understand concept. I'll concede that factory--and builder to a lesser extent--patterns can be sometimes hard to understand, but they are actually not meant to abstract away complexity; rather they are meant to enforce constraints. This may be unneeded complexity in small, data-driven domains with a small development team, but it is very helpful in larger teams dealing with complex domains as these patterns enforce constraints with public interfaces while hiding implementation details that only matter to the people working in that subdomain. This is similar to a business that contains forms for certain requests. The person filling out the form does not need to understand the details about how the information on the form is used, they just need to be able to fill out the form. Now, as mentioned in the video, it is definitely difficult to abstract away implementation details to the point that it can be described using real world concepts, but this is just the difficult nature of software design, not a shortcoming of OOP. This is analogous to saying it's hard to abstract away the mathematical complexity of structural mechanics, and therefore CAD software should just have the user input the center of mass and moment of inertia themselves. Abstractions and generality are probably the hardest things to master in computer science, but they are also extremely valuable when done correctly. 6.) Modularization is a great thing no matter what paradigm you use. Setting boundaries between responsibilities of your code minimizes the domino effect a change can have. The "fractionality" mentioned is actually a benefit in the long run. If you contain a big chunk of seemingly related functionality all in one place (rather than split it into smaller pieces), it becomes more likely that a small change in that big piece will bring the whole structure down. This idea of strict separation of responsibility is so fundamental to good design that you now have some of the greatest minds in software architecture (Martin Fowler, Chris Richardson, etc) focused on perfecting the microservice pattern. 7.) Lastly, "The kingdom of nouns" is no better than the kingdom of verbs that is procedural programming. The idea behind objects is that most verbs are attached to a noun--a doer if you will--or at least relate two objects together. For example, take the question posed in the video "should a message send itself?" No, a message service should send it, which describes the real world perfectly (be that the postal service, an email server, etc). I don't see what's so abstract or hard to understand about that. So the idea that OOP doesn't have verbs is utterly false, but I could argue that nouns are better described in OOP than they are in procedural code. Now I will concede that OOP is not always the right choice and I think this video addresses the issue of people always using OOP because that's what Java was designed to implement. It is often overkill for simple data-driven applications. But OOP is absolutely necessary for complex business domains. When the application becomes very difficult to reason about using basic control flow, it helps to construct objects that represent actual domain entities and then hide all the programming minutiae inside methods that represent actual business logic. It is also very beneficial when working with a large team to define public interfaces for objects and then delegate management of more detailed functionality to smaller teams that are able to master the concepts of their specific subdomain. This is similar to how every major business operates: your sales team is not responsible for accounting. They may at times need to interact with the accounting team, but there is no need for them to understand every detail of how the accounting department operates. The problem with procedural code when it comes to complex domains is that the entire team typically needs to understand the codebase in more detail. Most of the points in this video seem to pertain to a small codebase with functionality that is easily described using the basic control flow operations that have been around almost as long as computers. But this simplicity of design often breaks down once more complex logic is needed. Again, I want to say that this video makes some very good points about possible pitfalls in using OOP, but there are ways to avoid these pitfalls and it all comes down to taking the time to understand the problem (or domain) in great detail and carefully planning the structure of the application. If you don't want to put forth the effort to properly design the application then OOP will certainly fail you in all the ways mentioned in this video. If you're the kind of developer who gets an idea and wants to sit down and start coding immediately then OOP is not for you. But if you decide to take the time to apply an iterative, test-driven design process then OOP will reward you with a self-documenting, easy-to-understand, maintainable, and extensible codebase.
I also rue the epidemic incessant resistance to objective domain analysis that is occurring pandemically throughout humanity. You can see it in other areas of life too. Indeed more so! Money supply, politics, history, "education", mass media and blind acceptances of sinister, ever growing bureaucracies are the first examples that come to mind outside of engineering.
I feel like you missed some of the points. The problem with OOP is that most verbs are meant to attach to a noun. In the message example, a new coder doesn't know which class to check when it comes to finding out where messages are sent. Imagine you have the following classes: Message, Connection, MessageReader the new coder has no idea where to look exactly and any public methods in these classes will make them paranoid about side effects. In procedural you can still have namespaces/static classes/whatever to organize where the functions live. So now imagine you just have a "Messages" namespace. All actual business logic for messages is contained in this class. It has a function like "SendMessage(Connection, MessageReader, Message)". Bam you instantly know exactly what it's doing just from the parameters it takes. That's the FREEDOM of procedural. With Nouns (usually data) and verbs(business logic) you can now arbitrarily combine the nouns together with arbitrary verbs! Also if you like modularity pure Functions are THE MOST MODULAR thing in existence. In the above example SendMessage is pretty much a stand alone function (though not pure). But compared to OOP it way more modular. Someone could have programmed this so that it has a constructor Message(Connection). Now EVERY SINGLE METHOD in Message, might be using connection who the hell knows? That aint modular. Worse case it could be EDITING STATE in Connection. In the procedural example you know only that one SendMessage function could possibly be editing Connection. Lastly you can get all the same benefits you brought up with things like "business people readable flow charts" and have easy to follow call-graphs if you just name space and limit scope properly. Which is EASIER to do without OOP than with OOP because of the annoying noun.verb limitations.
@@sharp7j I don't understand how this is better than something like MessageService.Send(message) And as far as modularity goes, then you have a lot of options, like MessageService.SetConnection(connection).SendMessage(message) MessageService.SendTo(message, connection) MessageService.Send(message).To(connection) I'm also unclear as to the advantage of namespaces over static classes. And to be clear, I'm not an OOP fanatic. I use different paradigms in different scenarios and often find procedural to be the right choice for smaller applications. But it's rather dogmatic to argue that procedural code is always better than OOP. In large-scale applications, I think the encapsulation and expressiveness of OOP can be extremely valuable. The price you pay for this, as this video lays out, is that you have to be careful how you use it. But this video just lays out a bunch of problems that have been addressed (again, Domain Driven Design is a great book on this topic) and then asserts that procedural code must be the answer because OOP has flaws. I love programming in plain old C, but it's not the right choice for every project. I love messing around with functional languages like Haskell, but again it's not the right paradigm for every problem. OOP isn't the only way to code, but it is certainly a valid paradigm which is why Microsoft continues to put a lot of effort into maintaining C# (which is what a good OOP language looks like).
@@sharp7j Your example here is really nonsensical. Speaking just from an object modeling perspective, you've reversed the relationships between Message and Connection. Connections send Messages. Messages don't send themselves. Seeing Connection in a constructor for Message is an immediate code smell. If I ran across that in a code review I'd know immediately the developer didn't understand what they were doing. Whether your favor procedural code or OOP, single responsibility principals still apply. It's just as much a mistake for SendMessage to modify Connection's state as it is for Message to do so. I'm likewise perplexed where you get the notion that a procedural approach would give you a guarantee that only a given function could be responsible for modifying the state of a data structure since that structure could be passed to any number of functions any number of times. You'd have the same problem with each call to SendMessage as you would to each new Message the Connection was passed to. All you've demonstrated is that poor engineering can be carried out by anyone.
I watched the entire video and while it made me think I still can’t agree with the premise. To me a well thought out and designed OO program is easy to work on (whether you were the original programmer or not) and to extend. In my experience, the nightmares arise from programs that started as one thing and grew to something else or worse they were extended by a programmer who took shortcuts and no longer adhered to the original OO design. I’ve worked with both well designed OO and procedural programs and in my experience I have always found it so much easier and enjoyable if it’s OO.
What I like to do, is to basically fuck around and make some prototype , and then if I have the time, I draw up some random chart on how stuff is supposed to relate to eachother, and then organize it like that. It helps to be off a few adderal pills while at it. Im programming for fun, so i dont really know any of these industry level knowledge. Maybe people do that.
By that logic alone however, OO is difficult to extend because adhering to the original design plan while extending it in a meaningful way is exceedingly difficult.
@@CyrilCommando good luck changing thousands of lines of procedural code to a new design when it relies on the whole sequence of operations being maintained in an arbitrary order, otherwise your global state goes out of sync. What you'll have to do is rewrite the whole program.
Actually C++11 has the `use` feature that you show. In C++ lambdas doesn't capture (see) all variables from enclosing scope, you need to declare them explicitly in capture list. Instead of writing `use a, b { }` you end up writing `[a, b] {...}()`. Example: `const int a = 2, b = 3; const int c = [a, b] { return a + b; }();`
@@pcfreak1992 I missed that someone already said this and re-posted >.< My reply is a little more long winded though. I verified that Java and Ruby can't do anything close to what you can express in C++.
I would say it is difficult to write large systems that are well structured no matter what language is used. There is always 12 ways to do something and 10 of them are not good choices. Having designed and implemented and programmed in many Common Lisp systems, I've always found everything newer to be worse and disappointing, and it is satisfying that functional+object programming is now finally getting the respect it deserves. Javascript seems the closest to what we had in the 1980's..which is both disturbing and hopeful. I almost wonder if the time is right to bring back Common Lisp, or the simpler subset like NiCL that I created. The last 10 years all I do is write microprocessor apps using C to run in chips that have memory sized in kB. It is refreshing to be able to see the entire app in a few pages of code. I don't think Common Lisp would work in this environment or I would have tried to get it working. I see the chips have gotten big enough that they are running micropython these days...never tried it.
I really needed to see this. It's like I'm looking back at myself of the past, and this was always the obvious point I was supposed to reach but I somehow got lost along the way. This is what programming should have always been about for me but it never was, and it made me displaced because I was always looking at it from this artistic angle, but the pieces of the puzzle just never truly fit together, while nobody else ever gave a damn. That tree, the need to send a message up the nodes to send it back down, I felt that hard, it's the kind of minor thing that leaves me stuck for days working on until it kills the passion for what I'm working on and effectively kills the thing I'm working on. The way you talk about documenting your code, organizing things, it's like staring into a mirror. I wouldn't even know where to begin, but I would love to talk to you more about this, or even better, have some group pertaining exclusively to this kind of discussion, I cannot put into words how refreshing this would be to me. You did an excellent thing bringing light into this topic, I'm thankful.
Huzzah, thank you for writing this comment because now I don't have to. I feel exactly the same way in every particular. Fortunately for me, I am now retired and have taken up woodworking. Now I agonize over whether to use a hardening oil finish, shellac or water-based urethane.
If you haven't already seen it, you may appreciate Rob Pike's "Public Static Void" speech. Mr. Pike was one of the designers of Go, a language that Brian brings up in this video a few times himself (and does so in a positive manner).
Yep! Started out with Java in school and its good until you get out of the basic algorithms and data structures in isolation. Started making Android apps cause I figured that's how its done since I already know Java and wow, did I have this same frustrating experience to the point that I quit! I never much bothered to look into how people felt about these things and would sometimes read an article or two from people who should rather be designers with a keyboard instead of pencils, the web developer types, and would never really get anything concrete, very vague things that, at the time, I assumed I did not quite 'get' because they were simply better than me. Now that I am in a better position, it seems so surprising to me just how much contet surrounding this industry on the internet is utter pollution advising you to do things in a way that makes other things get in your way than get the job done!
Same here. I realized the importance of refactoring code but what killed my passion was thinking that everything had to fit together into labels and that it’s useful to do one thing in a million places for the future. I guess in any situation, one shouldn’t do X just for X’s sake
I’ve spent 15 years at a fortune 200 company where we have almost always been writing procedural code in OOP languages, mostly Java. We generally feel shame for it, and occasionally an overachiever builds something with “proper” OOP. Every time it ends up being a mess and more trouble than it’s worth. This has helped me finally see why that has been the case.
@@pyhead9916 I think OC and OP are both under the impression that writing with objects automatically makes you a good programmer. Hopefully this is obviously wrong because a pile of crap is still a pile of crap no matter what angle you look at it. If the object collection has weird hierarchical madness OO is not to blame, uness implemented so poorly in the language itself, but instead simply is a bad programmer/team.
@@Dovenchiko I agree. Yes, using OO doesn't make you make good programs magically. I think the authour is a good programmer, and as good programmer, doesn't really understand the problem of bad programmers. He is always comparing good procedural programs with good oo programs, and he has some really valid points. But the fact is business wants/needs a replaceable workforce and that causes a high rotation, so you are going to have some bad apples sometimes. And OO really excels limiting the amount of damage a bad programmer can do. I think every criticism the author makes about OO is correct, but he doesn't value enough its advantages, especially the management ones, and also dismisses the problems of procedural programming, i think because he is good at it so he can manage that. And finally, no two programmers are equal: some understand better (and consequently, are better at) some paradigms, an some other understand better another paradigms.
@@unformedvoid2223 No tool or technique prevents damage, but modularization makes the damage contained. OO forces that modularity, and that's why it is the popular choice for business. That was what i was pointing out. Remember that the 'best' language can be one from the point of view of the programmer, but another one from the point of view of the business.
1) in my university they tought us, that classes called -manager, -service, -handler, x-er and so on are considered to be bad for the reasons that you mentioned. They are actually bad practice and no contradiction to the use of the oop paradigm. 2) the "use" keyword that you suggested is basically the semantic of lambda expressions in pure functional languages like Haskell 3)even though you tried to give ideas, how to solve the problems instead, you did not give very concrete solutions. One main reason to impose a structure using oop is to avoid "spaghetti code", where everything is connected to everything and it is not really obvious what functions are called from what scope.
I have 35 years of software development experience. OO is a tool. A very effective tool when used correctly but not so much when you slavishly allow it to dictate every little decision you make. Leverage the strengths of OO (like polymorphism, high cohesion, low coupling) but minimize its pedantic usage (like unnecessarily complex inheritance trees for the sake of OO purity). Unless you are using a pure OO language like Smalltalk, most OO languages are flexible enough to be multi-paradigm. It's a poor workmen who blames his tools, unless of course it's VB.
Agree! In my simple understanding. Objects are stateful and most distributed and multithreading approaches assume stateless where data is separated from the handlers. OOP still is a working horse of all internal logic.
OO is a tool, but in language like Java uses OOP all over the places wherever it's not even necessary. New programmer checking the codebase gets confused when OOP used in places when it's not even necessary, with inheritance its just nonsense abstraction.
There is a programming tool that, when used incorrectly, causes compiling errors (and the code won't compile). No damage then. When used correctly, causes the debugger to be practically useless. Have you tried it? And yes, it is FP :-).
And the problem is there are too many purist that do exactly what you say it should not be done. I know a programmer and even have to work with that instead of making a single switch with 5 different settings will make 6 factories that will contain 15 classes to do it. If you are doing that just for a single switch you are bad on the head and heavily. And the switch on itself has only 1 or 2 lines of code. If you say "that is hard to read" trully cannot understand how compared to a total of 21 clases.
This was an absolutely fascinating video. I intended to just watch the start to get a vibe for the idea here and ended up watching the whole thing. I'm a SD student and we are using Java as our teaching language, and I personally am currently trying to square the circle; to extricate the cross-over from encapsulated objects. Something you said in the end about programmers taking good ideas to holistic extremes is something that even someone at my level can appreciate. A lot of the questions I have asked my lecturer, the ones that did not receive "good" answers anyway, relate to this idea. Super interesting video.
Oh, heck yeah. I normally hit the skip hard on lines like "One of the most important yada-yada." There was just enough nuance to promise the goods, and yeah. Maybe not a life-changer, but I'm with you!
Lol... the video just speaks above the audiences heads, where most listening aren't educated / experienced enough to refute the horrible ideas. OO is meant for large projects where architectural issues matter. Java was designed for enterprise-scale applications. It was designed by Sun microsystems specifically for this purpose. If your application is 3000 lines, you might not need OO. You might be able to get away with using global variables at that scale (though if your application grows much larger you're pushing it). In an enterprise scale codebase where you have 10 million lines, no architect would ever design for reliance on global state, this would only result in complete chaos. There would be so many unpredictable synchronization issues it would destroy your application. This has been known since the earliest days of high level languages... Most of the people in these comments agreeing with Brian are students, or people with 1 or 2 years experience, small scale web developers, electrical engineers / other scientists working only with C, for whom programming is not their main specialty... etc. I can guarantee you very few of these people are working at scale, these ideas simply do not work at scale. Which is why they are not used...
@@ianrust3785 Linux is millions and millions lines of code in a procedural style. It has more than half a million contributors and has been maintained for 20+ years. Every AAA game that pushes the boundaries is made in a procedural/data oriented fashion because OOP kills performance. These two examples absolutely dwarf any enterprise app by an order of magnitude in complexity.
@@cavasnel Despite the fact Linux is written in C, the code in the kernel leverages many tricks of the C language to mimic OO. For a detailed description of these various tricks and their use within the Linux kernel google "is linux written in procedural style". Also.... according to The Linux Foundation, since 2005, “some 14,000 individual developers from over 1,300 different companies have contributed to the kernel.”. Most of those contributions are drivers, ... but it's not half a million, sorry.
@@cavasnel Games are typically written with components based architectures these days, most are written with Unity or using various C++ libraries. Try describing a real game and how it was architected in detail. You just claimed the entire game was written in procedural style. What game was designed this way...? And what was the reason - was memory severely limited on the platform for some particular reason? This sort of problem was far more common in the days of 16-bit SNES than it is now. If you’re writing your own custom game engine or image processing functions, some piece of code that executes millions of times for a single update to the UI, then extra layers of abstraction could start to degrade performance. If performance is critical and resources are very limited, eliminating those layers will be one of two dozen optimizations you might do here - because you’re dealing with such a critical piece of code. The entire program would not be coded this way, this engine code will be tucked and the rest of the application will only interact with it through an API. In your main application, the part of the code that is user-input driven (not compute bound), memory management is the only thing that matters. A trip to disk takes about 8 orders of magnitude longer than a compute operation, processing extra layers is a compute operation… eliminating a single call to those layers might amount to a few microseconds worth of difference in performance, maybe. Depends on what’s in the cache. Depending on the compiler, it may actually make no difference whatsoever… To optimize in such a way any user-input driven code is purely idiotic, it is compromising the architecture for microseconds worth of performance. Meanwhile if you need to reuse any code, like an object model or behavior, in the entire game... you have to begin maintaining duplicated or global state for your code, and now you have a giant mess on your hands. I don't doubt some developers have done it, developers do all kinds of stupid things, but this is not a good design by any stretch. A more valid concern is potentially pulling in more from standard libraries than you need, increasing memory consumption, so some developers write their own custom algorithms that reimplement some things from standard libraries... but even this is usually a bad idea, the compiler recognizes & optimizes code in standard libraries, the code actually executes 3-4x faster than custom code. But this isn't a paradigm problem, it just has to do with memory management. Carry onward
Fantastic. And brave. I started with BASIC and machine code (Z80) in the eighties. We knew freedom. I write apps for Android in Java these days. I hate Java.
I have my complaints about Java. For instance, not having unsigned primitives is almost a deal breaker to me. However, It has some amazing features that you can't get from other languages without convoluted workarounds.
I hate Java too. I was in a Java User's Group for a while in the 90s when it seemed like the language was changing every week. Then I found other there were other web programming languages like PHP which didn't require a half dozen XML files to write a Hello World.
He was saying that the inner scope should not be able to access any data that's not explicitly passed in to it. This would let the reader temporarily forget all about any variable which is not immediately necessary for the next logical step. I don't know about c++ but local anonymous functions / closures in both php and js are not restricted in this way. He also mentions that the data is always passed by copy - not by reference - ensuring immutability.
I find I understand what Brian is saying about the tree graph and refactoring problem. In OOP class in Java I felt I totally didn’t do it right. I remember in Java I had to pass data all the way back up a hierarchy of classes and down again instead of just communicating directly. I had to add extra parameters in the class chain just to pass one or two pieces of data. It made looking at the UML diagram obfuscated. Other than this, there would be a lot to pick apart and a lot is about preference and can’t be proven whether one way is better or not. For example Brian you said how long functions are okay and better than abstracting away short functions because the reader of the code base has to sift through the names of the functions and that you can just comment the longer functions well instead. Orrr, you can just abstract the long functions away into the smaller ones and document your code well so that the reader can quickly glance at the comments and see which functions are sub-functions of others, and good IDEs like VS will do this for you, that is tell the reader where the function is referenced or called. This is a matter of style. It can’t be proven that one way is intrinsically better than the other. You either comment your long functions well or you comment your short functions well, but you document either way. The reason we break code up into smaller chunks is to forgo the law that complexity is always more error-prone. Breaking tasks into smaller chunks eschews accidentally updating a variable that a sub-task doesn’t need to touch. You threw in Agile, Test-driven development, and dependency-injection as bandaids for OOP. What?? I think you’re reaching. Maybe DI, but Agile has nothing to do with OOP. You can “fail-fast” in procedural and functional languages too. One more comment is on your saying that there’s no good way to connect behavior with a class, like should you send by a sender or receiver function. You basically said there’s multiple ways to do it, so therefore there is no right way to do it. Well… just pick the best one then-in the flow of the data. Sender. A sender method will import and call the method that does the sending. If you receive, then that would be an event-listener. The premise of this video is that we need to account for if data can be sent in multiple directions and that state… basically can’t be managed, so why bother at all. Data generally flows bilaterally though, and you’re always going to have to have some “god-object” to manage the state of any subset of data, which classes encapsulate, or in other words organize, whether it be a class or a namespace, or else this objection about maintaining “state,” or the state of your program, will always be unmanageable. This is my opinion anyway. I am certainly not wholly informed or don’t have experience with procedural I guess, but I do with OOP and functional. I appreciate the exercise of freedom of thought and opinion, but I think some things in this video are reaching.
I really liked this video! At the time of writing, so did more than 3 quarters of the people who gave it thumbs, which is really reassuring after reading through the horrible comment section. I have to admit that I related to every single example that was given, and it brought back painful memories of so many years of writing code. Thanks Brian! Now I don't have to feel guilty any more, and I have the courage to say it! :)
so.. i did get some way into the video. what i found strange is that first you state that oop shouldn't get all the credit for java has done right since a lot of it is not fundamentally about oop but then you go on to criticise oop for some things which are not per se a problem of oop but rather how things are done in best practice. i am currently at 22:17 and you just complained about shared state and how you shouldn't pass references to other objects (although i dont see why this is a problem when done well. sometimes shared state can help coordination of classes). you could just structure your oop code to always just pass copies of your objects like you do in functional programming. anyways, i wish you had used a different intro for your video, as it doesn't suggest that anything of non-subjective nature is going to follow..
YES 100% yes. you have managed to expertly put into words my exact frustrations with new coding paradigms and frameworks. I would much rather write functions instead of having weird datatypes and meaningless required declarations strewn all across multiple files that is hard to follow.
Great job on this video. I have come to most of the same conclusions from my own experience. I now love a functional style where most functions are pure. Greatly simplifies everything.
True, the C way is almost perfect, but there are higher level features which would make it better. Generic code generation is a good thing to have which saves you a lot of grunt work and bizarre macros, the issue is the C++ way this is done through the template system is a bit limiting and there's no way for example to store templates as DLLs as they don't really "exist" as something in binary. That's the other thing as well the inconsistency and complexity in which the C++ ABI is setup.
This sounds more an indictment of lack of design than OOP. I programmed both procedural and OOP. Having learned OOP using Smalltalk and Lisp, by the time I learned C++ and Java, I came at OOP programming very differently than those who started with the latter. Harlan Mills repeatedly demonstrated the value of mathematically proven design, yet the business call for creating a prototypes eventually overwhelmed the idea in the industry. Agile methodology further allows the programmer to really skip design and figure the whole thing out later. Early Java libraries and even MFC went a bit overboard on the use of inheritance and that started to turn programmers off to the idea. Nowadays, I see more code written with a procedural ideal, but using objects. Component frameworks like Spring and entity frameworks like Hibernate helped feed the creation of discrete objects, instead of the OOP concepts that actual delivery the power. The fallacy of his argument is the "truth" that a strictly OO coded program is a tree with only descending communication. The whole concept is that any two objects can communicate, each deciding whether to respond or not. His arguments are what lead C++ to have friend methods. Encapsulation is broken for convenience.
+Bryon Lape , I agree with the video content and also with your comments. The idea of class instances containing references to other instances for communication is obsolete. In essence there is no need for a complex tree of instances that are limited in their ability to interact with other code. Because there is not a reference availble. The idea I used long time was to implement Runtime Contexts (thread, request, session, application instance, server side instance, cluster instance) which can be used by any class. So any piece of code can be easily accessed. Even a procedure (static function) might hide where the reference to the instance is stored for further retrieval. So omni directional communication is possible.
exactly Dad Bryon Lape - exactly. Except one thing I'll add is that a cyclical object graph driven by some reactive data bindings is an industry standard at this point. Formula for making this video: come up with the worst OO design possible, use this to criticize the whole paradigm, ignore all problems in your alternative suggestions, rinse and repeat.
Interesting. I made a total-conversion ARPG mod for an RTS recently and found dealing with class organization to be the key struggle. The code, though powerful in the context of the mod, ended up as spaghetti as anything outside of the object's scope required additional subclasses or types to control things since they didn't necessarily belong to that class, or could exist in one class or another since they shared featurization (in game dev, this is probably a common issue as features are added). For every added layer of complexity I'd wager the effort to make a change increased by 3 to 4x (which is pretty crazy). It starts to make sense why the AAA titles coming out today are a mess, given the size of the stack at play and the corporations fiddling with them. Nightmare fuel.
A lot of games today are entity based, not OOP, so instead of a class for each NPC, the NPCs are just a number, that refers back to a container that has all the data of all NPCs, that is then updated by a "NPC" algorithm. I'm not familiar with those systems but that's the gist of it.
@@MechShark I have no idea, Unity seems to be OOP, but no one in their right mind would write a brand new engine using OOP these days. edit: apparently unity was refactored to an entity system
16:37 "What is the one abstraction bigger than a function and bigger than a data type?" Mathematical categories, of course. They provide an even higher level of abstraction than objects in OOP, since they are able to describe more than one object, and they remove the need to play the responsibility game that is referenced at 27:22. They can be most commonly implemented in software as a collection of data structures and a collection of pure functions that map between the data structures, with the additional constraint that functions are associative and data structures have some concept of persistent self-equality (so that there is a "do nothing" or "identity" operation that lets you leave a data structure alone without it changing its nature). This by itself is enough to meet the formal definition. You could implement categories as namespaces or as class constructs within code. Implementing a category using a namespace construct allows greater reuse across software projects, since homsets within the category can be stored within separate files. Since a homset captures a very small, perhaps even atomic aspect of interdependency, moving a single file across software projects will need only require importing the minimal set of other files needed to support it. If a category is implemented using a class construct, then there are many concepts within category theory that can be used to precisely describe many of the aspects of Object Oriented Programming that people actually like. For instance, categories often require other categories so that their behavior can be implemented. The categories that are used to implement another category can be tracked as dependencies passed to a constructor, which itself forms a category of categories, where objects are categories and arrows are constructors. This is similar to the way OOP makes use of dependency injection. In another example, certain kinds of functors within category theory can also be implemented using inheritance and interfaces. A design approach based around categories shows many of the positive attributes that people have found with OOP (dependency injection, code reuse, programming to interfaces) while eliminating many of the problematic attributes (state obscuration, state management, the banana-jungle problem), and it is based on a solid mathematical foundation that OOP lacks, which removes the need to discuss what it is we're even talking about, as is done for OOP starting at 1:01. The approach can be considered consistent with the "object oriented & functional" design approach mentioned at 6:12.
There seems to be a good reason why the term "abstract nonsense" is used about mathematical methods used with category theory... I just wonder how it is expected that the average Joe Coder - who may barely have grasped recursion as he drifted from HTML, CSS and copypasting and hacking JS by trial and error, to doing a little backend programming in Php - will understand category theory, and to what extent he should be expected to understand it?
I don't think they really can. Writing OO code that avoids this problem necessarily means embracing the procedural *mindset.* You have to abandon trying to make a taxonomy out of your code and instead write it so that the _flow of control,_ the _procedure,_ becomes the priority. Sure, you might have some data objects with specific functionality attached out of convenience, but that isn't widespread application of the OO paradigm. Instead, your classes (other than the ones that primarily hold data for passing around) end up becoming mere containers for procedure without state, which is only necessary because the language requires it. This does have the benefit of enabling faking for tests in OO languages, but again, this is a kludge around the missing infrastructure around first class, stand alone functions. (E.g., contrast with Python, where you can fake functions directly.) That isn't OO; that's just procedural in an OO heavy language.
Yeah, as a computer science student that took the supposed rules of the OOP paradigm and the supposed virtues it provides, I often wondered as the things I programmed became more complex just how on earth people tolerated writing code like this. I could whip up something fairly complex quickly with a procedural approach, but trying to force everything into boxes I was supposed to keep separate just paralysed me and I spent more time shuffling those boxes around to keep things neat than getting stuff done. All I could think was "this is so much extra work, but I guess it's good practice. It'll be worth it in the future, etc." So, can't say I was particularly impressed when I started looking into real world OOP source code and found out none of these rules are actually followed and people just threw around references to objects that were specifically there to allow global access to references of virtually every other object in a giant clusterfuck void of hierarchy. What's the point?! Now I still have to micromanage state myself, because your code will let me do things I probably shouldn't do, but I have no chance of understanding what's actually going on because most of your code is still pretending to respect encapsulation and is just juggling references to references which is now nothing more than a maze between me and the actual data or functionality.
@@coreyaruecker Depends what your version of "keep things separate" is. I don't understand how people write code that seems to contain more boilerplate than logic. Like, imagine code that has a comment every other line explaining verbosely what the code's doing in plain English. (Obviously I'm not comparing OOP to comments in terms of functionality, just using it as an extreme example.) Do the comments hurt the functionality? No. Is it useful? Maybe to someone who doesn't know how to read the code itself, or gets lost in it without plain English. Is it pointless busywork at best or a cluttered mess at worst to someone who doesn't need the comments? Absolutely. To be clear again, I'm not trying to say OOP practices are like writing needless comments. I'm just trying to illustrate how it can sometimes *feel* to work in OOP code written with a level of granularity that feels unnecessary. Like having an entire directory designed to contain a single file because to some people that's "more organised," when to others it's needless convolution.
@@harrywang4769Oh I know. Only half joking when I say I realised I wasn't cut out for working in the industry when I figured out most of the job is about working at the level of the lowest common denominator.
@@koacado Almost surely not - so often you find out a factory or manager or ??? is actually just a function, or a function with 2 arguments that should be passed on to another function, doesn't even need state.
What I like about this video, is that you really know your stuff. You're not someone who dipped into OOP, didn't like it, and posted a rant, rather, you've done extensive research into this field and have obviously put deep thought into everything you shared. I'll also note, when you first started talking about not being afraid of large functions, I was ready to pull out my own pitchfork. But then, as you started talking about breaking up that large function into sections, and making a bunch of small scopes within it, I realized the beauty of what you were proposing. Now, we have a single function we can go to to get a sense of how the logic flows, but we've also broken it up internally, so it's easy to understand individual parts of the function. That "use block" syntax would be really helpful for facilitating this sort of pattern. In fact, now that I think about it, I'm realizing that I've done this sort of thing before. I created a bunch of sub-functions within a bigger function to section it off, then I created a scope at the end where I invoked the different sub-functions and passed parameters around. I chose to put all of the data in a single scope at the end, to force each sub-function to explicitly list the parameters it needed. It wasn't allowed to just grab them from the enclosing scope. function doThing() { function taskA() { ... } function taskB(a) { ... } function taskC(a, b) { ... } { const a = taskA() const b = taskB(a) return taskC(a, b) } } A use-block syntax would have made this sort of thing much easier to do.
@@KebunH I haven't seen this sort of design used in JavaScript, nor do I see the resemblence with classes (you can emulate what I did here with classes, but IMO, it's wrong to use a class when there's no need for state or dependency inversion is not being used). In fact, I rarely see people create new bare scopes using "{}". Did you by chance mistake that code snippet for the factory pattern, which I know is popular in JavaScript? At the end, I did this: { //
@@thescottyjam8906 its similar to object oriented programming in the sense that the way your function resolves to its final value is by using internal encapsulated (private if you will) state and functions, albeit only temporary. Its not the best example of class emulation indeed as its not a reference but a value thats returned, but it still does something very oop like
@@KebunH Ok, I see what you're saying. I will mention that this type of temporary encapsulation and state-management isn't unique to OOP or classes. You can take the purist functional language, like Haskell, and find this sort of pattern all over the place. People will use the "let-in" construct to declare functions inside of functions, use them, and return a final value. (and, indeed, the use of functions inside of functions is common in JavaScript like you said, it's just the specific use of the sub-scope at the end that I felt was unique). So, we could say that this pattern is emulating something you can do with classes. Or, perhaps, we could say that when people use classes in this way (where you create an instance just to call a single function on it and throw the instance away), they're emulating this commonly found functional pattern ;). Either way, it is an interesting correlation.
Side note: your idea of anonymous functions that don't capture any enclosing scope is very much possible in C++ by just using a lambda with the empty brackets [] which does exactly that. It's a cool idea to structure long functions using that, although I'm not sure how I feel about the arguments being at the end rather than the beginning. The use syntax would be nicer. Personally, in C++, I'd rather just declare some temporary variables right before setting up the lambda, then have it capture just those variables (so that everything will be up front).
Lambas in C++ came later, the video is older ig. Btw, I like this pattern (over just creating a scope with braces) because it allows early return (no if else gymanstics needed)
A lambda "captures" the variables that it uses, but it can access some variables without capturing them. They're not necessarily trivial to factor out.
I was a Philosophy major first in college, then changed to Comp Sci. The part about 'abstract' actually made me laugh, because I literally thought that every single day.
@@unmellowthegamer8946 Abstract because you're extracting the core shared essential process from the specific implementations - just like abstract images aim to capture the core essence of an image, or a set of images. The word means exactly the same thing in both contexts. The language that surrounds programming can be daunting at first, but if you apply some rigour to your efforts to apprehend the meanings and purpose behind the names chosen for concepts it begins to make sense. I found such an aha! moments to be real breakthroughs in both my comprehension and my programming capabilities.
The definition of abstract is "existing in thought or as an idea but not having a physical or concrete existence." This perfectly described what OOP attempts to achieve. I dont see the problem.
It still has the same core meaning in both philosophy and programming but usually it's something that takes a lot of thought to make intuitive in the former, yet it's something that makes intuition easier in the latter.
This is a very interesting video. I had been writing procedural code for a long time. When i learned oop and design patterns, i never looked back to procedural programming. For sure it will work whichever paradigm you use, but a properly written oop code will give you more flexibility. I revamped multiple systems (products that are currently out in the market with revenues in millions of dollars) for different companies previously written in procedural languages into object oriented and it would always result in a more flexible software that can easily cope up with future changes. When i revamp procedural codes, 99% of the times is that they are very rigid and will require very invasive approaches to implement new features and make things more flexible.
I personally dislike OOP for many of the same reasons this video enumerates. However, I think the case you describe is the best possible use of OOP (or any (enforced, or accepted) formalized programming system. When you approach the rigid procedural system you’re refactoring (I assume), the “pieces” (“nouns”, “verbs”, higher level logic, etc) are already known and sketched out. For me, the bigger problem isn’t with OOP _exactly_ it’s with the attempt to design before the shape is known.
I can agree with most of the things you say in this video, except for the part about large functions. If you inline all code into large functions and just separate it by comments, you force me to read all this code or to constantly skip over it when I try to understand the higher-ranking logic of a function. The idea to split code into many small functions is that I don't have to read all that code unless I have to touch it for some reason and then I only need to read the code I want to touch. If the name of a function tells me exactly what this function does and I have no reason to assume that this function does not work correctly, I won't look at its code at all! Even if just called once in the entire program, why would I want a base64 decoder to be inline instead of having a call to "base64decode(...)"? I know that it does. Base64 data goes in, decoded data comes out, no need to see its code.
Yes, this. Using functions to spread apart your code can make it so much easier to read, when done correctly. Sure, if you're looking for a particular functionality, you will need to CTRL+F to find it. But if each function is longer than 5 lines of code, leafing through pages of functions to find exactly what the application is even _doing_ on the high level is annoying and inefficient. To add to this, you should NOT be afraid to make functions with long names, there is no real downside as long as it's smaller than ~10 words. A long and descriptive function name means that anyone who reads it knows exactly what it's doing, without needing to read some long extra sentence/paragraph in a comment. Abstract function names like "Build()" force a reviewer to actually look in multiple places to figure out exactly what the duck the code does. Both of these ideas complement each-other and lead to code that is both categorized neatly and quickly understood on the high level. However, you still need to use your own discretion on deciding what kinds of code to split up into functions like this. Sometimes it's just not worth it, as the video author describes.
@@LordOfLemon I don't have professional experience with coding yet so I might be wrong but I haven't seen situations that work better without separate functions outside smaller pieces of code. For a program that does multiple things, separating methods and commenting makes it much easier to read and understand. If you have an example of a long code where functions would be the worse option please do share
@@mistzzz4683 I also find myself disagreeing with the video's suggestion, but one place I prefer inlining is when the function in question is tiny, obvious, and only called in one or two places. For example, it's way less readable to see a call like "n = timesThree(n)" instead of just "n *= 3". I've seen real examples almost that bad. I love small functions, but at the point where the function name is just an English transcription of it's entire one-line body, I'm actually going to stop and read it because I'll be wondering what was so hard about this that you had to define a function.
@Mist zzz I have professional experience, and the answer is just that there are a lot of bad coders out there. I once spent an entire summer refactoring A SINGLE FUNCTION. It was thousands of lines long, and it meant I had to constantly delve deep into various levels of logic to understand the code. Some flows of code are long and complex. In this case it was the handling of a submission for a new user in a specific system. If you were A, you were gonna get Y and Z, if you were B, you were gonna get X. If you were C you were gonna get X and Z. And then with A B and C you were gonna do some common setup. Now, I was originally only going to add a new case D, but I couldn't. The code was so messy, I couldn't interject W into the code, because everything was mangled. That's why you want a function that describes what you're doing. submitForm() { createUser(); setupUser(); giveUserStuff(); sendOutMessagesToUser(); }. If I am going to give a new type of user new stuff, I need to change giveUserStuff() function. This is how you can see that you can drill down into the depth of complexity a lot better. Comments do not do this in the same way, because the code is still there. Good code is also comment enough. Show me what you're trying to do, then show me how you're doing it. Even when you're doing it, show me what you're drying to do, then show me at the deepest most complex level what you're doing. In my case, there was an API that was being used, so all over this original function, there were these extremely complex API-calls that were being set up in complex ways. Instead of saying giveUserASubscription(subscriptionType), there was a huge chunk that set up an API-call and just used the result. When it's my first time opening that code, I am not able to quickly scan that and ascertain that this part of the code is the API-call that's giving a user a subscription. I have to scan through low-level API-setup and see that the API call is something related to a subscription. That's why you first show what you're doing, THEN in a new function, show how you do that. That means that if I am giving the new userType D a new subscriptionType N, then I don't need to care about how you giveUserASubscription, I can just call giveUserASubscription(N). This means that it's not only good at showing you in coarse terms what the code is doing, you're also allowing me to not have to even relate to the lower levels of the code, if I don't need to. Say that later someone needs to implement a newer version of the API-call, then they can just find where all those API calls are used, and change how they're called. Then you don't need to relate to what you're trying to do. That's why abstracting code makes all tasks besides a complete rewrite a lot more easy. Maintaining code means that you should have to deal with as little as code as possible when you're doing a change. That way you lower cognitive load, which in turn means less chance of error. You also touch fewer lines of code, again making it less error-prone.
Yes. And smaller functions make the code better testable. If you have assumptions about the state of the relevant variables while entering a certain code block, and the desired result after passing it, it is testable if that code block has a name. Finding descriptive names, which make a longer comment superflous, is often time consuming but it teaches you in reflecting upon your code and what the parts of it do. If you are able to leave a helpfull comment, deriving a name from that should be possible.
Good video. I was long mired in the OOP world, but a few years ago I started delving into the functional and procedural side of things. At this point, I'd say that I FAR prefer a functional or procedural approach. To me, moving away from classical OOP has been liberating and empowering (and I'd say it's also helped me when I do need to work in OOP codebases). It's changed the way I work. I hardly ever need to step through my own code with a debugger, these days. And I no longer care much about Intellisense-like tools which I used to think of as an absolute requirement -- I no longer need them the way I used to. I know there are many who are deeply invested in the OOP paradigm, think of it positively, and want to defend it (I was one of those programmers). To anyone in that position, I'd encourage you to investigate functional and procedural paradigms with an open mind. You might be surprised to find that you are a happier and better programmer moving with a move away from the OOP world. If not, and you still prefer OOP, there's no loss. Learning different ways of thinking about code will help you in the OOP world, too, and make you a better programmer overall.
I am a big proponent of OOP, but I came to watch this with an open mind hoping to learn something. It turns out we agree on many points, but probably that's because I'm not taking OOP to the extreme, for example I don't agree with / don't do these: * Some OOP folks think there can be no such thing as simple types: even int and double must be a class. * Taking encapsulation too far, where you can't even access simple types from an object directly, no it must use setters and getters. * Trying to force everything into objects even when it doesn't really fit. Some things really are just plain data and plain functions. There are however some things about the other end of the spectrum that I find absolutely godawful: * Having your state scattered into a sea of global variables that are read and/or modified by functions which don't take them as a arguments, and in the worst case even having those globals be used across files. * Having lots of separate variables not organized into structs or classes, and a set of functions not organized into namespaces or classes that you have to guess what belongs with what and which are intended to be used together. * Having long complex procedures rather than capturing key parts into functions or at least scopes. This is analogous to writing a wall of text rather than using paragraphs.
For a while I thought I couldn't learn programming because people needlessly overcomplicate it on purpose. After watching this video, I now realize that I can't learn programming because people needlessly overcomplicate it on purpose.
People dont overcomplicate on purpose, that just doesnt happen. No one in the world is inherently evil, theyre just doing what seems right from their perspective. The problem isnt really overcomplication, it's that programming itself is complicated. Even with that, I dont think that being "unable to learn" it is ever correct. If you just stay curious and learn a little bit at a time and keep experimenting, you can probably become so much more skilled at it over time than you thought you ever could be. It takes effort but I think youre wrong in saying you cant.
Usually, it gets forced by some management, or some practices people learn. Extremely well designed object oriented programming is very easy to read. Much easier than functional or plain old C, because the functions are with the data.
Learn to program in C. It has a name of being hard, but that's incorrect imo. It's hard to write large software packages in C, but the language itself is pretty straightforward. And a great thing about learning that way is that sooner or later you will run into the limitations of the language, and then you will easily pick up on more abstract ideas like object-oriented programming. And you'll learn it not as the one true way, but as a technique that is in some cases helpful.
"Bundle globals into structs/records/classes. But be careful, there's an art to how you bundle these things together. " It's called object-oriented programming Also, I have an issue with this advice "Don't be scared of long functions". The reason why I generally have to avoid them is because I'll need to reuse the "components" of the function. There's a myriad of business areas with similar/overlapping business processes that needs to be mapped like this. And Brian's advice doesn't provide any solution to this. Is the solution to have copies of nested functions everywhere. How do you change your code later? Generally I think he's on point with the problems of stateful, OOP. But I don't think he quite manages to point to any useful remedies. Programming is hard, because mapping complex domains is just hard. Real life is messy, and code that tries to reflect that will also, inevitably, be messy.
I clicked on this video with skepticism because the video title is so "black and white". But unlike the title, the contents of the video is not so black & white. I think a lot of people who thumbs-down this video actually aren't watching the full thing. For your convenience, the 1st half-hour or so is all background and setup and analyzing the problems. He gets to his list of proposed best-practices (which are quite reasonable) at time index 34:46.
His evidence is anecdotal and many of the issues can be attributed to incompetent programmers as opposed to their paradigm. Also some of the advice is really bad and I hope I don't have to maintain code by people who took that advice to heart. Nested functions in favour of-you know- non-nested ones? Relying on comments? Like anyone writes those where they're needed. Properly named functions are the natural comments and make things more readable in the process. For the love of Zeus don't use large nested functions just because you "shouldn't be afraid of large functions". Give my scroll wheel a break.
It's hard to take anyone seriously when they start with this is the best video ever. It doesn't help when having decades of dealing with people who say that and you find out most of them are blowing smoke. It ruins any attempt at trying to make a change when he starts with a victory lap.
@@Kinglink He didn't say it's the best video ever. He says it's the most important programming video you are ever going to watch, and he's right. The fact of the matter is that the majority is not always right. Sometimes all of society can collectively make a mistake, and that's exactly what Java, and OOP was.
OOP is wonderful for large codebases. I have worked far too long on the IBM iSeries platform with its go-tos, and top-down programming. It doesn't work. It's nearly impossible to upgrade codebases that are gigantic and written top down quickly, and they're insanely error prone. Local versus global variables and isolating data in OPP makes life SO MUCH BETTER. I feel like you haven't ever worked on a gigantic codebase.
This is true, and i also feel like this video has nothing to do with areas of coding that arent database/website associated. Like simulations, or even game dev, where you dont need all data to be stored or returned somewhere. In that context youre not just "performing an action on data", youre using objects to capture behaviours, and having them split into classes with encapsulation allows that process to make sense, without ending up with a script thats 20000 lines of code. Entity-Component DOTS style coding is often a better choice, but its readability is atrocious compared to OOP, and thats important for quick testing and turn around times. The idea that a style as useful, flexible and readable as OOP is just blanket "bad", is really more just an admission that you dont know when to use it, or how to use it.
I really don't see how huge functions with only comments to break up logic are any more readable and, more importantly, easily testable. I would much rather modular functions that can be separately tested and read quickly rather than just a huge column that I have to memorize the components of. Also why wouldn't you just put the same comments you suggest in your longer function before the function calls?
I watched the video with an open mind... but then he said we don't need functions, we need inline functions. No matter how "good" the code you write is, it will break at some point. I fear debugging an unnecessary inline functions as much as a dependency hierarchy that is too deep. Debug a 2000 line function and you'll see.
It’s more readable because the sections are lexically contiguous, instead of being scattered around the source file. Wanting modular, separately testable functions is really just wanting the unit tests to automatically localize any bugs that occur. If the large function fails its test, you still know something’s wrong with it, you just have to manually examine more code to find exactly what. Writing code so that its tests can localize bugs is optimizing for the wrong thing.
@@richardblain4783 The separation into smaller functions does not imply testing all the little functions separately. A module can be broken in 15-20 functions instead of a huge pile of 2000 spaghetti code, and still you would test the same behavior. I'm working on a pile of functions that are thousands of lines long, with no tests. I can not refactor because well, there are no tests for the big function either. The auto refactor failed because it is written in the 'C' style with 'goto cleanup;' instructions. If the programmers had split the code into smaller pieces it would be easier to reuse the logic AND to write unit tests for them step by step instead of testing a 1000 logic (which is way beyond my timeline). While I agree with most of the things said in this video I'm totally against having long 1000 lines functions
Also, I think longer functions makes the code harder to read because it implies that all the details are in one place. You start seeing things at a high level of abstraction. Ex "makePayment" entangled with ultra annoying details like a for loop over char*. You start reading code that is like 4 levels of abstraction in just 10 lines. which makes the cognitive load way higher than just see nice separated subroutines.
"There are only two kinds of languages: the ones people complain about and the ones nobody uses." - Bjarne Stroustrup
@@gl3nda96 so you did the deed?
- every video or post about programming languages anywhere on the internet
Neither being used, nor being complained about make any language good.
It's just another way to say the set of all languages is distinct from the empty set. Every language has it's flaws and so there will be complaints about every language. Even if there was a perfect language people would still have different opinions on it. So basically what you posted is a true statement with no use.
I still never met a person who uses C and complains about C.
The irony of this video is most people watching it are just going to be better at OOP by the end of it.
that is because Brian Will understands OOP. to criticize something you first have to understand it. and if you believe he know OOP better than most of us and also procedural, why is there so much disbelieve in his statement that procedural is better?
the way Java does OOP is just a special case of procedural programming. it forbids a lot and what it does not is called OOP in Java.
Yes, and for yet another level of irony, this was at least as boring and confusing as are the tutorial lessons and especially those advocat-preaching speeches of OOP.
Well, literally couldn't agree more - only the limits of understanding make barrier for agreeing. This proves that to the limit of understanding and beyond, OOP sucks!
The objective was to get better not perfect
I understood maybe one third of the video but between the lines it seemed to make a pretty solid case for OOP.
@@75hilmar I understood all of it and I can say that if you understood the other 2/3rds, you wouldn't say that. He's really destroying OOP.
Hm, yes, this video has definitively shown me that I don’t actually know enough about programming to properly understand this video.
Just give it a bit of time and one or two serious/professional projects and you will know the problem firsthand!
After two years in university you will understand it.
@@GuerreroMisterioso95 Give him 1-2 years in industry and he will definitely understand it if not already
No worry, you have nothing worthy to understand from this video. Better read Gang of 4, Martin Fowler, Robert Martin and other references. His whole argumentation, especially about encapsulation (eg from: 23:04) , is extremely biased and perverse. His arguments are based on object graph examples which are badly designed in first place. If you respect reasonably SOLID and general OOP principles, you should NEVER get that kind of spaghetti object graph in first place. Then the code would be encapsulated already (SRP, DIP, ISP, LSP, demeter law, and so on...). Then you wouldn't need to encapsulate non existing spaghettis such as he's attempting to, to make his point about how "OOP is bad".
And about his statement: "Abstraction = simplified complextity; abstract = hard to understand". Well why abstraction simplifies complexity is because we do NOT have to understand what is abstracted if it's abstracted, so again a perverse argument... Furthermore, abstraction is not more related to OOP than to procedural or any other paradigm.
"Mistery why industry tends by far toward OOP"... Did you try developping then maintaining complex enterprise applications (changing often) in procedural, compared to OO? That wouldn't be a mistery anymore for you then...
"Procedural languages are more polymorphic than OO languages", lol, special mention for that collector one...
And the same goes for a lot of his arguments... That guy is a clickbait sophist misleading least experimented people...
@Clément Cazaud Have you seen the guy's video where he breaks down four real OOP examples from proponents of OOP and how they could be rewritten in procedural form? I think one of the examples in that video was from Uncle Bob (Robert Martin). To be clear, I really respect Uncle Bob and I've bought and read his Clean Code book (his chapters about comments, naming things, and code rot really convinced me to kick some bad habits while programming), but I think the guy's video brings up an interesting point on how one example of Uncle Bob's code being convoluted because it was written in OOP fashion. I'm still relatively new to programming, so my opinion probably isn't really worth much at this point.
“Better to start with a free form structure than to eagerly create one that turns out to be wrong.” I learned that lesson the hard way.
As long as you don’t abandon structuring in lieu of working code later on 😐
@@life_score Indeed. That’s why it’s “Start with a free form structure” and not “Only use a free form structure”.
@@atalhlla yeah, but that usually ends up being a trap. Most people don’t take time to improve code once it’s in working condition!
given that requirements are always changing, a free form approach is more appropriate. If you start out with a rigid plan/structure you are in fact doing waterfall.
The biggest takeaway I get from this, is that this man really, really hates jumping around. He wants to read the function in one place. I can respect it.
Well and I don't want to read the function at all when using it in a smaller context somewhere else, what I don't have to if it is on e well named, well abstracted and I know it's tested.
@@googleuser2016 This exactly. The author of the video would have you believe that readability is just about aesthetics, while in reality it's arguably one of the most important things and we spend a lot more time reading existing code than writing it.
Jumping around makes debugging other people's code an abysmal experience
@@s0mbres Could you elaborate on that? I work with (old-ish C-based) rail control systems with dozens of processes and as you can consequently imagine, hundreds of source files and thousands of functions all around. Nothing makes me feel more thankful than coming across a function that doesn't try to do everything. Core dumps are easier to analyze if the stack isn't completely flat, it's easier to set breakpoints, etc.
@@rndszrvaltas It probably depends on how the code is done. Good code would be lots of small functions that do their own thing, they work, and you can ignore the ones that work. A true nightmare would be lots of small functions that are all interlinked in some bizarre way where X relies on Y, Y relies on X, both of them call Z, Z sometimes calls X or Y, etc. It's lots of possible issues and you're never really sure which one is fucking up, so you end up jumping back and forth just to figure out wtf is going on and how it's supposed to work.
I spent too much time asking myself whether a specific function should be in the ControllerAdapterFactory class or the FactoryAdapterController class.
This sounds like a Monty Python joke
I spent too much time searching Google with "should I..." questions instead of "how do I..." questions.
May God help you! BANISH THE DEMON THAT HAUNTS YOU!!! (Get rid of that evil naming convention you have)
When in doupt, put it in ControllerFactoryAdapter.
Sounds like poorly defined clasess to me. Do you really need both of them?
i do a different coding style . its called "what ever gets the job done"
right? like there's no silver bullet, there's a lot of these "oop bad functional good" or vice versa bullshit on every forums and it always strikes me as biased. there's soo many different kinds of weird real world problems that one coding style isn't enough.
Seems like too many developers have lost sight of this. The end user doesn't give two shyts about whether you used functional or OOP programming in your coding. They only care that it works...and as long as it does what difference does it make? Code it the way you are comfortable coding it instead of constantly chasing after what's in vogue right now.
Define “the job”. If it’s anything other than “make it work this week and then delete it”, these kinds of discussions are vital to the long term flexibility and maintainability of the code you write.
Transparent “working code” may be working, but it still falls on various spectrums of maintainability, flexibility to change, complexity, performance, etc.
I’m tired of people using “different tools for different jobs” as a hand-wavy excuse to use shitty tools just because they feel familiar.
Sure, don’t go chasing fads, but it’s worth considering whether programming really IS that difficult, or whether we just make it that way out of ignorance and inertia.
Sounds like an indian to me
At company after company I have dealt with confusing messes of deep inheritance hierarchies, sometimes 10-15 levels deep, where each subclass adds just a couple of lines of code. The prime consideration was keeping the classification pure in a philosophical way, like we were inventorying the animal kingdom. A good portion of the developer efforts were targeted towards dealing with the structure of these classifications and not getting any actual features done. Factories that make factories. Singletons with pages of boilerplate to do one simple thing, no object ever necessary in the first place.
Spot on. Anyone who writes lots of JavaScript, Python, Ruby or Go and then dips into the C# or Java world knows exactly what this nonsense feels like. It's developers scratching their own itches of systems level thinking, trying to create the ultimate "system" to solve code complexity. Like a mini-game where they're trying to create an encyclopedia for the world. But by trying to systemamitize away complexity they create even more of it. Now you need your devs to understand dozens of different "design patterns" that supposedly decouple your program and make the codebase easier to work with. Why then are they so complex, require constant abstract thinking and make the codebase even more complicated to work with? Why does every design pattern come with the disclaimer that it doesn't actually fully decouple anything and that it's just a different form of coupling? The coupling comes from needing the cross cutting code. There is nothing we can do about that problem other than make it easier to understand for our developers. Creating all sorts of wacky service locators, inversion of control containers, factories, dependency injectors doesn't decouple anything.
Generic interfaces with only one implementation, factories that just pass the value back to the constructor, subjects and observers that only ever get called once, Massive mutators receiving several different strategies do to the same thing an if/else already does.
In general web dev, a lot of "problems" we solve are so simple, and well supported by the base features of the language, that any attempt to show off how smart you are by "using patterns" is guaranteed to generate bloat.
your companies need architects, and start writing some baselines before writing code
Yes. Some senior developers take pride in turning everything into layers and layers of inheritance.
I've known some very narcissistic programmers who write this kind of code for huge organizations like banks. Usually they have zero experience outside their language of choice from 15 years ago and think "standards" and "good practices" are gospel. They usually have zero creativity and think things like "you should use DateTime to store a year value because it's made for it" instead of using a bleeping int.
That guy built a framework for every single little thing, and thought Microsoft invented marshalling. He literally could not define the term outside the context of the .Net Framework. And big companies paid him well to write overengineered crap. Of course, he constantly bragged about all this amazing stuff he'd learned 'at the top.' 😂
I come back to this video a lot. Here's some timestamps:
4:38 Definition of Terms (Procedural, Imperative, Functional)
8:00 Why does OOP dominate the industry? (Java)
15:50 What is the appeal of OOP?
17:18 The One True Way to do OOP (Bandaids)
18:08 What's wrong with OOP (Encapsulation)
20:05 Shared State (Not too different than a global variable)
21:10 Encapsulation requires direct hierarchy (Problems.)
25:41 Premature erected wall building = cool-aide man solutions (OOoooH YeaaaaH!)
26:14 When starting bad structure is worse than an absence of structure
26:44 The mind games of OOP (Unnatural data types, kingdom of nouns, Manager classes)
28:54 Stupid questions you have to ask yourself (Analysis paralysis)
29:52 Abstractions hide complexity (The princess is in another castle)
31:43 Spreading your code out unhelpfully (Increases the surface area of code)
33:19 Solution! Good procedural code:
34:24 What to do about shared state?
34:46 Parameterize! Try not to use globals.
35:15 Bundle globals you do use into a single datatype
35:48 Prefer pure functions
36:19 Use namespaces / packages / modules
37:15 Long functions are fine! Logic in sequence = code in sequence. Use "section comments"
38:55 Use nested functions. (Functions inside a function, so you know it only gets used multiple times there.)
39:50 Constrain scope of local variables (Anonymous functions, use blocks, Jai programming language)
43:32 Conclusion - liberate yourself.
For those who enjoyed this, I also recommend talks from Casey Muratori, Jonathan Blow, and David Acton. Thanks for creating this Brian!
Deserves a separate bookmark: 43:50 you don't need to read any of these books
Refactoring by Fowler
Test Driver Development by Kent Beck
etc.
44:36 trump 2020
Seth Archambault thanks
"Use namespaces / packages / modules" = OO concepts...
Just now watching this video. Thanks for the nice index!
Employee: _We're practicing POOP here._
Applicant: _What's POOP?_
Employee: _Proper OOP._
employeR?
@@drygordspellweaver8761 Nope, employeE. In each of my 20yr past cases, applicants (including myself) were talking to their potential teammates about some tech details.
Guess most employeRs actually don't even have a plan what the heck their devs are doing in detail... 😁
Should be top comment 😂
People Order Our Patties, POOP
No....
POOP actually stands for Python Object Oriented Programming.
My (C)ommunist Coding style:
A classless society
Your coding style is marxism. Communistic paradigm be like...
Every variable, every method public. No private APIs. No borders between namespaces. No inheritance - every variable is common! Get rid of bourgeoisie class.
Lines of the code, unite!
@@mastermati773 That would explain why communist societies failed ... they should've learned to be object-oriented :-D
@@a0flj0 I find some things from programming outside IT: My favourite example are ads. Ads are nothing other than some unreliable message from some host to us clients. Normal people see products. I see that some outside API tries to change my internal state in suspicious way, so I need to project my thinking, so I will be able to handle this spam.
(Sorry for my eng)
PS Minutes before writing my previous comment I had watched Zizek xD
dontlikemath -.- Nice.
@@mastermati773 Its maybe more like functional programming :P
As much as I love object-oriented programming, I have to admit that you strike some good points that I haven't considered.
After backing up my original code for a particular project, I deleted the whole thing (because it was a mess) and started over from scratch.
I will try to implement these principles in that project in hopes of making my code more manageable.
My advice is to write procedural code inside classes and only when it really makes sense use inheritance, interfaces, design patterns and other OO stuff.
The fault was not with the tool, rather your design.
@@HumanBeingSpawn Beware, OOP fanboys always gonna try to gaslight you into thinking you are the problem after your code turns into a mess by following their advice.
@@tongobong1 I have started doing this without realizing what I was doing. My inheritance seldom goes much deeper than 2 levels and even then usually only for datatypes
@@MartinSparkes-BadDragon Objects are great for modeling business logic. When you have a domain expert telling you what he wants software to do it is great to represent his knowledge with objects.
I am a mother one of those old timers - started programming back in 1980. That was with an HP calculator with 15 lines of code and 8 memory registers. There were also the ‘optical cards’. You scribble your program onto cards by selecting various numbers, send them off to the mainframe, then a few days later back comes a printout. No keyboards. No monitors, and the printout just as likely to say ‘syntax error on card 3’ as it is to provide any meaningful result. Imagine, A code, run, test cycle measured in days!
I have been through almost every ‘revolution’ there has been. Several times. These new fangled technologies come and go out of fashion on roughly a 10 year cycle, just with new TLAs. I have also been responsible for pushing a fair few of them myself as well as resigning myself to ‘here we go again’.
One thing to keep in mind is the Turing Machine. Turing proved that all computers, and all languages, are equivalent. Pretty much, once they can do basic logic, then all languages have equal expressive power. Anything you can do in one, you can do in another. So arguments about procedural vs OO vs Functional are moot. I can write an OO compiler using a procedural language and vice-versa. Not efficient, perhaps, but doable.
So what it comes down to, as mentioned in this video, are the practicalities. In practice, in a real team, with real people with real business problems and challenges. The efficiency of writing and maintaining the code. The efficiency of new hires getting up to speed and the risks of losing people with ‘the knowledge’. Building a system that is easy to adapt and extend that Ames the users/customers happy.
This video singles out ‘excessive OO’ or ‘extreme OO’ as a bad thing - in particular encapsulation. Quell surprise. Excessive anything is a bad thing!
For the comp sci students out there, just remember all these technologies are tools. To be successful in a programming career you will need to master a fair few of them. No real world problem, worth solving, can be done well with a single tool. As a wise boss said to me once, ‘this is technology. With the right tool it will go 100 times faster’. This was when I was writing my own ORM layer - not realising there was already a library to do that. Similarly, I once had 85 lines of procedural code replaced with a single line of (damn clever) SQL. Being a programmer is a bit like being a doctor. You can’t solve every problem with a scalpel. You can’t solve every problem with antibiotics. Every person is a bit different. You need years of training and a wide variety of diagnostic as well as preventative as well as curative tools, medicines, machines and hands on experience - book learning alone will not cut it.
The main factor for deciding what tools to use is the Problem Domain. One thing I have noticed, time and time again, is that the ‘best’ computer code accurately reflects the Domain it is working in. I guess it is called DDD these days. From that perspective, and getting back to the video, your encapsulation level should reflect the natural encapsulations of the domain you are working in. Though this is more of a heuristic than an absolute rule.
Let’s call it ‘Domain Oriented Heuristic’ programming or DOH programming for short. 👍
Thanks for the information. I'm still an under-graduate. If anything, this gives me insight on my future in computer science.
I liked that last paragraph of yours. Do you know of any resources to study this further?
Shirom Makkad mmm. Well, Domain Driven Design is the thing to search for.
Most custom written (let’s say in-house) systems, built to support a specific business, end up with libraries, objects and data models that reflect the nouns and verbs and datasets that the business users use themselves. Encapsulation is also reflected in the various departments and functions of the business. Where this gets really interesting, is dealing with mergers/takeovers and internal re-organisations. Over time, what happens is that the system moves towards an ‘industry standard’ model - particularly as engineers and users are hired and fired between companies but within that industry.
Well said old timer. Just left my own comment on this 45 minute rant. I beat you though. Had to carry my cards to the computer building in a wheelbarrow. Those were the days - not. You still programming?
graham287 yeah - spent quite a few years ‘managing’ and ‘enterprise architecture’ but back to programming now. Mainly Python - which I think is now my favourite language.
Mainly because it ‘defaults’ the right way, most of the time and automatically deals with edge cases. So less code.
Q:What is the object oriented way to become wealthy?
A: Inheritance.
nice
create state, create controllers, design gui. besides that you need a state that allows function for your site/app the controls to manipulate it towards the users objective and then the GUI for an end user to manipulate the controllers they have access to.
Kiss ppl's asses and procedurally control their minds.
think functionally about money
Multiple inheritance ;-)
I know from experience, anytime you put an opinion out on something like this it brings out the pitchforks for sure.
Well he shouldn't have claimed this to be the most important video, or whatever similar.
Jerome Potts I mean, this is one of the ONLY videos on UA-cam which explicitly calls out OOP as a bad paradigm. If his assertion is more true than false, then I would say this video is indeed one of the most important programming-related videos on UA-cam right now. If it’s not, then whatever, maybe it’s not that important, and then the pitchforks wouldn’t be necessary... just a thought.
Most people blindly believe what they have been taught at school.
It's hard to fight decades of brainwashing
@@fwefhwe4232 lol
@@fwefhwe4232 Most people do what will actually get them employed in the real world.
This is so good. My first C++ job, 20 years ago, was writing code in corporate codebase where a genius code architect had de-crufted a horrible architecture of earlier encapsulation. Nine layers of OOP encapsulation were collapsed into a single layer in an epic refactoring. That happened shortly before I joined the company and I supported both the "old" product and the "new" product. I had started this job fresh out of procedural coding and the excesses of unsupportable "isa" / "hasa" confusions were freshly there on day one.
I still like C++, I guess. But this is such a well thought out, battle scarred view. Thank you for this!
C++ is not an OOP
The beauty of C++ is that you can practice OOP (classes, encapsulation) etc. where it seems to fit the problem, but not where it doesn't.
The problem with Java is *everything* has to be a goddamned object, whether that concept fits the problem or not. Not everything "is-a" object or "has-a" object or "is-a-kind-of" object. Some things just "is".
@@UncaAlbyGmail Except java was, is, and forever will be a million times better than C++. At least in java you can actually code a functional, logical code, whereas in C++ the best you can do is some fibonacci
@@morbiusfan3176 Somehow I get the feeling you're trying to tell a joke, but it went over my head. Sorry.
@@UncaAlbyGmail I still like C++, and having OOP in places saves time. I am not sure I'd like to inherit somebody's codebase who goes through all the mistakes of OOP. I also never warmed to Java.
Me who has done just a simple calculator in python:
"He does have a point"
A calculator is a very decent program to compare differences between languages, you observed right. You don't need complexity to know the rightness, only how logically sound and consistent it is. A calculator program fills these requirements very well.
@@equinox2584 But you have to limit the input to a certain set of characters, otherwise you can end up with a code-injection vulnerability in your program.
@@SapioiT it's not the thing you'd worry about writing your first program
@@badunius_code Not the first, true, but among the first hundred or thousand, maybe sooner if you hit that problem early on.
@@badunius_code my first calculator was a graphical calculator made with tkinter, mostly to avoid messing around with eval and hardening against code injection lol
In reality though the "combination of objects and functional programming" is already the real world way people use OOP and i might argue the standard. The problems with OOP are really just strawman uses of OOP. In the real world, encapsulation and abstraction are only used if it is beneficial, like reducing complexity and improving readability. In school, it's probably the case that everything is over-engineered, but that has educational benefits too. You get to experience the friction that comes with objectifying everything
One hundred percent agree. For example, he introduces a theoretical OOP practice(not sending object references), completely admits that almost nobody follows this practice, and then spends minutes following that explaining why this practice that nobody follows is bad. Well yeah, no shit, thats why nobody does it.
Agreed. Even when in java, a lot of big projects end up having one or several "Tool" static classes which are used as a repository of procedural functions, used with parameters and all that. Because it makes sense, even if it is not ideologically perfect.
If I'm not mistaken, he acknowledged that people don't really program that way, didn't he? Anyways, I don't really care about that point.
The thing you said that bothered me was the "school fail upwards by teaching you to do things that way", kind of part. I agree, but I just don't like that reality. Could schools be explicit about this and tell you that full OOP doesn't work? Shouldn't students learn something useful instead of that?
If anything, I think we should push back against the literature that pretends that full OOP works.
in every project i was involved in, there has at some point been a situation where we went "damn, if only we hadnt cut corners back then"
@@MjolnirFeaw yes, those repositories suck ass, because they are usually strewn all over the code base, and the next guy is just gonna write his own "Helper" function rather than see if something already does the job. They result from not actually working on the architecture.
This is how I saw scientists program intuitively. I was an experimental physicist before changing to the industry and most programs, I have seen back then, written by physicist followed the functional programming idea. Maybe, because all of us were comming from a mathematical background. Some programs were a mess though and comments were in Portugese, German and English on different layers of code. ... Good memories. And good video.
@@Bayo106 not even close. engineers.
i knoww XD
Procedural coding has its name literally because its how most would learn to code if self guided.
i think.
usually accurate but not probably not correct.
Scientists don't know how to program, this is well known to software engineers...
Well, the quirky thing is that we're doing OOP when deep "in the beast of the machine" it's literally running the code in a sequence vis a vis procedural.
You must understand that every programming paradigm solves those problems, that other paradigms can't solve. And different tasks require different paradigms
I've made enough OOP mistakes that much of this resonated with me and wasn't really surprising. But the bigger take-away is much of these same problems apply to microservices. Each microservice couples it's own data storage with the service and they don't share databases. This causes all sorts of problems managing the indirectly-shared state. It is a similar problem, by trying to force a small encapsulation, problems get spread around but not made any easier.
That is a very insightful analogy.
microservices don't solve any problem, they just move the complexity of the monolith to the infrastructure, it makes no sense
Why would you put data storage into every microservice?
@@Ian-eb2io That is a fundamental concept in microservices. Each service has it's own data store containing all the data it needs to use. The point is to reduce how chatty your system is, rather than call another service for the information, read it from your own data store. This prevents other issues such as deadlocks or infinite recursion. It breaks DRY, but is an intentional duplication.
@@username7763 Are you telling me each services having it's own data store is so that it CAN have access to the data? Otherwise it has to call other services to get the data for it?
11:43 "It seems like real programming, it has curly braces after all."
I see, a fellow man of culture.
I recently ported a PHP web api client into my project without thinking . It has 1300 lines of code, 8 classes 4 files. Pure unadulterated OOP straight out of College... The amount of lines performing the actual function of the application? 8 lines. Which was what was left after i pilfered it for my own use (i had a duh day not knowing php uses CURL library for webapi.)
*python has left the chat*
haha goto go brrr
@@williamdrum9899 Underated comment.
@@TheBelrick Who did that and why? 😨
Whether I agree or not is unimportant. This was very thought provoking indeed. Many thanks!
Blasphemy. How refreshing.
That's the attitude I like to see !
My goodness, as a student learning OOP I found the criticism here so relatable. I spent DAYS thinking about how my project could be conceptualized into classes and which methods belong to which type. I decided it was my own lack of OO design experience but I'm so glad to learn I'm not alone and it's never possible to make perfect object-behavior encapsulation in the real world. The "matchmaking game" is absolutely real.
Fellow student here, I see inheritance useful if the main code changes a lot and the clases that inherit do not change. Otherwise, composition is useful. I find it rarely in my programming that I used OOP succesfully, but the idea that I want emphasize is that if this style of codijg is hard to implement, maintain and change for any programmer at any level, it's probably not good. And I haven't even talked about scalability...
Also a student! I know, I’m so happy I came across this video.
Then when you graduate you'll be writing webapps that make almost zero use of OOP despite you likely writing them in a class-based language.
@@bobbycrosby9765 so typical nodejs smelly shit? (Even though js does not really have classes ... or does it now with a new standard? not sure)
I've been in the industry for 20+ years and thought I would share my experience.
These are just tools. Imagine 2 handymen arguing that the hammer is THE tool and another claiming a screwdriver is THE tool. They go back and forth pointing out the inadequacies of the other while showing the advantages, elegance, power and more than anything personal aesthetic preference for their preferred tool. Meanwhile the thousand other handymen in town use both tools alongside each other as well as a dozen other tools in their toolbox.
Out of the 2 arguing handymen one is certainly more right than the other, though no one is likely to know which. All the other handymen in town know how and when to use specialized tools and they are getting paid the same if not more than the 2 arguing. Learn what you can, try it out, use it when it applies, get paid.
School is just one of the tools along the way. There are many problems with the educational system, but that's too much to cover here. What the education systems effectively do is certify you can solve problems and you can see things through. You will learn more about effective design in your first year working than you did in the 4 years to get your undergrad. The learning never stops. Your intuition will continue to develop through your entire career.
Listen and extract from these hot takes but please don't take them as gospel. Mostly don't take anything as gospel, just figure out how things are useful.
Enjoy your journey.
I return to this video periodically, every year or so, to read comments from people coming back to this video periodically, every year or so
The anonymous functions denoted by the “use” keyword exist in C++ lambdas, right down to the explicit list of named function variables that can be shared with the anonymous function.
Is that what a lambda function is? I have been trying to understand lambdas for so long that it's embarrassing. The "use" idea described at 41:40 makes perfect sense to me. Is this all a lambda is?
@@1u8taheb6 Yes that is exactly what a C++ lambda is, it allows you to capture any or all variables from the enclosing scope either as references or values, and also allows you to pass in regular variables like you would to any function, you can combine these options in any way you like, they are very handy.
@@CapteinObvious Thank you so much!
yes, python lambdas are the same concept
@@koacado Python Lambdas allow only one expression. Python lambdas are also closured which means they capture the variables of the surrounding scope. The proposed "use"-block should be like an immediately invoked function that is NOT a closure.
I return to this video periodically, every year or so, as I improve my knowledge of programming concepts, each time I understand it a little bit better :)
Me too!
same
@@mfilipe7778 Good to know there are others out there who do likewise :)
@@smoothbeak Hahahah this is too much of a coincidence. I returned to the video today again and here I find a comment from you from 1 day ago.
@@amadeusk525 I agree!
My first programming language was Fortran IV in 1979. For over 30 years I got paid to write COBOL code. During the time I’ve seen languages and theories come and go. I learned PL/SQL in 1997. As a procedural programmer, switching to OOP drove me nuts for many of the reasons you mentioned. All those factories and patterns made it nearly impossible to find the business logic. I retired and came back to do pl/sql again.
COBOL and Fortran are optimized for a specific area of problems - and inside their area they are fine because they are very limited - it is difficult to write code in COBOL or Fortran which is hard to understand.
But imagine you have to implement a graphical user interface having a dynamic number windows containing complex dynamic layouts of widgets of different type (Checkboxes, Radiobuttons, Textedits) using COBOL or Fortran - it would be a nightmare...
reminds me of a time I was on a Java project, one part was an incomprehensible kludge and the developer responsible answered every question with "oh this is the (whizbang) design pattern, refer to (book) if you want to understand it." When I finally dove into the code, I found the synchronisation was completely broken, and that he didn't have a clue about multithreading but always managed to blame the resulting crashes on other things.
That is the main problem the OOP programmers just use 80% time to refactor and only 20% (or less) of the time to make the program how can they not take 5 times more?
I disagree, because cognitive complexity is a real thing. Our human brains can only hold so much in memory at once, so it is our goal to design a software architecture that optimizes for that. I can say definitively from my own experience that giant procedural code I wrote in my early years is far more difficult to read and comprehend. Once you study it for a while it becomes understandable, but only after a while. Code where I broke up into smaller pieces is much faster to read and comprehend, and I make fewer mistakes with code written like this.
Agree. I think more importantly, breaking code up into smaller pieces, sets things up for unit testing later on. Rather, what I got from this video is to try really hard to avoid needlessly increasing surface area thereby displacing complexity. But in procedural programming's defense, procedural programming often helps me build momentum until i need to break the code up to make it easier to unit test.
@@solstice5767 all the cool kids do tcr these days
Encapsulation does not fly out the window. While object B method is processing message from object A, object C cannot affect the innards of object B.
Also do not forget that, going into more of a domain driven design paradigm, eventing allows for decoupling when things get too far apart from a domain standpoint. Think about how much UI systems rely on eventing, but often still use OO concepts as well.
Yes, cognitive complexity is a real thing. But OOP does not help with it. All it does is throw the performance aspect out the window so that the problem you are trying to solve is a simpler one. Dumbing things down means you aren't solving the same problem anymore. For a given problem, if you aren't disregarding the codes performance, the problem is necessarily as complex as the problem is. You can't just willy-nilly group things up without affecting how the code runs. When you are trying to solve the more complicated version of the problem (which is not done for nothing, it is useful), thinking in OOP terms will not do anything useful.
It drives me nuts that people act like Common Lisp never existed, exclaiming "Oh, why can't I have " when it was available for the past 30+ years. Or worse -- crapping on all that Lisp provided and then embracing it as soon as someone adds curly braces instead of parentheses.
As someone relatively new to programming, I appreciate how thoroughly you managed to explain your argument in this video despite all the jargon that was involved. It also taught me more about how and when to use both procedural and OOP, so thank you for that.
Thank you so much for this. As a programmer from the days of Fortran IV through Pascal, C, Java, JavaScript etc I always felt a bit of a fraud when not writing "proper" object oriented code. I much preferred writing in a procedural way. Now I know why!!! And yes teaching OOP is a real hassle as I'm always thinking, but really I don't need to do this... 😀
Minimizing state is something I think a lot of people gradually learn over time. Function side effects make your code difficult to use if your project grows enough, function side effects can just cause an overwhelming amount of bugs.
I really disagree with people saying OOP is bad, but that's mostly because if it weren't for OOP, modding minecraft would be fuckin *impossible*, like holy hell it would suck.
@@technoturnovers7072 Coding efficient game engines and game logic would also be a pain with only functional or procedural programming.
@@daedalus6433 not really, doom classic and quake 1 are those that were written in C very optimize and modable.
@@hongkyang7107 Moddable because it's small. For bigger size ones like Minecraft it's practically impossible to do with just procedurals and even if you could, you'll need a degree on Minecraft to mod it.
@@hongkyang7107 Doom was written in Objective-C. Quake was written in QuakeC.
While I agree with most of your arguments, what you've mostly described is just bad programming practice. Procedural programming has its own pitfalls, and an inexperienced coder will turn it into crap just the same.
you missed the point that oo is not needed
Gavin Schuette neither is procedural programming. They're just two different ways of abstracting the logic so you don't have to code on the bare metal. Use whichever floats your boat.
@@gavinschuette9826 .. "to turn it into crap all the same" ? - I agree.
That is a sound premise. I've always felt that C++ is one of the better languages out there because it's a Swiss Army Knife of tools. You can write pure OOP or pure Procedural or mix and match. We even have his "use a, b {...}" example in C++ with lambdas. I think it comes down to picking the right tool for the job. Don't make everything a class, and don't make everything a 64 parameter function. Use what fits the situation best, and remember bad programmers will always choose the wrong tool.
@@djpeterson7479 why couldn't they define classes for c++ too and other functionality of other languages so we could have one single language and either use it as oop or make our own universe. Why is that so hard if c++ is as flexible as you describe? Im just asking i dont know nothing about these but whoever gonna respond to this comment just dont be a dumb
“Procedural code is better”
Me, a COBOL programmer: Is it finally my time?
Always has been ;-)
Hahhaa :) I've done COBOL in the past, don't really want to go back to it sincerely )
procedural is not an essay ! COBOL is so damn verbose, is it still ADD X TO VARIABLE or
MOVE BlahBlah TO thisOtherVariable ?
Everyone else found that Variable += X or variable =variable + x is as informative without screaming in prose.
@@TricoliciSerghei I cut my teeth on SNOBOL. It's not really SNOBOL I wouldn't want to go back to (although frankly I can't remember much that far back) it's the primitive Dec MicroVax hardware I'd rather not go back to. "Hey guys, is the tape deck free?" or "is anyone running anything at the moment?" Glad we're not heading back there!
Yes, you would be very successful working on OOP projects. You could join all the teammates not understanding and hating OOP. ;-)
Just found this video, found it really informative and I'm glad someone pointed out that OOP isn't some magical methodology you learn and becomes default. From what I found with OOP is that most cases it's either a) Trying to be an overengineered struct, or b) Trying to be an overengineered function (or group of functions). It seems most people use them because it's normally the most advanced thing they learnt and they assume "it's what the real programmers do" (Like a kid mimicking adults). Unless you specifically need inheritance or instancing (e.g., gamedev), FP makes way more sense than OOP
There's been a shift in gamedev away from OOP because objects are slow and hard to multithread in practice and class hierarchies are simply too restrictive to express complex game logic nicely.
These days many big titles are built using entity component systems which in many ways mimic relational databases, with every entity being an ID with components like position and health associated with it.
There is a other way of seeing it, and it's fairly simple : OOP is good to learn to learn how to suffer in it and know not to touch it later on, while still learning about UML stuff (which can be transposed in FP)...
This year, I took the plunge and decided to write an entire project using just functions and interfaces, in TypeScript. I had to stop and regroup a few times along the way, but the exciting thing is, the stuff I've had to learn along the way really seems more generally applicable - whereas a lot of the OO patterns and principles often seems like (as he says) a workaround or a bandaid. What you learn by committing yourself to this is much more general - simple techniques that work every time, not just for one specific problem or scenario. I have 23 years of developer experience and pushing myself to go all in on this feels like leveling up - for the first time in a long while.
I was on the fence about this video in 2016. I knew there was something true to it, but I had no idea where to start. It does have something to do with the language you're using - I was using mainly PHP, where functions and interfaces aren't really practical or useful the way they are in TypeScript.
I don't agree with absolutely everything in this video. Part of me wonders if the author hasn't changed his mind about some of the finer points in this video over the years - we all grow and learn, right? I kind of wish he would post an updated version of this presentation. Anyhow, I gave it my like today, 5 years later. 😄
I also don't agree with everything stated, particularly about functional programming. Computers just don't work that way so will never be efficient enough. I was a true believer of OOP as a C++ game programmer for many years. One day a friend suggested I try writing a game engine in C instead of C++. I was skeptical as I loved my fancy C++ systems and templates etc, but in the end I found my code to be much simpler, smaller and more elegant. Unfortunately I still have to work on OOP game engines and nobody will believe me that procedural is better.
I've started coding in a more procedural style i think, but i like to slap most stuff into a class anyway ... just with more emphasis on making many methods pure functions that don't touch the object's properties. And i reduce the object dependencies. I find when my code is hard to test, it's hard to think about, and it's hard to maintain. I also make most stuff public because idk when i might need to mess with things.
Idk. I used to put a lot of weight into designing things perfectly for extensibility or whatever & talks like this (and moreso molly rocket) have lead me more toward just ... get it done, clean it up a little bit, and don't worry so much about design. I feel like I'm explaining myself poorly though
@@reed6514 I don't really understand the point of pure functions in a class.
@@aaronmacdougall if it's generally only used by other methods in that class
@@aaronmacdougall
in my experience, people know that namespaces are good, but they are used to classes being the way you do that.
"Kernels of good ideas have been taken to holistic extremes." That's a very succinct way of pointing out a bad tendency people exhibit in many areas not just coding. I'm adding that to my "great quotes" list and attributing it to Brian Will.
Yeah, I liked that line a lot, too. (When you do the quote, though, bear in mind that the word is 'holistic'.) 😉
@@SpiritmanProductions Corrected. My mistake.
Poo-pooing autocomplete as "groping your way through" is a nonsense argument. Programming isn't about memorizing rote syntax, and having that belief is a recipe for disaster in a codebase of any significant size.
Imo a good programmer is the one that comes out with te best and most efficient solution, not the one that memorized all syntax. Autocomplete is a really good tool that improves workflow exponentialy
Absolutely. Especially if you're modifying someone else's code or code you wrote more than a year ago, autocomplete is an absolute lifesaver. This is why it takes me 3 times as long to write JS as Java.... I have to constantly look for function names in other files that were written years ago by other people.
If you can't hold what you want to do in your head before you write it, then your program is shit.
Typical computer scientist completely failing at comprehension lmao. What a terrible, bad-faith strawman.
it seems like these "oop critics" have only ever woked on single person small scale applications where they could reinvent the wheel without it taking years - its also proably a very shitty wheel if they just came up with it themselves.
It really depends on the problem. I programmed the first ever constructed OOP-language in 1970 which was SIMULA and has since then used various OOP-languages including Object Pascal, Java, Javascript, C++ and Ruby. Some problems are really difficult to tackle without objects and for some others objects just distract from the task what you actually want to achieve. Nowadays I mostly use procedural programming but occasionally add in objects for special tasks. It just boils down to common sense when to use it. I agree to that OOP is overused and often just adds confusion when one wants to understand what the code of others actually do.
This is exactly how i do OOP.
Objects when it matters / procedural for everything else.
it's the most sane path!
This is exactly how I do OOP... in Haskell.
Agreed. In some domains there are natural relations between data and methods that are never going to change, so objects work well here. In others there are cross-cutting concerns, so standalone data structures and functions work better. It's not really so difficult if you take a pragmatic, undogmatic approach.
All my apps end up with modules that contain objects, standalone data structures, and libraries of functions. Works for me.
Seems to me that it's not objects that are the issue - they have their legitimate uses. It's dogmatic OO thinking that attempts to force every domain into the OO paradigm.
What am I doing here, I dont even know what object-oriented programming is
1. Women are objects. I am object-oriented programmer, I'd know.
Hey guys, we have an object that needs to be garbage collected before we get a segfault.
ahahahaha
+The Foun Neither do I and according to the video it would be a waste of time figuring it out. O well, I don't even program yet. I ordered a book on structured BASIC today. when I am confident in making 3D Games in Basic, I plan to learn C. When I am able to make AI for Arduino in my sleep in C, I plan to learn Varilog. When I am able to synthesis instincts for robotics in Varilog, I plan to learn Assembly. When I can wright a kernel in Assembly, Then I will consider myself a programmer. It all begins with Structured Basic and game programming.
that's the most retarded language progression i've ever heard
Most programmers are perfectly capable to make a mess of any project without oop, good programmers can make a gem with any style of programming.
Must admit that oop can rely boost the clusterfuckery of a project.
I like the name spacing iOOP provides but I don't think you should be hiding internals of an object unless you actually need to. Other developers just hide things by default and cringe when they are forced to add accessors to their private variables. :/ I go the other way around untill I see abuse I'll keep accessors to private variables and remove them if they start being abused.
😂🤣🤣😂
@@carlosgarza31 when you expose variables or methods they become part of your public API. removing them in big projects later when you discover that people abuse them (probably because you have exposed something you never intended to be used by someone else than you), then you have to mark those variables and methods as deprecated, release a new version and give time (depending on the size of the project or the user base that process might take years), and then finally you can mark them private to hide them again. information hiding is one of the key principles of oop and a powerful tool in the hands of the right programmer.
Good procedure code is much easier to understand than 'good' OOP code, whatever that means
@Minori Housaki Not necessarily. It's the difference between `queue_push(q, elem)` and `q.push(elem)`, or even `queue.push(q.elem)` if you put your queue handling functions in a module.
I think Gilfoyle did a great job narrating this epic story!
I LOVED that show. It was so weird seeing him in Spiderman, though...
@@hazevt04 Wow, never recognised him, I guess that's his real voice?
As a Software Developer with over 20 years in, having developed in many different languages using multiple methodologies and patterns, this was an interesting presentation, Thank You Brian! The pitfalls you point out and complexities are absolutely real. While I don't think that falling back to the 90's is the answer, as you seem to insinuate, your argument points to something very important, which is too keep architectural and design concerns at the forefront as you write and modify code. "Bolting on" without understanding the design of a program is a fast way to create spaghetti, not matter functional or OO. I find that good program layering and being familiar with Design Patterns is crucial, and helps avoid some if not all of the pitfalls. If I am a doctor doing surgery, I better know how the organs (objects) are laid out and connected (patterns). Just knowing how everything functions is not sufficient in a complex system. As far as Agile goes, that is not a coding "thing", it is a shift in the way we think about building and maintaining things, but should also involve many non-coders in order to correctly set expectations about what gets done when. I will be watching some more of your content. There are some great titles, and you obviously have a lot of real world experience to draw from.
did you just thank yourself?
You're right to bet in design and architecture, but it doesn't change the fact that once in implementation you'll face the dilemma for full/partial encapsulation as he mentions around 25:00
- full encapsulation results in lots of additional classes created to help you deal with state (design patterns). Those classes are not related to the problem being solved. They increase complexity because it doesn't make the code any easier to read and leads to poor performance
- partial encapsulation results in spaghetti code.
@@KXBeats accidental click :-)
@@visitante-pc5zc you are certainly right. I deal with this in two ways. 1. I'm aware of most standard pattern classes and their functions. They are just part of the world I live in 2. A LOT of those classes are relegated to the framework and work magically.
1) OO was proposed as a solution to problems inherent in functional, top-down design. It was first implemented in languages like Simula (1960s) / Smalltalk (1972), Python (1991). Java wasn’t created until 1996, OO was well established by then. OO was adopted after struggling with many problems in industry, having a long theoretical debate, then doing some experimentation.
2) Brian fails to mention the problems with functional / procedural design. Top level interfaces become very fragile without encapsulation, Iterative development often leads to repeated refactoring of the higher level interfaces. Reuse within programs is common, having every consumer maintain its state for an interface adds complexity and will become unwieldy at scale. It's also simply unnecessary. And no, making the shared state global is not a good solution.
3) Test driven development is not consistent with OO. Test driven development was pretty much a fad imposed by managers. It is just design by interface applied to testing with some other rituals built in, it has more in common with functional design than OO. It suffers from the same shortcomings.
4) Brians biggest mistake is he pitches functional and OO as opposite one another. OO is not about data modeling, or data driven design, or modeling real life objects. The early theoreticians overemphasized the data modeling aspect of OO because they were contrasting it with the prevailing design style at the time, functional design. Had data modeling been popular instead, OO would have been described in terms of designing for behaviors.
OO is simply the bundling together of functions and data, with some support for minimizing duplication and access. That ideal balance of functional and data-oriented styles you keep alluding to - that is a good use of OO.
OO is also totally compatible with FRP, infact FRP is used alongside objects in most code implementations of FRP. It actually compliments OO very well as it solves most of the state management problems. And FRP is an entirely separate paradigm from functional / procedural programming that dominated in the 60s - 70s.
5) Your object graph is just an undirected mess. You claim the only alternative is a hierarchical structure, this is simply not correct. You can maintain a cyclical object graph and trigger update cycles... you can have mid-level objects that manage lifecycles, and then have an object-graph of them at scale, or a hierarchical structure of them. You can come up with bad designs for anything, this is not a criticism of the paradigm.
6) refactoring code into smaller components is a functional technique, I'm not sure where you got the association with OO. Many functional programmers advocate super-small functions that confuse people ... I've never seen OO designers advocate using tons of super small classes. That's just bad design, the fault of that is on the programmer, not the paradigm. The programmer has to choose the right granularity.
7) Inheritance is used all over the place, there's just a good and bad way to use it. There is not some uniform consensus it should be avoided. When you use inheritance you just have to make sure that the base class won't be torn in different directions. Usually this means using it on smaller components without complex responsibilities.
8) OO is meant for large projects where architectural issues matter. Java was designed specifically for this purpose. If your application is 3000 lines, you might not need OO. You might be able to get away with using global variables at that scale (though if your application grows much larger you're pushing it).
In an enterprise scale codebase where you have 10 million lines, no architect would ever design for reliance on global state, this would only result in complete chaos. You exaggerate the flaws of OO without providing a real alternative, and don't address the scalability problems in procedural programs at all, hardly.
45 minutes of someone with a nice voice telling me OOP is bad, I'm in
sounds like you just want to hear someone say something that you agree with to make you feel good, rather than actually think critically.
@@ChristopherGray00 I actually just like listening to people talk about how something sucks which I know nothing about.
rn I do know what OOP is about and its usecases and use it myself when needed
The complaints are valid but I can't agree with all the suggested fixes.
This is the story of humanity
I think the video could be more digestible if you didn't sensationalize it with "This is the most important programming video you will ever watch". Not that I am one to talk though.
+Jesus Bejarano On the other hand, it's also a good hook to get people to watch the whole presentation.
+fburton8 I think is a little bit cheap but I know was he was going for none the less and its merits.
+Jesus Bejarano With OOP being a standard of the industry for 20 years now, I don't think it's too cheap.
+JBeja M
I almost didn't watch it only because of that grandiose promise.
+Shawn McCool
What does my or his ego have to do with this?
Hey, just want to thank you for this video. This explained something really well that I’ve noticed from trying to grapple with existing codebases at work and have intuited as a kind of encumbrance.
Chasing classes down gives me a headache - when I want to understand what code does ideally I want it as sequential as is sensible so that I can follow the changes of values and potentially interrogate by pulling values out at different stages.
Having classes built on classes built on classes all the way down means that often I get to the functionality I want but then can’t find weird the values get into that piece of functionality in the first place as whilst I’ve followed the hierarchy down one leg I now need to follow it up another!
I’m a Data Engineer but have come into it without a degree in any computing field, and many of the things you’ve mentioned in this video I have come to on my own. Such as passing global variables into functions explicitly not just calling them for the sake of it.
I totally agree that there is a place for objects/classes for data structures makes sense but not making all functionality Objects. DataFrames are excellent for my work and what they do makes sense, but that doesn’t mean that style of structure makes sense for everything.
Here’s a fun project: write your own Flattenizer. Scan through a code base and inline all the scattered logic into one location as if it were procedural.
Your description of modelling real world things with OOP makes me think about a large corporation and all the pointless layers of middle management. Its like we found a way to replicate *that* relationship in our code.
Another manifestation of Conway’s law?
Middle management is a necessary “evil”
it's like taxation. we all hate it, but it's necessary
@@yousefabdelmonem3788 Well yeah. Furthermore, those necessary evils are a fundamental problems arising from scale. As the problems become bigger and harder to solve by one person, you need to make bigger and bigger compromises. It's also useful to remember when criticising governments as "less optimal" than corporations. They only have more clutter because they have harder problems to solve. Same goes for big code.
@@ekki1993 No. Thats not a fair comparison with the governments.
Some governments dont even use version control for their engineers when they make their software. People dont say governments are less optimal because of their messy solutions, people say governments are less optimal because bad managers are even harder to fire, and bad employees are the only ones that flex their worker rights. Its just inherently bad.
41:49
auto foo = [x, y]() {
// Code here
};
lambda in C++
the “use” block is a c++ lambda
I'd say MOST good java code is actually written in the style above. The things which are encapsulated in classes tend to be a) various functions which are logically grouped together and b) configuration (which is a bit like global variables as it pointed out in the video ). The reason for this is that most server side codes needs to be entirely re-entrant, so storing state (other than configuration ), is a no no. State ends up in databases (of various forms ). The only exception to this is various caching type functionality.
Thats why frontend code is more complex as backend code. But even on server there is so much algorithms and processing to do.
I had second hand experience with Java web developers back in my early day as a software engineer. I saw way too many static class methods to the point I thought to myself: Are they even doing OOP? Well I was still an OOP fanatic at that time -- my foolish younger self.
Functional programming has also its clear limits when you need mutable variables/objects/functionality because that is available only in run time: for instance user interface devices and dynamic inputs/outputs. In this case it is better and easier to code the business with dynamic objects than with monads.
Try Clojure. Great constructs for safe mutable state like atoms.
@@user26912 Clojure/FP makes things much slower and trashes memory so you have to ask yourself why are you using multithreading parallelism at all. It's done only for performance and if you waste this on clean code i don't understand. Then write just zero shared data multitasking.
lol, idiots trying to run from state, must be in a state of fear
@TheJacrespo Uhhh- no? If your buffer is 256 bytes long and the user inputs 300 bytes, just increase the size of the buffer at runtime.
This is how it is ALWAYS done- the difference is that OOP incurs a heavy performance cost by doing that stuff under the hood to account for all general cases. The garbage collector for example…
In the end, it all comes down to error handling. Monads do this at the expense of performance and cognitive overhead. I've always preferred simpler pragmatic approaches, and they seem to get the job done.
I think this is part of larger problem of adhering too strictly to a specific doctrine.
+djbanizza instead of searching for god
Yeah, it's a group think issue, as with so many other things.
I agree.....for instance one specific doctrine is OOP is bad.
Holy wars.
Object oriented programming has some good ideas, which applied intelligently can help produce good code. But when those ideas are reduced to formulas and applied blindly, you can get some very bad code. And the latter seems to be the norm for much of the industry.
When you only have a hammer, then everything looks like a nail right?
That’s why as programmers we should have a range of tools. Hammers have their uses
@@justgame5508 That's one of the reasons why I like C++, it's the whole toolbox of paradigms.
Even if there are a couple weird dangerous tools in there that can prick your fingers (i.e. placement new), it's still nice to have them.
The biggest problem I have with this video is that it completely forgoes one crucial aspect of development - testing. Not having modular components and having "God functions" that do everything in one place makes your software really hard to test, you can't really test individual bits of logic and it will become a nightmare to debug your functions if you need to make even tiny changes. Besides that this obviously introduces a very high barrier of entry for understanding your code, there is so much that a human brain can process and keep track of while reading and grasping the full picture will take much longer than if the same logic was split into separate components, and in many cases the ability for your colleagues to quickly understand what the code does without delving too deep into implementation details is really important, they in general would have other shit to do and spending an hour to understand what a function does is not a good use of their time
Working on large, highly complex projects, I've found the opposite to be true. Human brains struggle with interlocking complexity, not cardinality. We use the functional core imperative shell apporach -- even a non-technical person can open the "shell" procedure, read it top to bottom and understand what it does, hell they might even be able to make minor edits. Uncle Bob-style OOP does the opposite -- it fragments business logic into as many small, individually meaningless units as possible, making it much harder to build out a mental model in your head. Our code practically has a UML sequence diagram built-in, and our engineers love it.
@@dukiwave Its funny because I've also worked on large and complex projects, and modular code with broken down function points is absolutely crucial if you ever want to extend or maintain your code. Having smaller function blocks allows fast isolation to problematic code.
Not being able to form a mental model is more of a skill issue and can be overcome with time, code comments and documentations.
@fetherfulbiped No. Nononononono no. You have it COMPLETELY backwards.
It is infinitely easier to debug procedural code and reason about it when it is right there inlined for you. And figuring out a codebase is made HARDER by fragmentation it into 10,000 little pieces scattered everywhere. Your coworker can just as easily use a FUNCTION over some class with a built in method.
Procedural programming doesn’t mean write everything in a single function. You can still build single responsibility functions and test them.
41:55
C++ has a solution for this: lambda functions, so instead of doing "use x, y { ... }", you do "[x, y] { ... }". It won't have access to any other variable than x and y.
Edit: and it does exactly what you want, it makes clones of those variables, and if you really want to change them, you can pass them as reference for ex, with &x. It's really like having a function but without having to name one, and it can be placed anywhere.
lambda a simply nonamed function.. i.e: not reusable. why not simply make it a named function, and you can re-use it!
@@moestietabarnak C++ lambda is just an anonymous f() inside a normal f(). It can be reused inside it.
@@moestietabarnak C++ lambdas have a feature called CAPTURE. Basically they can capture variables in the scope as static values.
@@jukit3906 SO, it emulate a function with args passed by Value.. great
note : capture can also do references.
Still the same functionality as a named function, but the named function can be shared more , re-used more
and syntax is cryptic ..
@@moestietabarnak nope, it is more useful: it is dynamic (that is, a function can hold lambdas with different captures and function bodies)
Only virtual member functions are comparables.
Why is C such a rude language?
cos it has no class :V
Ba dum tiss!
C++ exposes the programmer to all variants of STDs. Versions prior to 17 make exceptions.
Who need class when you have (void*)?
With C you can shoot yourself in the foot. With C++ you can reuse the bullet. With Java you would have a Bullet factory. With C# you would have a bullet that would query the subject to harm itself.
Haha :v +10 lol
I appreciate your insight, this has allowed me to view my code from another perspective. Thank you for taking your time to express and share these ideas.
I propose the idea that, in programming, there is no good to be found through inconsistency.
Great job, Brian. I was in the business for over 35 years, and when OOP was first announced and I got my first taste of it, my thinking swirled around the realization that in the attempt to add clarity, OOP would add layers of complexity that have nothing to do with code functionality. I'm grateful to hear someone else take OOP apart on this basis. Thank you for this presentation.
Domain Driven Design (DDD) addresses--and in my opinion eradicates--all the issues described. Now DDD is language agnostic, although Evans himself points out that OO languages are usually better suited for it. I think the issue is that most people confuse the paradigm of OOP (which has matured over the years) with Java's implementation of it.
Now first I want to say I think this video makes excellent points about some of the shortcomings of old OOP paradigms, although I think the title of this video should be "Strict Encapsulation is Bad." There are, however, some points I think need to be addressed. (I've addressed test driven design in another comment).
1.) The analogy of the over-architected house is (pretty much by definition) describing a poorly planned application. He mentions walls that were prematurely established, but this is a planning flaw and not directly related to OOP. I could argue that procedural code is like a winding country road, it just randomly weaves and has no coherent sense of an actual destination. This is not a flaw with procedural programming, this is just poorly planned code.
2.) He states early on that he will ignore inheritance--rightly so--because even OO programmers warn against it. He then later references complex hierarchy graphs, which are due to inheritance. The modern approach to OOP is much more focused on dependency injection and interfaces. As far as complexity goes, the relations between objects in well designed software are only as complex as the domain requires. Any other complexity is almost certainly due to poor design choices and not the paradigm itself.
3.) "In procedural code there is only the call graph." I think this is very misleading because, in a complex domain, the call graph will only be decipherable by the software designers/engineers themselves. It is very unlikely that an "outsider" would understand the call graph outside the context of the software itself. This is why OOP is such a great tool for complex business domains. I worked on a healthcare application and one of the most valuable aspects of our design was that we could use the same graphs and flowcharts our developers were using to discuss the application with doctors (the domain experts). There was no intermediate translation needed to describe the business logic (interactions between objects) to the people who were actually in charge of implementing it (the developers). This is the point behind DDD's ubiquitous language.
4.) There definitely is a need for functionality that lives outside of objects. Anonymous functions now exist in most languages, as mentioned in many other comments. I would only like to add that creating a "container" for these functions is very much like a namespace and can actually be helpful in describing the context of certain functions and grouping them together in ways that are self-documenting.
5.) Abstraction is a term that has a specific meaning in computer science: it refers to the level at which the actual inner-workings of the computer itself are hidden from the client/programmer. For example, even a simple expression like "int y = x + 5;" is an abstraction that hides some of the implementation details (allocate a memory slot for an integer, add the value in the x memory slot with 5 and store it in the allocated memory). The abstraction of OOP is described properly in this video as "simplified interface over complex inner workings." But this is a GREAT thing! This means once an algorithm (business logic) is perfected (using test driven design) it can then be abstracted to an easier to understand concept. I'll concede that factory--and builder to a lesser extent--patterns can be sometimes hard to understand, but they are actually not meant to abstract away complexity; rather they are meant to enforce constraints. This may be unneeded complexity in small, data-driven domains with a small development team, but it is very helpful in larger teams dealing with complex domains as these patterns enforce constraints with public interfaces while hiding implementation details that only matter to the people working in that subdomain. This is similar to a business that contains forms for certain requests. The person filling out the form does not need to understand the details about how the information on the form is used, they just need to be able to fill out the form. Now, as mentioned in the video, it is definitely difficult to abstract away implementation details to the point that it can be described using real world concepts, but this is just the difficult nature of software design, not a shortcoming of OOP. This is analogous to saying it's hard to abstract away the mathematical complexity of structural mechanics, and therefore CAD software should just have the user input the center of mass and moment of inertia themselves. Abstractions and generality are probably the hardest things to master in computer science, but they are also extremely valuable when done correctly.
6.) Modularization is a great thing no matter what paradigm you use. Setting boundaries between responsibilities of your code minimizes the domino effect a change can have. The "fractionality" mentioned is actually a benefit in the long run. If you contain a big chunk of seemingly related functionality all in one place (rather than split it into smaller pieces), it becomes more likely that a small change in that big piece will bring the whole structure down. This idea of strict separation of responsibility is so fundamental to good design that you now have some of the greatest minds in software architecture (Martin Fowler, Chris Richardson, etc) focused on perfecting the microservice pattern.
7.) Lastly, "The kingdom of nouns" is no better than the kingdom of verbs that is procedural programming. The idea behind objects is that most verbs are attached to a noun--a doer if you will--or at least relate two objects together. For example, take the question posed in the video "should a message send itself?" No, a message service should send it, which describes the real world perfectly (be that the postal service, an email server, etc). I don't see what's so abstract or hard to understand about that. So the idea that OOP doesn't have verbs is utterly false, but I could argue that nouns are better described in OOP than they are in procedural code.
Now I will concede that OOP is not always the right choice and I think this video addresses the issue of people always using OOP because that's what Java was designed to implement. It is often overkill for simple data-driven applications. But OOP is absolutely necessary for complex business domains. When the application becomes very difficult to reason about using basic control flow, it helps to construct objects that represent actual domain entities and then hide all the programming minutiae inside methods that represent actual business logic. It is also very beneficial when working with a large team to define public interfaces for objects and then delegate management of more detailed functionality to smaller teams that are able to master the concepts of their specific subdomain. This is similar to how every major business operates: your sales team is not responsible for accounting. They may at times need to interact with the accounting team, but there is no need for them to understand every detail of how the accounting department operates. The problem with procedural code when it comes to complex domains is that the entire team typically needs to understand the codebase in more detail. Most of the points in this video seem to pertain to a small codebase with functionality that is easily described using the basic control flow operations that have been around almost as long as computers. But this simplicity of design often breaks down once more complex logic is needed.
Again, I want to say that this video makes some very good points about possible pitfalls in using OOP, but there are ways to avoid these pitfalls and it all comes down to taking the time to understand the problem (or domain) in great detail and carefully planning the structure of the application. If you don't want to put forth the effort to properly design the application then OOP will certainly fail you in all the ways mentioned in this video. If you're the kind of developer who gets an idea and wants to sit down and start coding immediately then OOP is not for you. But if you decide to take the time to apply an iterative, test-driven design process then OOP will reward you with a self-documenting, easy-to-understand, maintainable, and extensible codebase.
I also rue the epidemic incessant resistance to objective domain analysis that is occurring pandemically throughout humanity.
You can see it in other areas of life too. Indeed more so!
Money supply, politics, history, "education", mass media and blind acceptances of sinister, ever growing bureaucracies are the first examples that come to mind outside of engineering.
I feel like you missed some of the points. The problem with OOP is that most verbs are meant to attach to a noun. In the message example, a new coder doesn't know which class to check when it comes to finding out where messages are sent. Imagine you have the following classes: Message, Connection, MessageReader the new coder has no idea where to look exactly and any public methods in these classes will make them paranoid about side effects.
In procedural you can still have namespaces/static classes/whatever to organize where the functions live. So now imagine you just have a "Messages" namespace. All actual business logic for messages is contained in this class. It has a function like "SendMessage(Connection, MessageReader, Message)". Bam you instantly know exactly what it's doing just from the parameters it takes. That's the FREEDOM of procedural. With Nouns (usually data) and verbs(business logic) you can now arbitrarily combine the nouns together with arbitrary verbs!
Also if you like modularity pure Functions are THE MOST MODULAR thing in existence. In the above example SendMessage is pretty much a stand alone function (though not pure). But compared to OOP it way more modular. Someone could have programmed this so that it has a constructor Message(Connection). Now EVERY SINGLE METHOD in Message, might be using connection who the hell knows? That aint modular. Worse case it could be EDITING STATE in Connection. In the procedural example you know only that one SendMessage function could possibly be editing Connection.
Lastly you can get all the same benefits you brought up with things like "business people readable flow charts" and have easy to follow call-graphs if you just name space and limit scope properly. Which is EASIER to do without OOP than with OOP because of the annoying noun.verb limitations.
@@sharp7j I don't understand how this is better than something like MessageService.Send(message)
And as far as modularity goes, then you have a lot of options, like
MessageService.SetConnection(connection).SendMessage(message)
MessageService.SendTo(message, connection)
MessageService.Send(message).To(connection)
I'm also unclear as to the advantage of namespaces over static classes. And to be clear, I'm not an OOP fanatic. I use different paradigms in different scenarios and often find procedural to be the right choice for smaller applications. But it's rather dogmatic to argue that procedural code is always better than OOP. In large-scale applications, I think the encapsulation and expressiveness of OOP can be extremely valuable. The price you pay for this, as this video lays out, is that you have to be careful how you use it. But this video just lays out a bunch of problems that have been addressed (again, Domain Driven Design is a great book on this topic) and then asserts that procedural code must be the answer because OOP has flaws. I love programming in plain old C, but it's not the right choice for every project. I love messing around with functional languages like Haskell, but again it's not the right paradigm for every problem. OOP isn't the only way to code, but it is certainly a valid paradigm which is why Microsoft continues to put a lot of effort into maintaining C# (which is what a good OOP language looks like).
@@sharp7j Your example here is really nonsensical. Speaking just from an object modeling perspective, you've reversed the relationships between Message and Connection. Connections send Messages. Messages don't send themselves. Seeing Connection in a constructor for Message is an immediate code smell. If I ran across that in a code review I'd know immediately the developer didn't understand what they were doing. Whether your favor procedural code or OOP, single responsibility principals still apply. It's just as much a mistake for SendMessage to modify Connection's state as it is for Message to do so.
I'm likewise perplexed where you get the notion that a procedural approach would give you a guarantee that only a given function could be responsible for modifying the state of a data structure since that structure could be passed to any number of functions any number of times. You'd have the same problem with each call to SendMessage as you would to each new Message the Connection was passed to. All you've demonstrated is that poor engineering can be carried out by anyone.
I had the same feeling watching nthis video.
I watched the entire video and while it made me think I still can’t agree with the premise. To me a well thought out and designed OO program is easy to work on (whether you were the original programmer or not) and to extend. In my experience, the nightmares arise from programs that started as one thing and grew to something else or worse they were extended by a programmer who took shortcuts and no longer adhered to the original OO design. I’ve worked with both well designed OO and procedural programs and in my experience I have always found it so much easier and enjoyable if it’s OO.
Code debt makes all issues clear.
What I like to do, is to basically fuck around and make some prototype , and then if I have the time, I draw up some random chart on how stuff is supposed to relate to eachother, and then organize it like that. It helps to be off a few adderal pills while at it.
Im programming for fun, so i dont really know any of these industry level knowledge. Maybe people do that.
By that logic alone however, OO is difficult to extend because adhering to the original design plan while extending it in a meaningful way is exceedingly difficult.
"programs that start as one thing and grew to something else" pretty much describes every piece of commercial code ever written :-)
@@CyrilCommando good luck changing thousands of lines of procedural code to a new design when it relies on the whole sequence of operations being maintained in an arbitrary order, otherwise your global state goes out of sync.
What you'll have to do is rewrite the whole program.
Actually C++11 has the `use` feature that you show. In C++ lambdas doesn't capture (see) all variables from enclosing scope, you need to declare them explicitly in capture list. Instead of writing `use a, b { }` you end up writing `[a, b] {...}()`. Example: `const int a = 2, b = 3; const int c = [a, b] { return a + b; }();`
It's almost as if "OOB == Java" was a bad premise for the video...
avoid temptation to [&] :(
+1! I was scrolling through the comments just to see if someone posted this.
@@pcfreak1992 I missed that someone already said this and re-posted >.< My reply is a little more long winded though. I verified that Java and Ruby can't do anything close to what you can express in C++.
@@kitten-inside Java is kind of a bad language by oop standards.
I would say it is difficult to write large systems that are well structured no matter what language is used.
There is always 12 ways to do something and 10 of them are not good choices.
Having designed and implemented and programmed in many Common Lisp systems, I've always found everything newer to be worse and disappointing, and it is satisfying that functional+object programming is now finally getting the respect it deserves. Javascript seems the closest to what we had in the 1980's..which is both disturbing and hopeful.
I almost wonder if the time is right to bring back Common Lisp, or the simpler subset like NiCL that I created.
The last 10 years all I do is write microprocessor apps using C to run in chips that have memory sized in kB.
It is refreshing to be able to see the entire app in a few pages of code. I don't think Common Lisp would work in this environment or I would have tried to get it working. I see the chips have gotten big enough that they are running micropython these days...never tried it.
I really needed to see this. It's like I'm looking back at myself of the past, and this was always the obvious point I was supposed to reach but I somehow got lost along the way. This is what programming should have always been about for me but it never was, and it made me displaced because I was always looking at it from this artistic angle, but the pieces of the puzzle just never truly fit together, while nobody else ever gave a damn.
That tree, the need to send a message up the nodes to send it back down, I felt that hard, it's the kind of minor thing that leaves me stuck for days working on until it kills the passion for what I'm working on and effectively kills the thing I'm working on. The way you talk about documenting your code, organizing things, it's like staring into a mirror.
I wouldn't even know where to begin, but I would love to talk to you more about this, or even better, have some group pertaining exclusively to this kind of discussion, I cannot put into words how refreshing this would be to me. You did an excellent thing bringing light into this topic, I'm thankful.
Huzzah, thank you for writing this comment because now I don't have to. I feel exactly the same way in every particular. Fortunately for me, I am now retired and have taken up woodworking. Now I agonize over whether to use a hardening oil finish, shellac or water-based urethane.
If you haven't already seen it, you may appreciate Rob Pike's "Public Static Void" speech. Mr. Pike was one of the designers of Go, a language that Brian brings up in this video a few times himself (and does so in a positive manner).
Yep! Started out with Java in school and its good until you get out of the basic algorithms and data structures in isolation. Started making Android apps cause I figured that's how its done since I already know Java and wow, did I have this same frustrating experience to the point that I quit! I never much bothered to look into how people felt about these things and would sometimes read an article or two from people who should rather be designers with a keyboard instead of pencils, the web developer types, and would never really get anything concrete, very vague things that, at the time, I assumed I did not quite 'get' because they were simply better than me. Now that I am in a better position, it seems so surprising to me just how much contet surrounding this industry on the internet is utter pollution advising you to do things in a way that makes other things get in your way than get the job done!
Same here. I realized the importance of refactoring code but what killed my passion was thinking that everything had to fit together into labels and that it’s useful to do one thing in a million places for the future. I guess in any situation, one shouldn’t do X just for X’s sake
def looking at it from an autistic angle for sure
I’ve spent 15 years at a fortune 200 company where we have almost always been writing procedural code in OOP languages, mostly Java. We generally feel shame for it, and occasionally an overachiever builds something with “proper” OOP. Every time it ends up being a mess and more trouble than it’s worth. This has helped me finally see why that has been the case.
Your company problem stems from too many programmers who don't understand OOP.
@@pyhead9916 I think OC and OP are both under the impression that writing with objects automatically makes you a good programmer. Hopefully this is obviously wrong because a pile of crap is still a pile of crap no matter what angle you look at it. If the object collection has weird hierarchical madness OO is not to blame, uness implemented so poorly in the language itself, but instead simply is a bad programmer/team.
@@Dovenchiko I agree. Yes, using OO doesn't make you make good programs magically.
I think the authour is a good programmer, and as good programmer, doesn't really understand the problem of bad programmers. He is always comparing good procedural programs with good oo programs, and he has some really valid points. But the fact is business wants/needs a replaceable workforce and that causes a high rotation, so you are going to have some bad apples sometimes. And OO really excels limiting the amount of damage a bad programmer can do.
I think every criticism the author makes about OO is correct, but he doesn't value enough its advantages, especially the management ones, and also dismisses the problems of procedural programming, i think because he is good at it so he can manage that. And finally, no two programmers are equal: some understand better (and consequently, are better at) some paradigms, an some other understand better another paradigms.
OP’s OOP beats OC’s PP as far as OO goes
My god, the naming conventions. I would hate to read code from yallz
@@unformedvoid2223 No tool or technique prevents damage, but modularization makes the damage contained. OO forces that modularity, and that's why it is the popular choice for business. That was what i was pointing out.
Remember that the 'best' language can be one from the point of view of the programmer, but another one from the point of view of the business.
1) in my university they tought us, that classes called -manager, -service, -handler, x-er and so on are considered to be bad for the reasons that you mentioned. They are actually bad practice and no contradiction to the use of the oop paradigm.
2) the "use" keyword that you suggested is basically the semantic of lambda expressions in pure functional languages like Haskell
3)even though you tried to give ideas, how to solve the problems instead, you did not give very concrete solutions. One main reason to impose a structure using oop is to avoid "spaghetti code", where everything is connected to everything and it is not really obvious what functions are called from what scope.
I have 35 years of software development experience. OO is a tool. A very effective tool when used correctly but not so much when you slavishly allow it to dictate every little decision you make. Leverage the strengths of OO (like polymorphism, high cohesion, low coupling) but minimize its pedantic usage (like unnecessarily complex inheritance trees for the sake of OO purity). Unless you are using a pure OO language like Smalltalk, most OO languages are flexible enough to be multi-paradigm. It's a poor workmen who blames his tools, unless of course it's VB.
Agree! In my simple understanding. Objects are stateful and most distributed and multithreading approaches assume stateless where data is separated from the handlers. OOP still is a working horse of all internal logic.
OO is a tool, but in language like Java uses OOP all over the places wherever it's not even necessary.
New programmer checking the codebase gets confused when OOP used in places when it's not even necessary, with inheritance its just nonsense abstraction.
No, some tools really are garbage.
There is a programming tool that, when used incorrectly, causes compiling errors (and the code won't compile). No damage then. When used correctly, causes the debugger to be practically useless. Have you tried it? And yes, it is FP :-).
And the problem is there are too many purist that do exactly what you say it should not be done. I know a programmer and even have to work with that instead of making a single switch with 5 different settings will make 6 factories that will contain 15 classes to do it. If you are doing that just for a single switch you are bad on the head and heavily. And the switch on itself has only 1 or 2 lines of code. If you say "that is hard to read" trully cannot understand how compared to a total of 21 clases.
This was an absolutely fascinating video. I intended to just watch the start to get a vibe for the idea here and ended up watching the whole thing. I'm a SD student and we are using Java as our teaching language, and I personally am currently trying to square the circle; to extricate the cross-over from encapsulated objects. Something you said in the end about programmers taking good ideas to holistic extremes is something that even someone at my level can appreciate. A lot of the questions I have asked my lecturer, the ones that did not receive "good" answers anyway, relate to this idea. Super interesting video.
Oh, heck yeah. I normally hit the skip hard on lines like "One of the most important yada-yada." There was just enough nuance to promise the goods, and yeah. Maybe not a life-changer, but I'm with you!
Lol... the video just speaks above the audiences heads, where most listening aren't educated / experienced enough to refute the horrible ideas.
OO is meant for large projects where architectural issues matter. Java was designed for enterprise-scale applications. It was designed by Sun microsystems specifically for this purpose. If your application is 3000 lines, you might not need OO. You might be able to get away with using global variables at that scale (though if your application grows much larger you're pushing it).
In an enterprise scale codebase where you have 10 million lines, no architect would ever design for reliance on global state, this would only result in complete chaos. There would be so many unpredictable synchronization issues it would destroy your application. This has been known since the earliest days of high level languages...
Most of the people in these comments agreeing with Brian are students, or people with 1 or 2 years experience, small scale web developers, electrical engineers / other scientists working only with C, for whom programming is not their main specialty... etc. I can guarantee you very few of these people are working at scale, these ideas simply do not work at scale. Which is why they are not used...
@@ianrust3785 Linux is millions and millions lines of code in a procedural style. It has more than half a million contributors and has been maintained for 20+ years. Every AAA game that pushes the boundaries is made in a procedural/data oriented fashion because OOP kills performance. These two examples absolutely dwarf any enterprise app by an order of magnitude in complexity.
@@cavasnel Despite the fact Linux is written in C, the code in the kernel leverages many tricks of the C language to mimic OO. For a detailed description of these various tricks and their use within the Linux kernel google "is linux written in procedural style".
Also.... according to The Linux Foundation, since 2005, “some 14,000 individual developers from over 1,300 different companies have contributed to the kernel.”. Most of those contributions are drivers, ... but it's not half a million, sorry.
@@cavasnel Games are typically written with components based architectures these days, most are written with Unity or using various C++ libraries. Try describing a real game and how it was architected in detail. You just claimed the entire game was written in procedural style. What game was designed this way...? And what was the reason - was memory severely limited on the platform for some particular reason? This sort of problem was far more common in the days of 16-bit SNES than it is now. If you’re writing your own custom game engine or image processing functions, some piece of code that executes millions of times for a single update to the UI, then extra layers of abstraction could start to degrade performance. If performance is critical and resources are very limited, eliminating those layers will be one of two dozen optimizations you might do here - because you’re dealing with such a critical piece of code. The entire program would not be coded this way, this engine code will be tucked and the rest of the application will only interact with it through an API. In your main application, the part of the code that is user-input driven (not compute bound), memory management is the only thing that matters. A trip to disk takes about 8 orders of magnitude longer than a compute operation, processing extra layers is a compute operation… eliminating a single call to those layers might amount to a few microseconds worth of difference in performance, maybe. Depends on what’s in the cache. Depending on the compiler, it may actually make no difference whatsoever… To optimize in such a way any user-input driven code is purely idiotic, it is compromising the architecture for microseconds worth of performance. Meanwhile if you need to reuse any code, like an object model or behavior, in the entire game... you have to begin maintaining duplicated or global state for your code, and now you have a giant mess on your hands.
I don't doubt some developers have done it, developers do all kinds of stupid things, but this is not a good design by any stretch.
A more valid concern is potentially pulling in more from standard libraries than you need, increasing memory consumption, so some developers write their own custom algorithms that reimplement some things from standard libraries... but even this is usually a bad idea, the compiler recognizes & optimizes code in standard libraries, the code actually executes 3-4x faster than custom code. But this isn't a paradigm problem, it just has to do with memory management.
Carry onward
Fantastic. And brave. I started with BASIC and machine code (Z80) in the eighties. We knew freedom.
I write apps for Android in Java these days. I hate Java.
Then do something else. Is someone forcing you to use high level languages with strict compilers?
I have my complaints about Java. For instance, not having unsigned primitives is almost a deal breaker to me. However, It has some amazing features that you can't get from other languages without convoluted workarounds.
I hate Java too. I was in a Java User's Group for a while in the 90s when it seemed like the language was changing every week. Then I found other there were other web programming languages like PHP which didn't require a half dozen XML files to write a Hello World.
@@MrCmon113 Yes, it's called having to go to work to earn money to feed your family and eating whatever shit your boss throws at you.
you can always use Kotlin, it's like java but better
42:51 C++:
int a = [x,y]() -> (int) {
..
return 3;
} ();
It does exactly what you are looking for..
;) ...
and javascript:
var a = 3, b = 4;
var c = function(d, e){
return d + e;
}(a, b);
console.log(c);
php also...
@@zegarek840525 Even easier in ES 6 with a pointer function!
In C++14, you don't even have to specify the return type yourself :)
He was saying that the inner scope should not be able to access any data that's not explicitly passed in to it. This would let the reader temporarily forget all about any variable which is not immediately necessary for the next logical step. I don't know about c++ but local anonymous functions / closures in both php and js are not restricted in this way.
He also mentions that the data is always passed by copy - not by reference - ensuring immutability.
@@zegarek840525 No... You have the outside scope inside the function and thus not the same thing. Did you even watch the video?
This is such a profound video. I have always struggled with trying to force my code into an OOP design just for the sake of doing it "right".
I find I understand what Brian is saying about the tree graph and refactoring problem. In OOP class in Java I felt I totally didn’t do it right. I remember in Java I had to pass data all the way back up a hierarchy of classes and down again instead of just communicating directly. I had to add extra parameters in the class chain just to pass one or two pieces of data. It made looking at the UML diagram obfuscated.
Other than this, there would be a lot to pick apart and a lot is about preference and can’t be proven whether one way is better or not. For example Brian you said how long functions are okay and better than abstracting away short functions because the reader of the code base has to sift through the names of the functions and that you can just comment the longer functions well instead. Orrr, you can just abstract the long functions away into the smaller ones and document your code well so that the reader can quickly glance at the comments and see which functions are sub-functions of others, and good IDEs like VS will do this for you, that is tell the reader where the function is referenced or called. This is a matter of style. It can’t be proven that one way is intrinsically better than the other. You either comment your long functions well or you comment your short functions well, but you document either way.
The reason we break code up into smaller chunks is to forgo the law that complexity is always more error-prone. Breaking tasks into smaller chunks eschews accidentally updating a variable that a sub-task doesn’t need to touch.
You threw in Agile, Test-driven development, and dependency-injection as bandaids for OOP. What?? I think you’re reaching. Maybe DI, but Agile has nothing to do with OOP. You can “fail-fast” in procedural and functional languages too.
One more comment is on your saying that there’s no good way to connect behavior with a class, like should you send by a sender or receiver function. You basically said there’s multiple ways to do it, so therefore there is no right way to do it. Well… just pick the best one then-in the flow of the data. Sender. A sender method will import and call the method that does the sending. If you receive, then that would be an event-listener.
The premise of this video is that we need to account for if data can be sent in multiple directions and that state… basically can’t be managed, so why bother at all. Data generally flows bilaterally though, and you’re always going to have to have some “god-object” to manage the state of any subset of data, which classes encapsulate, or in other words organize, whether it be a class or a namespace, or else this objection about maintaining “state,” or the state of your program, will always be unmanageable.
This is my opinion anyway. I am certainly not wholly informed or don’t have experience with procedural I guess, but I do with OOP and functional. I appreciate the exercise of freedom of thought and opinion, but I think some things in this video are reaching.
I really liked this video! At the time of writing, so did more than 3 quarters of the people who gave it thumbs, which is really reassuring after reading through the horrible comment section. I have to admit that I related to every single example that was given, and it brought back painful memories of so many years of writing code. Thanks Brian! Now I don't have to feel guilty any more, and I have the courage to say it! :)
if a video itself states "this is the most important video you will ever watch" (twice) it's already super suspicious
so.. i did get some way into the video. what i found strange is that first you state that oop shouldn't get all the credit for java has done right since a lot of it is not fundamentally about oop but then you go on to criticise oop for some things which are not per se a problem of oop but rather how things are done in best practice. i am currently at 22:17 and you just complained about shared state and how you shouldn't pass references to other objects (although i dont see why this is a problem when done well. sometimes shared state can help coordination of classes). you could just structure your oop code to always just pass copies of your objects like you do in functional programming.
anyways, i wish you had used a different intro for your video, as it doesn't suggest that anything of non-subjective nature is going to follow..
@@homeXstone Joshua Bloch Effective Java third edition. Good read.
YES 100% yes. you have managed to expertly put into words my exact frustrations with new coding paradigms and frameworks. I would much rather write functions instead of having weird datatypes and meaningless required declarations strewn all across multiple files that is hard to follow.
41:00 This is exactly how C++ implements lambdas :)
Yes.
Great job on this video. I have come to most of the same conclusions from my own experience. I now love a functional style where most functions are pure. Greatly simplifies everything.
True, the C way is almost perfect, but there are higher level features which would make it better. Generic code generation is a good thing to have which saves you a lot of grunt work and bizarre macros, the issue is the C++ way this is done through the template system is a bit limiting and there's no way for example to store templates as DLLs as they don't really "exist" as something in binary. That's the other thing as well the inconsistency and complexity in which the C++ ABI is setup.
This sounds more an indictment of lack of design than OOP. I programmed both procedural and OOP. Having learned OOP using Smalltalk and Lisp, by the time I learned C++ and Java, I came at OOP programming very differently than those who started with the latter. Harlan Mills repeatedly demonstrated the value of mathematically proven design, yet the business call for creating a prototypes eventually overwhelmed the idea in the industry. Agile methodology further allows the programmer to really skip design and figure the whole thing out later. Early Java libraries and even MFC went a bit overboard on the use of inheritance and that started to turn programmers off to the idea. Nowadays, I see more code written with a procedural ideal, but using objects. Component frameworks like Spring and entity frameworks like Hibernate helped feed the creation of discrete objects, instead of the OOP concepts that actual delivery the power.
The fallacy of his argument is the "truth" that a strictly OO coded program is a tree with only descending communication. The whole concept is that any two objects can communicate, each deciding whether to respond or not. His arguments are what lead C++ to have friend methods. Encapsulation is broken for convenience.
+Bryon Lape , I agree with the video content and also with your comments. The idea of class instances containing references to other instances for communication is obsolete. In essence there is no need for a complex tree of instances that are limited in their ability to interact with other code. Because there is not a reference availble. The idea I used long time was to implement Runtime Contexts (thread, request, session, application instance, server side instance, cluster instance) which can be used by any class. So any piece of code can be easily accessed. Even a procedure (static function) might hide where the reference to the instance is stored for further retrieval. So omni directional communication is possible.
You can't design so future changes don't ultimately destroy decoupling.
exactly Dad Bryon Lape - exactly.
Except one thing I'll add is that a cyclical object graph driven by some reactive data bindings is an industry standard at this point.
Formula for making this video: come up with the worst OO design possible, use this to criticize the whole paradigm, ignore all problems in your alternative suggestions, rinse and repeat.
Interesting. I made a total-conversion ARPG mod for an RTS recently and found dealing with class organization to be the key struggle. The code, though powerful in the context of the mod, ended up as spaghetti as anything outside of the object's scope required additional subclasses or types to control things since they didn't necessarily belong to that class, or could exist in one class or another since they shared featurization (in game dev, this is probably a common issue as features are added). For every added layer of complexity I'd wager the effort to make a change increased by 3 to 4x (which is pretty crazy). It starts to make sense why the AAA titles coming out today are a mess, given the size of the stack at play and the corporations fiddling with them. Nightmare fuel.
A lot of games today are entity based, not OOP, so instead of a class for each NPC, the NPCs are just a number, that refers back to a container that has all the data of all NPCs, that is then updated by a "NPC" algorithm. I'm not familiar with those systems but that's the gist of it.
@@Vitorruy1 Does this mean that OOP languages are wrappers for the entity engine? i.e. Unity and Unreal?
@@MechShark I have no idea, Unity seems to be OOP, but no one in their right mind would write a brand new engine using OOP these days.
edit: apparently unity was refactored to an entity system
Unity DOTS is an ECS framework which represent the phrase _"composition over inheritance."_ I don't use Unity, btw.
16:37 "What is the one abstraction bigger than a function and bigger than a data type?" Mathematical categories, of course.
They provide an even higher level of abstraction than objects in OOP, since they are able to describe more than one object, and they remove the need to play the responsibility game that is referenced at 27:22. They can be most commonly implemented in software as a collection of data structures and a collection of pure functions that map between the data structures, with the additional constraint that functions are associative and data structures have some concept of persistent self-equality (so that there is a "do nothing" or "identity" operation that lets you leave a data structure alone without it changing its nature). This by itself is enough to meet the formal definition.
You could implement categories as namespaces or as class constructs within code. Implementing a category using a namespace construct allows greater reuse across software projects, since homsets within the category can be stored within separate files. Since a homset captures a very small, perhaps even atomic aspect of interdependency, moving a single file across software projects will need only require importing the minimal set of other files needed to support it.
If a category is implemented using a class construct, then there are many concepts within category theory that can be used to precisely describe many of the aspects of Object Oriented Programming that people actually like. For instance, categories often require other categories so that their behavior can be implemented. The categories that are used to implement another category can be tracked as dependencies passed to a constructor, which itself forms a category of categories, where objects are categories and arrows are constructors. This is similar to the way OOP makes use of dependency injection. In another example, certain kinds of functors within category theory can also be implemented using inheritance and interfaces.
A design approach based around categories shows many of the positive attributes that people have found with OOP (dependency injection, code reuse, programming to interfaces) while eliminating many of the problematic attributes (state obscuration, state management, the banana-jungle problem), and it is based on a solid mathematical foundation that OOP lacks, which removes the need to discuss what it is we're even talking about, as is done for OOP starting at 1:01. The approach can be considered consistent with the "object oriented & functional" design approach mentioned at 6:12.
I'm going to need a reminder to read this reply in full...
@@nullpoint3346 same
@@nullpoint3346 yeah
@@nullpoint3346 I hereby remind you
There seems to be a good reason why the term "abstract nonsense" is used about mathematical methods used with category theory...
I just wonder how it is expected that the average Joe Coder - who may barely have grasped recursion as he drifted from HTML, CSS and copypasting and hacking JS by trial and error, to doing a little backend programming in Php - will understand category theory, and to what extent he should be expected to understand it?
Thank you for this video. Your point of view was very interesting. Though, the problems you adress can all be avoided, so I will keep using OOP.
I don't think they really can. Writing OO code that avoids this problem necessarily means embracing the procedural *mindset.* You have to abandon trying to make a taxonomy out of your code and instead write it so that the _flow of control,_ the _procedure,_ becomes the priority. Sure, you might have some data objects with specific functionality attached out of convenience, but that isn't widespread application of the OO paradigm. Instead, your classes (other than the ones that primarily hold data for passing around) end up becoming mere containers for procedure without state, which is only necessary because the language requires it. This does have the benefit of enabling faking for tests in OO languages, but again, this is a kludge around the missing infrastructure around first class, stand alone functions. (E.g., contrast with Python, where you can fake functions directly.) That isn't OO; that's just procedural in an OO heavy language.
This argument applies precisely to bureaucracy, it’s tendency to form chains of command, and why that doesn’t work so well.
Onus probandi
OOP is the ultimate bureaucracy!
Yeah, as a computer science student that took the supposed rules of the OOP paradigm and the supposed virtues it provides, I often wondered as the things I programmed became more complex just how on earth people tolerated writing code like this. I could whip up something fairly complex quickly with a procedural approach, but trying to force everything into boxes I was supposed to keep separate just paralysed me and I spent more time shuffling those boxes around to keep things neat than getting stuff done. All I could think was "this is so much extra work, but I guess it's good practice. It'll be worth it in the future, etc."
So, can't say I was particularly impressed when I started looking into real world OOP source code and found out none of these rules are actually followed and people just threw around references to objects that were specifically there to allow global access to references of virtually every other object in a giant clusterfuck void of hierarchy. What's the point?! Now I still have to micromanage state myself, because your code will let me do things I probably shouldn't do, but I have no chance of understanding what's actually going on because most of your code is still pretending to respect encapsulation and is just juggling references to references which is now nothing more than a maze between me and the actual data or functionality.
I’m quite literally the exact opposite. How do people not write code and keep things separate.
@@coreyaruecker Depends what your version of "keep things separate" is. I don't understand how people write code that seems to contain more boilerplate than logic.
Like, imagine code that has a comment every other line explaining verbosely what the code's doing in plain English. (Obviously I'm not comparing OOP to comments in terms of functionality, just using it as an extreme example.) Do the comments hurt the functionality? No. Is it useful? Maybe to someone who doesn't know how to read the code itself, or gets lost in it without plain English. Is it pointless busywork at best or a cluttered mess at worst to someone who doesn't need the comments? Absolutely.
To be clear again, I'm not trying to say OOP practices are like writing needless comments. I'm just trying to illustrate how it can sometimes *feel* to work in OOP code written with a level of granularity that feels unnecessary. Like having an entire directory designed to contain a single file because to some people that's "more organised," when to others it's needless convolution.
A computer science student doesn't really know what a production code base looks like.
@@harrywang4769Oh I know. Only half joking when I say I realised I wasn't cut out for working in the industry when I figured out most of the job is about working at the level of the lowest common denominator.
assuming people knowledge based on academic levels is just arrogance. maybe this person knows much more than you think@@harrywang4769
God this is my favorite video in the universe. Perfectly explains the frustration and skepticism I've constantly encountered browsing OOP codebases.
I thought I was just dumb for not being able to figure out which classes should be factories and which managers and how to structure adaptors
@@koacado Almost surely not - so often you find out a factory or manager or ??? is actually just a function, or a function with 2 arguments that should be passed on to another function, doesn't even need state.
Translation: confirms my opinions, love it!
@@koacado I find a good role of thumb for naming is that nothing should ever be a manager because it implies you don't even know what it does.
@@trejkaz yeah that's a good rule
This lecture is a masterpiece. I've come back to it over and over again and I credit it for really being my breakthrough to how to write OOP well.
What I like about this video, is that you really know your stuff. You're not someone who dipped into OOP, didn't like it, and posted a rant, rather, you've done extensive research into this field and have obviously put deep thought into everything you shared.
I'll also note, when you first started talking about not being afraid of large functions, I was ready to pull out my own pitchfork. But then, as you started talking about breaking up that large function into sections, and making a bunch of small scopes within it, I realized the beauty of what you were proposing. Now, we have a single function we can go to to get a sense of how the logic flows, but we've also broken it up internally, so it's easy to understand individual parts of the function. That "use block" syntax would be really helpful for facilitating this sort of pattern.
In fact, now that I think about it, I'm realizing that I've done this sort of thing before. I created a bunch of sub-functions within a bigger function to section it off, then I created a scope at the end where I invoked the different sub-functions and passed parameters around. I chose to put all of the data in a single scope at the end, to force each sub-function to explicitly list the parameters it needed. It wasn't allowed to just grab them from the enclosing scope.
function doThing() {
function taskA() {
...
}
function taskB(a) {
...
}
function taskC(a, b) {
...
}
{
const a = taskA()
const b = taskB(a)
return taskC(a, b)
}
}
A use-block syntax would have made this sort of thing much easier to do.
The exact same thing can be done with static methods in a Java OOP class.
The irony in that design (and I know its very popular in js) is that you’re basically emulating classes
@@KebunH I haven't seen this sort of design used in JavaScript, nor do I see the resemblence with classes (you can emulate what I did here with classes, but IMO, it's wrong to use a class when there's no need for state or dependency inversion is not being used). In fact, I rarely see people create new bare scopes using "{}". Did you by chance mistake that code snippet for the factory pattern, which I know is popular in JavaScript?
At the end, I did this:
{ //
@@thescottyjam8906 its similar to object oriented programming in the sense that the way your function resolves to its final value is by using internal encapsulated (private if you will) state and functions, albeit only temporary. Its not the best example of class emulation indeed as its not a reference but a value thats returned, but it still does something very oop like
@@KebunH Ok, I see what you're saying. I will mention that this type of temporary encapsulation and state-management isn't unique to OOP or classes. You can take the purist functional language, like Haskell, and find this sort of pattern all over the place. People will use the "let-in" construct to declare functions inside of functions, use them, and return a final value. (and, indeed, the use of functions inside of functions is common in JavaScript like you said, it's just the specific use of the sub-scope at the end that I felt was unique).
So, we could say that this pattern is emulating something you can do with classes. Or, perhaps, we could say that when people use classes in this way (where you create an instance just to call a single function on it and throw the instance away), they're emulating this commonly found functional pattern ;).
Either way, it is an interesting correlation.
Side note: your idea of anonymous functions that don't capture any enclosing scope is very much possible in C++ by just using a lambda with the empty brackets [] which does exactly that. It's a cool idea to structure long functions using that, although I'm not sure how I feel about the arguments being at the end rather than the beginning. The use syntax would be nicer. Personally, in C++, I'd rather just declare some temporary variables right before setting up the lambda, then have it capture just those variables (so that everything will be up front).
Well, for the thing he proposed you just use capture by value, and the syntax is almost identical to his use: [x, y]() { doStuff(); }();
Using lambdas is also technically functional style.
Lambas in C++ came later, the video is older ig. Btw, I like this pattern (over just creating a scope with braces) because it allows early return (no if else gymanstics needed)
@@sanjarcodeThis Video is 2016, lambdas are C++11
A lambda "captures" the variables that it uses, but it can access some variables without capturing them. They're not necessarily trivial to factor out.
I was a Philosophy major first in college, then changed to Comp Sci. The part about 'abstract' actually made me laugh, because I literally thought that every single day.
my thought though is "abstract" means abstract to the computer, not the human
wasm is abtract because it's not like the real machine
@@unmellowthegamer8946 Abstract because you're extracting the core shared essential process from the specific implementations - just like abstract images aim to capture the core essence of an image, or a set of images. The word means exactly the same thing in both contexts. The language that surrounds programming can be daunting at first, but if you apply some rigour to your efforts to apprehend the meanings and purpose behind the names chosen for concepts it begins to make sense. I found such an aha! moments to be real breakthroughs in both my comprehension and my programming capabilities.
The definition of abstract is "existing in thought or as an idea but not having a physical or concrete existence." This perfectly described what OOP attempts to achieve. I dont see the problem.
It still has the same core meaning in both philosophy and programming but usually it's something that takes a lot of thought to make intuitive in the former, yet it's something that makes intuition easier in the latter.
Many portable operating systems have something called a Hardware Abstraction Layer. Most folks I knew called it a Hardware Obstruction Layer. 😉
This is a very interesting video. I had been writing procedural code for a long time. When i learned oop and design patterns, i never looked back to procedural programming. For sure it will work whichever paradigm you use, but a properly written oop code will give you more flexibility.
I revamped multiple systems (products that are currently out in the market with revenues in millions of dollars) for different companies previously written in procedural languages into object oriented and it would always result in a more flexible software that can easily cope up with future changes.
When i revamp procedural codes, 99% of the times is that they are very rigid and will require very invasive approaches to implement new features and make things more flexible.
I personally dislike OOP for many of the same reasons this video enumerates. However, I think the case you describe is the best possible use of OOP (or any (enforced, or accepted) formalized programming system. When you approach the rigid procedural system you’re refactoring (I assume), the “pieces” (“nouns”, “verbs”, higher level logic, etc) are already known and sketched out.
For me, the bigger problem isn’t with OOP _exactly_ it’s with the attempt to design before the shape is known.
I can agree with most of the things you say in this video, except for the part about large functions. If you inline all code into large functions and just separate it by comments, you force me to read all this code or to constantly skip over it when I try to understand the higher-ranking logic of a function. The idea to split code into many small functions is that I don't have to read all that code unless I have to touch it for some reason and then I only need to read the code I want to touch. If the name of a function tells me exactly what this function does and I have no reason to assume that this function does not work correctly, I won't look at its code at all! Even if just called once in the entire program, why would I want a base64 decoder to be inline instead of having a call to "base64decode(...)"? I know that it does. Base64 data goes in, decoded data comes out, no need to see its code.
Yes, this. Using functions to spread apart your code can make it so much easier to read, when done correctly. Sure, if you're looking for a particular functionality, you will need to CTRL+F to find it. But if each function is longer than 5 lines of code, leafing through pages of functions to find exactly what the application is even _doing_ on the high level is annoying and inefficient.
To add to this, you should NOT be afraid to make functions with long names, there is no real downside as long as it's smaller than ~10 words. A long and descriptive function name means that anyone who reads it knows exactly what it's doing, without needing to read some long extra sentence/paragraph in a comment. Abstract function names like "Build()" force a reviewer to actually look in multiple places to figure out exactly what the duck the code does.
Both of these ideas complement each-other and lead to code that is both categorized neatly and quickly understood on the high level. However, you still need to use your own discretion on deciding what kinds of code to split up into functions like this. Sometimes it's just not worth it, as the video author describes.
@@LordOfLemon I don't have professional experience with coding yet so I might be wrong but I haven't seen situations that work better without separate functions outside smaller pieces of code. For a program that does multiple things, separating methods and commenting makes it much easier to read and understand. If you have an example of a long code where functions would be the worse option please do share
@@mistzzz4683 I also find myself disagreeing with the video's suggestion, but one place I prefer inlining is when the function in question is tiny, obvious, and only called in one or two places. For example, it's way less readable to see a call like "n = timesThree(n)" instead of just "n *= 3". I've seen real examples almost that bad. I love small functions, but at the point where the function name is just an English transcription of it's entire one-line body, I'm actually going to stop and read it because I'll be wondering what was so hard about this that you had to define a function.
@Mist zzz I have professional experience, and the answer is just that there are a lot of bad coders out there. I once spent an entire summer refactoring A SINGLE FUNCTION. It was thousands of lines long, and it meant I had to constantly delve deep into various levels of logic to understand the code. Some flows of code are long and complex. In this case it was the handling of a submission for a new user in a specific system. If you were A, you were gonna get Y and Z, if you were B, you were gonna get X. If you were C you were gonna get X and Z. And then with A B and C you were gonna do some common setup. Now, I was originally only going to add a new case D, but I couldn't. The code was so messy, I couldn't interject W into the code, because everything was mangled.
That's why you want a function that describes what you're doing. submitForm() { createUser(); setupUser(); giveUserStuff(); sendOutMessagesToUser(); }. If I am going to give a new type of user new stuff, I need to change giveUserStuff() function. This is how you can see that you can drill down into the depth of complexity a lot better. Comments do not do this in the same way, because the code is still there. Good code is also comment enough. Show me what you're trying to do, then show me how you're doing it. Even when you're doing it, show me what you're drying to do, then show me at the deepest most complex level what you're doing.
In my case, there was an API that was being used, so all over this original function, there were these extremely complex API-calls that were being set up in complex ways. Instead of saying giveUserASubscription(subscriptionType), there was a huge chunk that set up an API-call and just used the result. When it's my first time opening that code, I am not able to quickly scan that and ascertain that this part of the code is the API-call that's giving a user a subscription. I have to scan through low-level API-setup and see that the API call is something related to a subscription. That's why you first show what you're doing, THEN in a new function, show how you do that. That means that if I am giving the new userType D a new subscriptionType N, then I don't need to care about how you giveUserASubscription, I can just call giveUserASubscription(N). This means that it's not only good at showing you in coarse terms what the code is doing, you're also allowing me to not have to even relate to the lower levels of the code, if I don't need to. Say that later someone needs to implement a newer version of the API-call, then they can just find where all those API calls are used, and change how they're called. Then you don't need to relate to what you're trying to do. That's why abstracting code makes all tasks besides a complete rewrite a lot more easy. Maintaining code means that you should have to deal with as little as code as possible when you're doing a change. That way you lower cognitive load, which in turn means less chance of error. You also touch fewer lines of code, again making it less error-prone.
Yes. And smaller functions make the code better testable. If you have assumptions about the state of the relevant variables while entering a certain code block, and the desired result after passing it, it is testable if that code block has a name.
Finding descriptive names, which make a longer comment superflous, is often time consuming but it teaches you in reflecting upon your code and what the parts of it do. If you are able to leave a helpfull comment, deriving a name from that should be possible.
Good video. I was long mired in the OOP world, but a few years ago I started delving into the functional and procedural side of things. At this point, I'd say that I FAR prefer a functional or procedural approach. To me, moving away from classical OOP has been liberating and empowering (and I'd say it's also helped me when I do need to work in OOP codebases). It's changed the way I work. I hardly ever need to step through my own code with a debugger, these days. And I no longer care much about Intellisense-like tools which I used to think of as an absolute requirement -- I no longer need them the way I used to. I know there are many who are deeply invested in the OOP paradigm, think of it positively, and want to defend it (I was one of those programmers). To anyone in that position, I'd encourage you to investigate functional and procedural paradigms with an open mind. You might be surprised to find that you are a happier and better programmer moving with a move away from the OOP world. If not, and you still prefer OOP, there's no loss. Learning different ways of thinking about code will help you in the OOP world, too, and make you a better programmer overall.
I am a big proponent of OOP, but I came to watch this with an open mind hoping to learn something. It turns out we agree on many points, but probably that's because I'm not taking OOP to the extreme, for example I don't agree with / don't do these:
* Some OOP folks think there can be no such thing as simple types: even int and double must be a class.
* Taking encapsulation too far, where you can't even access simple types from an object directly, no it must use setters and getters.
* Trying to force everything into objects even when it doesn't really fit. Some things really are just plain data and plain functions.
There are however some things about the other end of the spectrum that I find absolutely godawful:
* Having your state scattered into a sea of global variables that are read and/or modified by functions which don't take them as a arguments, and in the worst case even having those globals be used across files.
* Having lots of separate variables not organized into structs or classes, and a set of functions not organized into namespaces or classes that you have to guess what belongs with what and which are intended to be used together.
* Having long complex procedures rather than capturing key parts into functions or at least scopes. This is analogous to writing a wall of text rather than using paragraphs.
Video should be called OOPs, my bad.
lol
In your country, you like to create "object," but in Java's "Inversion Of Control" paradigm, object create YOU!
nice one
OOPs i did it again
OOPsy poopsy
Although I'm mostly in favor of OOP, there are some really great points mentioned in this video, thanks!
For a while I thought I couldn't learn programming because people needlessly overcomplicate it on purpose.
After watching this video, I now realize that I can't learn programming because people needlessly overcomplicate it on purpose.
People dont overcomplicate on purpose, that just doesnt happen. No one in the world is inherently evil, theyre just doing what seems right from their perspective.
The problem isnt really overcomplication, it's that programming itself is complicated. Even with that, I dont think that being "unable to learn" it is ever correct.
If you just stay curious and learn a little bit at a time and keep experimenting, you can probably become so much more skilled at it over time than you thought you ever could be.
It takes effort but I think youre wrong in saying you cant.
Usually, it gets forced by some management, or some practices people learn. Extremely well designed object oriented programming is very easy to read. Much easier than functional or plain old C, because the functions are with the data.
Learn to program in C. It has a name of being hard, but that's incorrect imo. It's hard to write large software packages in C, but the language itself is pretty straightforward. And a great thing about learning that way is that sooner or later you will run into the limitations of the language, and then you will easily pick up on more abstract ideas like object-oriented programming. And you'll learn it not as the one true way, but as a technique that is in some cases helpful.
"Bundle globals into structs/records/classes. But be careful, there's an art to how you bundle these things together. "
It's called object-oriented programming
Also, I have an issue with this advice "Don't be scared of long functions".
The reason why I generally have to avoid them is because I'll need to reuse the "components" of the function. There's a myriad of business areas with similar/overlapping business processes that needs to be mapped like this. And Brian's advice doesn't provide any solution to this. Is the solution to have copies of nested functions everywhere. How do you change your code later?
Generally I think he's on point with the problems of stateful, OOP. But I don't think he quite manages to point to any useful remedies. Programming is hard, because mapping complex domains is just hard. Real life is messy, and code that tries to reflect that will also, inevitably, be messy.
I clicked on this video with skepticism because the video title is so "black and white". But unlike the title, the contents of the video is not so black & white. I think a lot of people who thumbs-down this video actually aren't watching the full thing. For your convenience, the 1st half-hour or so is all background and setup and analyzing the problems. He gets to his list of proposed best-practices (which are quite reasonable) at time index 34:46.
His evidence is anecdotal and many of the issues can be attributed to incompetent programmers as opposed to their paradigm.
Also some of the advice is really bad and I hope I don't have to maintain code by people who took that advice to heart. Nested functions in favour of-you know- non-nested ones? Relying on comments? Like anyone writes those where they're needed. Properly named functions are the natural comments and make things more readable in the process. For the love of Zeus don't use large nested functions just because you "shouldn't be afraid of large functions". Give my scroll wheel a break.
The tl;dr of this video is no one even does the crap he says is bad about OOP. At least no one any good does.
He's saying OOP is bad. Sometimes things genuinely are black and white.
It's hard to take anyone seriously when they start with this is the best video ever.
It doesn't help when having decades of dealing with people who say that and you find out most of them are blowing smoke. It ruins any attempt at trying to make a change when he starts with a victory lap.
@@Kinglink He didn't say it's the best video ever. He says it's the most important programming video you are ever going to watch, and he's right.
The fact of the matter is that the majority is not always right. Sometimes all of society can collectively make a mistake, and that's exactly what Java, and OOP was.
I'll just grab some popcorn before reading the comments
Better grab a bucket.
Eatin’ some white cheddar flavored popcorn right now ... LOL
OOP is wonderful for large codebases. I have worked far too long on the IBM iSeries platform with its go-tos, and top-down programming. It doesn't work. It's nearly impossible to upgrade codebases that are gigantic and written top down quickly, and they're insanely error prone.
Local versus global variables and isolating data in OPP makes life SO MUCH BETTER. I feel like you haven't ever worked on a gigantic codebase.
This is true, and i also feel like this video has nothing to do with areas of coding that arent database/website associated. Like simulations, or even game dev, where you dont need all data to be stored or returned somewhere. In that context youre not just "performing an action on data", youre using objects to capture behaviours, and having them split into classes with encapsulation allows that process to make sense, without ending up with a script thats 20000 lines of code.
Entity-Component DOTS style coding is often a better choice, but its readability is atrocious compared to OOP, and thats important for quick testing and turn around times.
The idea that a style as useful, flexible and readable as OOP is just blanket "bad", is really more just an admission that you dont know when to use it, or how to use it.
I really don't see how huge functions with only comments to break up logic are any more readable and, more importantly, easily testable. I would much rather modular functions that can be separately tested and read quickly rather than just a huge column that I have to memorize the components of. Also why wouldn't you just put the same comments you suggest in your longer function before the function calls?
I watched the video with an open mind... but then he said we don't need functions, we need inline functions. No matter how "good" the code you write is, it will break at some point. I fear debugging an unnecessary inline functions as much as a dependency hierarchy that is too deep. Debug a 2000 line function and you'll see.
I disagree with the idea of long functions, we need a way to test our code by unit instead of a bunch of logic at the same time
It’s more readable because the sections are lexically contiguous, instead of being scattered around the source file. Wanting modular, separately testable functions is really just wanting the unit tests to automatically localize any bugs that occur. If the large function fails its test, you still know something’s wrong with it, you just have to manually examine more code to find exactly what. Writing code so that its tests can localize bugs is optimizing for the wrong thing.
@@richardblain4783 The separation into smaller functions does not imply testing all the little functions separately. A module can be broken in 15-20 functions instead of a huge pile of 2000 spaghetti code, and still you would test the same behavior.
I'm working on a pile of functions that are thousands of lines long, with no tests. I can not refactor because well, there are no tests for the big function either. The auto refactor failed because it is written in the 'C' style with 'goto cleanup;' instructions. If the programmers had split the code into smaller pieces it would be easier to reuse the logic AND to write unit tests for them step by step instead of testing a 1000 logic (which is way beyond my timeline).
While I agree with most of the things said in this video I'm totally against having long 1000 lines functions
Also, I think longer functions makes the code harder to read because it implies that all the details are in one place. You start seeing things at a high level of abstraction. Ex "makePayment" entangled with ultra annoying details like a for loop over char*. You start reading code that is like 4 levels of abstraction in just 10 lines. which makes the cognitive load way higher than just see nice separated subroutines.