I love how CodeAesethic just came from nowhere, proceeded to release some bangers and now is easily one of my top 3 fav programming youtube channels lol
Nemean was even crazier, just dropped the Quake q_rsqrt video and got like a million views and tens of thousands of subscribers with only one video on his channel
@@animowany111 Not programming, but Bobby Fingers' view to upload ratio is insane. Guy comes out of nowhere and drops a 20 minute video that is currently sitting at half a million views 10 months later.
I dislike quite a few of his videos. They feel like they were made by person that writes 100 lines of code and names it a project, and therefor it's not relatable to actual production code.
@@pixelpox11 I agree with you. Even the actual name of channel "code aesthetics" already sounds very weird, since code is not a book or poem, it's a machine instructions :)
Inheritance only works on paper because it forces you to go backwards. Programming tends to happen bottom-up, you start with some core functionalities, and write more and more stuff on top. OOP requires you to work top-down, which necessitates knowledge of the full problem upfront. That’s why it works on paper (small exercises) but fails on the true stuff. I’m not sure if you can build inheritance to work the other way round, but it’s clear Rusts traits are much closer to this, that’s why they feel nicer.
Underrated comment. I would expect if one were to poll pro-OOP vs anti-OOP programmers there would be a strong correlation between top down vs bottom up design mentality.
Depends on what your true stuff is. In embedded stuff, especially safety critical stuff you get a much tighter contract than any "big tech" company has ever written. Not uncommen to get a close to full coverage unit test alongside a project. But most of the time we use inheritance to fit some code into memory constraints. Something most desktop or server programmer don't need to bother with. At least not to the extend of, if we really need the bigger chip we lose millions of dollars.
I like Jonathan Blow's take on design paradigms: " The benefits are obvious because the thing got proposed because of its benefits and you'll hear about the benefits. The cost tend to be subtle and/or swept under the rug by the people that want to sell you on the benefits." - 'Jonathan Blow on unit testing and TDD' (7:25)
@@JohnZakaria I really don't think that calling codeAesthetic the 3blue1browne of CS is very accurate, programming is significantly more accurate. Like the topics he covers is: - how to make code easier to read - how to make code more extendible - how to design code effectively (yes this is a mix of the above). What he isn't covering is: - how the computer works fundamentally - complexity analysis - math What he's covering is significantly more accurate to call it programming than CS. And this isn't a bad thing btw. I find it funny when people think monkeybrain "CS good, programming bad". But saying that, they aren't the same topic, computer is understanding why we design code, how we can analyse code, and even lower how a computer works. It's even to find the limitations of code. Programming is a part of CS, and this is what CodeAesthetic is explaining (really well).
I definitely agree with that whole "when you're shown this it makes perfect sense". Like the idea of inheritance is smart, and practical, and obviously a decent way to do it. Then you see a codebase that actually uses it and it's such a complete cluster fuck and it's so hard to keep track of what anything is supposed to be.
I think the issue is that inheritance is often taught with examples that describe existing properties or functionality. People will often use the example of having a square class that inherits from a polygon class. The issue is that this works because you’re classifying existing functionality. When creating something new, the analogy falls apart. Companies don’t have a Vehicle blueprint that they then extend to either get a Car blueprint or a Motorcycle blueprint to the build. They build the Car and Motorcycle completely separately.
Yeah I think people take it too far, I worked for a company which only employed "Senior Developers" and had 2 juniors to do really menial stuff. Some of the software those "Senior Devs" wrote had so much inheritance if you wanted to go down the rabbit hole it would take you hours to figure it out. Thats really bad code architecture and I don't think you can call yourself a "Senior" anything if you don't think about how your code reads to a person who's not familiar with the structure.. At the same time though, these devs would be very critical of how you named your classes and methods, so you don't care about how your code reads but you really care about what a class is called? That doesnt make any sense to me.
doesn't it just depend on how it's implemented? For example, in my game, every item inherits from the base world object class. This class handles temperature and other things like that, so anything that affects temperature can easily do that. Then an example of an item is a tool. The tool has base things like, how it can hold items, how you can remove items from it, and whatnot. And then there's specific tools that have their own specific behaviors for items in the tool. I just want to know, how in god's name would you approach any of that without inheritance? I genuinely want to know. And I think if done well it's just common sense what things are supposed to be
@@moonashaDoes your game have weapons? If so, would a weapon be a subclass of item? What would happen if I wanted a hammer to act as both a tool and a weapon, or a baguette as a weapon and food etc. This flexibility might not be necessary for whatever you're building but it's something to consider
Yep, and honestly there’s some codebases that spoil us rotten and make it seem so good because the focus is so constrained that object orientation is good. Perfect example is game engines and really popular large web frameworks. Even then, there’s limits and people end up running into issues. Want to see a HORRIBLE OOP library? Read LangChain, and try to figure out how Conversational Agents pass through chat_history, this nearly drove me mad. If you’re wondering why it’s that way, they release a new version of the library roughly every 1-2 days (absolute chaos of a commit history) and they prioritize maximum flexibility above LITERALLY ALL ELSE.
I love how Prime calls out CodeAesthetics for the next line curly braces and then rhetorically asks whether it's C# because C# standards infamously use next line curly braces. Then I double check the code and realize that it is actually C# code.
I know it's an old video but I'm so glad I found this channel. TH-cam addiction is my greatest struggle as a junior dev that just has been coasting in his work. I know this channel is primarily entertainment but replacing some of the political trash that I normally binge with your content is starting to improve my mindset and problem solving skills and I'm noticing improvement and growing competence at work as a full stack dev. Coming from a game dev degree that I'm sometimes not proud of you're really helping me fill gaps in my knowledge base that I missed in my formal education. That said I'm happy I went to college for game dev, got me into coding and even if I was disillusioned by the industry during salary negotiation it got me started down this path and it's something I'll never take back.
Inheritance has its place. It can definitely get out of hand. Composition has its issues, but he made the point that some of those issues might be due to putting too many responsibilities on the composes objects. Some of the issues with inheritance can surely be traced to the same.
Never heard of the triple method but thinking back, its seems to be categorically true in my experience. I've seen lots of new products fail after months of effort because we didn't know all the things we learned in the first attempt but the 2nd pass usually ends up shipping and doing well. Whenever I've worked on proprietary framework level things it does seem like it takes at least two but often three passes for it to be fully baked and desirable to the org.
The file savers should really just be functions that take some sort of writable object, an image, and options. There really doesn't need to be a class, since that just creates an unnecessary alias to the image and options.
"Avoid protected member variables with direct access" - when I was learning Android Development in Java, this KILLED my motivation entirely when trying to reason with the code, some view operations intensively use protected members from the parent class and I kept asking myself where those members were coming from...
Just add correct code convention. m_ - protected member variables. p_ - private variables. Well in any case you can just look into a codebase and find what this member variable is actually doing. I didn't really understand your concern.
I love my next line squirly braces... especially in a language where things can occur between functions and their scope, such as declaring a function const, or member initializers, or metaprogramming(?) such as std::enable_if, or any other number of things. Lines come free, readability doesn't. MyClass::MyClass(int value_1, int value_2) : member1(value_1), member2(value_2) { //stuff }; ^^ isn't too bad until you have longer type names, or more parameters, or more going on with the class/function/etc. MyClass::MyClass(int value_1, int value_2) : member1(value_1), member2(value_2) { //stuff }; ^^ is so much easier to read and is much easier to be consistent with in different situations.
I quite like these 2nd channel videos. The volume of clips and more of a focus on Actual Programming concepts, rather than the 'hot-takes' to just drive engagement like the main channel. I like seeing the M who just likes to help people get better.
the rust default impls of traits is available in C# as well (also C# allows Data Members in interface but not entirely since Properties are methods and not fields so it feels like it does when it actually doesn't)
The biggest problem with the term NoSQL is that it's like "We run NotLinux on our servers"... well, what DO you run? Windows NT? FreeBSD? QNX? iOS?! FreeDOS?! "Everything except ..." is a worthless way to group things.
I did java for several years and i suffered at the hands of forced inheritance. Now i use literally anything other than java and i hate inheritance because it overcomplicates everything and forces you to fit the problem to the code rather than the other way around
i use inheritance only for derivatives, the child classes only add new functionality, they may or may not alter some functionality (i usually prefer not to), and maybe implement some virtual methods to fit an API (or use interfaces) aside from that, i avoid it like the plague, if it's not needed, dont use it
C++, JavaScript, Python, C#, TypeScript and fucking COBOL has inheritance, and are fine, but not Java? Do people just not know how to write Java or something?
@@HrHaakon I almost didn't reply because your comment severely missed the point, but I have a minute so I'll explain again: Java forces you to use classes and heavily encourages you to think about the problem in an object oriented way. (You don't have to of course if your classes are just function containers or bags of data but that is terrible Java. and as someone who knows how to write Java I'm sure you wouldn't approve.) This is not good because most problems are not cleanly represented in the pure OO inheritance world of Java. So while other languages may have inheritance, they also have more sensible ways of modeling code so that you never have to touch inheritance. Caveat with TypeScript: you should know about prototypical inheritance but you rarely have to actually use it. You can write a NodeJS application in TS that doesn't touch prototypes or explicit classes ever.
@@HrHaakon I didn't sneak it in. I made it pretty clear that you don't have to use inheritance in Java but if you don't then you're writing what Java people would call "bad Java." Maybe next time read the comment.
2:40 that is because they don't tell/show you how painful it is to change the structure of it after it has been coded . They show it from the perspective of a world where everything is easy, and is correct on the first try. (I have not gone to college for programming, so I don't know if that is the case, but it sounds like that is why)
I have studied OOP in college a long time ago and this is exacyly how it is shown, OOP has its place but we have gone too far in making everything abstract and that in turn made things worse.
Schools teach you how, not why. So when they introduce a new hammer, people want to try and hit everything that looks even remotely like a nail with it.
I really connect with Prime's idea of 3 stages of writing an application. Coming from embedded development, I am very used to separating data and methods on data, and even now that I work on higher plane of abstraction I often always write my first version as a combination of data containers and methods on them. What I find is that often times on subsequent re-factors that pattern alone is sufficient without even going towards OOP way of structuring your code. I am not generally opposed to it, but in my experience it's far easier to first do the simplest thing first (data + methods on data) and if you spot patterns - refactor them however it makes sense to refactor them. Using classes to structure your code from the beginning IMO is hard to get right even for experienced devs, inheritance or not, you are still effectively dealing with lots of local namespaces and are dealing with "this object owns this and that object owns that", if you then need to share some data, it's more work to make these changes compared to if they were simply passed in as a parameter to some method I also feel that having a very clear way of identifying data within your program is often very beneficial and illuminating. I suppose this is more towards data oriented programming approach
My experience is that if you're having a clump of data, and then some operations on that data, it's pretty easy to get it done using OOP. Will you have to refactor a lot? Maybe. But it's pretty easy to think in terms of. Then again, I model the database FIRST these days, as nailing that is pretty reasonable, and then the OOP layer on top handles messaging between the data model and the front end. So it's like the controller in the way MVC was originally conceived. It works really well for enterprisey solutions.
That description of developing software in a 3ply-fashion is brilliant. Totally tracks sometimes you build it to understand what it does, then build it to make it ready for others, then build it to make for you and others in the future.
dependency injection was our bread and butter in high school robotics. We split the robot into subsystems and everything would be injected into relavent methods and classes. Was a massive pain
The solution to the image problem the video presents could be far more easily solved in practice with a single Image struct and some associated methods for editing + separate load/write functions to different image formats (maybe a ImgConverter interface if needed). The video is good but usually these OOP examples are overly complicated for no reason other than trying to fit a mental model of classes
That's the point, though. You have some class that works just fine, but then one day you want to extend it. Doesn't matter how cleverly your inheritance structure solves the current requirements. The moment a new requirement is added, the resulting refactor is huge. Sure, you can solve the problem given here, with a drawable / non-save-and-load image just fine if you know that that's a requirement from the get-go. But then something new comes along.
Actually the method I proposed is way more flexible, as the image manipulation logic is separate from the conversion logic. Want more formats? Simply write a new save/load function pair and pass an Image struct to it.
@@MasterHigure The problem with inheritance when I used it (maybe bad at OOP) is mutation/global state is order dependent, and if you inherit something the child classes may sometimes want to call the super method first before the body sometimes last and it could lead to subtle bugs because of state, you sometimes have to have the super method be a higher order function so it can be called around your code. Also sometimes you have to have the parent call the parent version of a method other times you want the parent to call the child method...
Data duplication is good when it leads to a more decoupled codebase. You don't always want a giant users table with 100 columns because every part of your system needs parts of the table. Adding an extra column suddenly breaks everything because other parts of the project are now trying to insert null into the new column.
I believe pretty strongly inheritance is only good for composing orthogonal behaviors. Inheritance for specialization opens the door to overriding, calls to super and having to understand multiple versions of the same artifacts but jumping back and forth between modules to understand what each does. My brain doesn't compute that.
Code Aesthetics messed up with the inheritance diagram right before moving on to composition. The object on the right started with only blue and red square but ended up as a child of the green and blue object.
So true. When learning programming, inheritance seems so logical and endlessly useful. Yet much of the more advanced programming literature/videos points out that composition is often better.
I agree with you on smooth scrolling. There's a reason why scrolling was originally in steps and it's because it causes less motion sickness. I am not particularly prone to motion sickness, but I find it much harder to read stuff when they're smooth scrolling, even on a 144hz monitor. The greatest demonstration of this are youtube live comments vs twitch, where the former is smooth scroll and the latter is bulked into steps. Twitch chat is legible when it's being -SPAMMED-, I'm talking 100 messages a second spammed. TH-cam chat is barely legible with 5 messages per second, because it's constnatly scrolling up, there's no rest for your eye to actually read the words.
An example of where inheritance makes sense is Pokémon: you have the definition of a pokemon(name, evolution, moves it can learn, abilities it can have, base stats). Then you have a class to express a pokemon in your team (has ivs, evs, only four moves out of whatever it can learn, a personality, only one ability) and from the pokemon class it helps to validate if it can learn the move, calculate actual stats based ob ivs and evs and so on. But then I wouldn't do a battle class that inherits from pokemon, items and so on
I got to work on an old project where all the important bits happened in business objects that were pure spaghetti of inheritance. EVERYTHING happened through protected variables and methods. You would see a 5-line function and think that that should be easy enough to understand... but then it ends up triggering 500 lines of code across 5 levels of inheritance. And it would go up and down, up and down... Line 1 out 5 executes something in its parent, but that is passed right through to the grand parent. So the grand parent does some things, then passes it to the GRAND-GRAND-parent, which executes like 10 lines before it goes back to the grand-parent. The grand parent does a few more lines before calling something that passes through to the GRAND-GRAND-GRAND parent... which calls some completely other module... and from there it falls back to the original child object I was working on. One line done, get ready for 4 more...
as a silver haired coder from the 1980s OOP & inheritance was the bees knees. on paper. awesome. in reality - i watched programmers inherit the crap out of stuff, till it was just crap. So deep that even a Matryoshka doll would have gasped. also kudos for your music tastes there with miss monique.
Now I understand why Rust traits work in such a way. It has the advantages suggested in this video, without the disadvantages that comes from inheritance.
@@betterinbooks I agree. Once I was coding a server backend. I created the struct for the internal state that is going to be accessed by the endpoints. I then created a trait with the methods for accessing the state. For each endpoint, I created a unique trait for it with the behavior of that endpoint. The trait applies to any struct that implement the original trait type. Thanks to that, each endpoint is its own trait. We just have to implement these in the state type. Instead of one file with all the business logic, I was able to divide each into its own trait. Extremely easy to scale. Loved it so much. Using that structure, we could have several people working and pushing to the same repository at the same time. Each developer needs to simply work on their trait.
This video so purely speaks to an idea I've been thinking about for a long time. I keep seeing these videos floating around about why OOP is a lie, and why it doesn't do what it was made to do. Which was solve the lack of modularity in imperative programming. To be fair, it doesn't solve that problem, we have modules now. However, this is exactly why OOP has evolved over time. It's not about trying to make infinitely reusable code snakes, it's about breaking down large frameworks into easily definable objects that can be instanced to serve a higher purpose.
I clang-format on save with whatever defaults it considers the best. So because IDGAF enough to create a .clang-format file it does my braces on new lines. At first it would cause me geniune aggrievement but honestly you learn to accept it and move on. I work with C++, my life is shitty enough.
This is the first reaction video I liked - ever. You complement each other so well. It is like you inherit his evil genius and implement an emotional amplifier of profanity into it.
How about people think long and hard about what their classes look like before writing them. Why the hell would an Image have any concept of save and load when the sole purpose of an image should be to represent an image as data? Having anything extra is over abstracting, class should be the bare minimum and nothing more. I'd keep image the way it was (without save and load, and maybe even the transformations) and have the abstract class for saving and loading. Interfaces to me are for functionality which isn't restricted to any concept. For example if you wanted to have a collection of things which are all unrelated but all have save and load then I'd use an interface, that way I can just create a collection of what ever interface, ex: ISerializable, IEnumerable. If I want to save and load strictly for images, I'd have an abstract class inherit from Image and put abstract save and load methods in there. Now I can have a collection of images which can contain all image types as well as a collection of all "FileImage" types if I wanted only images that have save and load. This also doesn't break implicit conversions from Image to any FileImage. And for those saying null-conditional operator is hiding a problem, I'd argue that it exposes a problem that was previously hiding. And null in itself is not an issue, its like saying instead 0 as a guid is an issue, its just a state. It is your responsibility to handle what state its in. Asking to get rid of null entirely is like asking to get rid of 0 entirely.
9:15, there should be quotes around the word perfect in "change is the enemy of perfect design". A truly perfect design enables and embraces change and provides seamless extendability without needing breaking changes.
I great thing I heard to use was "Is" vs "Has". If something "Is" something else, then you use inheritance. If something "Has" something else, you use composition. For example, "Is" a Png an image? No. Png "Is" a format for saving images. As are the other formats. So we can separate these from image and have them inherit from some ImageFormat class which contains methods to take in an image and save it and return a saved image. "Is" a DrawableImage an Image? No. But a DrawableImage "Has" an image that it can draw to. So we can separate DrawableImage to its own class, and give it an image to perform drawing operations on.
I like your point at 10:00. First build is discovering what you don't know. Second build is learning the things you didn't know. Third build is starting from the beginning with full knowledge.
24:06 This describes my hatred of DI *frameworks*. I love dependency injection and I use it a lot in Go, but the dependencies are created manually and passed around in code in such a way that's easy to trace and debug. I hated DI in other languages (C#, TS [Angular], PHP) because it almost always involves some sort of DI Framework that over complicates everything and turns DI into a magic global black box that's a pain in the ass to debug and manage configurations.
The video doesn't present a problem with inheritance. If anything, all it does is reveal the necessary foresight and experience required to construct a feature-rich, well-designed API. It just so happens that the lack of such things are naturally inherent in humans, and this by default. Also, inheriting from an interface *is* inheritance.
6:25, just use the virtual keyword instead of abstract. This will allow you to define and write code in the function inside the abstract class. This allows you the ability to override it in inherited classes while not being forced to.
My biggest problem with composition in Rust, especially those default implementations for functions of a trait... is that when these default implementations rely on state, you always have to provide accessor functions within the trait... and in rust you often need two for the same value, one for immutable and one for mutable state... this forces the struct that implements said trait, to implement these accessors even if they just point to a single field in the struct... depending on how many such traits you have this can cause a lot of boilerplate code
it's C#, the opening brace on a separate line is the default and the norm there. the methods that start with a small letter, though... A SIN! AN UNFORGIVABLE SIN! the only things that start with small letters in C# are local variables, parameters and private fields! (private fields also typically start with _ but then a small letter follows) (I'm talking identifiers, of course keywords (int is also a keyword) start with small letters)
i put squiggly braces on the next line in protest; if a language is forcing me to define blocks with braces, then I am going to make them as prominent and annoying as possible
Those curly braces on the same line are standard practice in C#. Any linting tool would complain if you did anything different to that. They are called automatic properties and they are awesome.
The one thing I've never understood is why full on complex class objects is considered better code reuse than separate functions for common tasks. Like classes make sense if the class covers every single use you have, but that's rarely the case in any moderately complex system. I guess this is what they mean by composition over inheritance, but inheritance seems more like a convenience and organization advantage in certain cases, not a code reuse advantage.
Java Swing goes full in on inheritance, and it's the only one case that I found actually useful and easy to hook into. Building UI, quickly changing a draw method here, or making a table behave like a tree. Once I got it, it was really easy. But designing that stuff myself to be flexible like that, no thanks.
I also made my first game with xna, a tower defense, and I also had like 4 levels of inheritance for only the attack effects lmao Oop obsesión was so bad when I was young
How to execute: 1. Dicks out for Harambe 2. Shit code 3. Full rewrite no matter what "statistics" tells you Took me over 10 years to understand how true this is
25:03 Yeah, I've thought about this a bit and have been curious if it's not a bad idea to have an `embodies`, `compose`, or a similar keyword. The idea would be that within your class you would say that it embodies another object, which you can then access it's methods as if it's your own, but they would be private. Mostly a convenience feature, it would benefit from the IDE informing you which component you are fetching the method from, but I haven't thought of a scenario where it would be dangerous (yet.)
Is it though? In a dynamically typed language, if I implement a common set of operations on some classes, then I have implicitly implemented an interface. That Java requires implementors to be explicit doesn't suddenly make it inheritance. There's also the question of what is being inherited: besides default methods, there isn't any behavior being shared by an interface to its implementors. I don't think there's much inheritance at all.
Imo the only good way to have this conversation is by looking at what the compiler does. For example each time the compiler sees a dereference (dot operand) it performs an extra instruction for null checking, while local variables are accessed directly. And if we get even more exquisite... Local assignment(dereferencing a field to the method stack is supposedly even faster... But the assignment tradeoff needs to be matched against the number of times the reference will be accessed.
The performance hit is probably insanely negligible since many instructions are pipelined and executed partially in parallel. By this logic, all OOP languages would be slow, but they arent Inheritance is not bad. The code reuse alone is much better than any type of convoluted dispatching argument. The best, most bug free code is code that is never written or re-written. Without inheritance, you are guaranteed to write code over and over for no reason. People just really dont understand inheritance in my opinion. And if you are so concerned about performance, write everything in assembly.
@@jshowao I like OOP, but I won't deny it's drawbacks, specially in such limited scope tests like a simple loop. JIT is better at optimizing code than a well written C, maybe even a very good C. But the time it takes for the optimization to kick in... Takes some time. Another thing about VM languages is that their compilers are really not that smart... Maybe because of exceptions and extensibility but devirtualization should happen a lot more... yet is something that rarely happens. Each virtual call adds a whole instruction... which eventually adds up... Each extra instruction call is ONE LESS op code of instruction sequence, which is one less op code for the speculation sequence. The worst offender is that each virtual function call forces a memory order boundary, which means that the compiler is not allowed to interleave beyond the virtual function call... this is the worst offender.
@@jshowao I love OOP, I just wish compilers were more confident when devirtualizing and parallelizing up front, maybe even add keywords to hint devirtualization. Also I'd love for compilers to not implement 'freezing' so strictly, instead make a flow graph and ignore the constructor memory barrier if the logic allows it... do it up front instead of JIT. This would allow swaths of lengthy op code sequences... Making massive and faster speculations uninterrupted. I keep being told that if full de-virtualization would be applied, some pattern would break when dynamic dispatch is not obeyed, yet I keep getting some weird/nonsensical and incomprehensible examples of this...
in React, if you making heavily nested reusable "functional" components - they're effectively classes with multi layers of inheritance and you run into same problem of everything being too interwoven
~8:00 OOP inheritance can easily go out of hand when doing games, classic orc merchant npc issue. It's not nearly as bad when you have hard defined scope, although thats probably the crux that you tend to expand that scope over time and now suddenly your child classes bring along garbage thats only relevant to some children and you need to override stuff
This feels like almost an exact replica of a Raymond hettinger talk, but there was obviously a lot of care and love put into visualization and video development.
It just depends how many "branches" of your core class there end up being. If say there are 30 variations.... then you'll probably have a hierarchy of inheritance with 3 or 4 "generations" to stop objects from inheriting too much. That can require a lot of pre-planning and can easily end up a mess. I'd usually prefer to use the modular approach of composition to form each sub type, but it really just depends on how many variations we plan on using and how worried we are about memory.
These design patterns help a lot when you need change the code with minimum impact.. sometimes can be over engineering, but most pf cases not!! Prime continue with these videos because your lives most of time is in my work time and sometimes I can't watch and the Vod is to long!!!
The problem with "dependency injection" is that ppl get carried away with it, and you end up in a situation where inheritance is actually better. The best way is to get rid of classes all together - in c# you're rooted though cos everything is a class.
That guy in the comments "grepvyne" made a great point: "This is my fear writing the piece of shit version, that my manager will just say yep off to prod it goes and it'll never get re-written properly" My tech lead told me this is exactly what happened years ago to the software we're currently maintaining and why the code base is a mess Also, from hence day fourth, I will now pronounce NoSQL as "no squeal". Thank you
my current strategy is, "Can I do it with an interface?" If the answer is yes, I don't use inheritance. There are some cases though where inheritance is super useful, but I live mostly in the game coding world so it probably doesn't apply elsewhere. As an example, a game would have a base object, that keeps track of temperature, who owns it, etc. Then more specific objects with certain functionalities can inherit from it, without having to duplicate a bunch of code. Basically for me it comes down to how much code you're duplicating
You create classes for each of your properties, like temperature, and add them to the class of the object you are creating. Then, you make a factory with methods for instantiating your object with valid combinations of your property classes for creating different types of objects.
companies should be interviewing people by having them configure the linter for the language they’ll have to write. nevermind if they can cook up a linked list problem in 40min, find out of they want all code to be wrapped to exactly 80 columns and convert tabs to 5 paces
Did XNA in college senior year, I loved it. Glad to hear I wasn't the only person who used it. I made a pacman and tetris clone in it and was so excited to deploy it to the first generation of windows 7 phone. RIP windows 7 phone, never got flash support and was too late to the market
What do you think of query builders? I honestly can't stand query builders for sql. I'd rather look at a basic string with some replacement keywords or embedded variables.
@@ThePrimeTimeagen just... Why should I be learning a new structure/language with your query builder? If you just keep it as sql I can read it without having to figure out your custom representation of a query.
String replacement can take a long walk of a short pier, but SQL is better than 90% of everything else the SECOND you're no longer doing basic CRUD that's 1-1 with a single table with no joins.
When implementing a compiler inheritance is elegantly straight forward. Functionality is presumed from the way objects are structured in memory. Where as composition there has to be an active mechanism in determining if such particular object contains a particular functionality. Maybe inheritance is a house of cards that may apart due to design issues. But when it works it should be more efficient then composition
It's indeed the best way to produce quality software. Developers who think they can do it right the first time just haven't learned that lesson yet. RIP Harambe
What happened there is that the parent class Image was actually transformed into a mutable isolated class for handling the operations; a better name from this image would have been ImageBuffer (which is what it actually represents). The error at the beginning was that thinking having PngImage and JpegImage import from ImageBuffer means it could be a good idea to place save and load there. There was a problem of cohesion: save and rotate shouldn't have been placed on the same class. If you want to solve this issue using inheritance, one possible solution is to declare save and load in an interface, not in the abstract Image class. I'd rather prefer what haskell does with GADTs and type classes, so that the cohesion problem disappears, you just add instance methods in the correct module without having them to be in the same file as in OOP.
The semantic meaning for inheritance has always been taught as “X is a Y” is X inherited Y. The reality is that the true semantics is “X is only a Y”. Reasoning being X is/can be many things, however inheritance only allows for a single “is” and therefore “is only”
So...I don't see why the proposed solution isn't functionally equivalent to multiple inheritance. The entire idea of interfaces being "lightweight" but (in some languages) allowing you to give default implementations that reference other methods of the interface... isn't that just a smaller parent class? I don't see how inheritance precludes any of the solutions to the given problems. "ImageFile" could also be an abstract parent class, as could "Image." Then your JPEG, PNG, and Bitmap all inherit from both parents, while drawable only inherits from "Image." Am I missing something?
5 หลายเดือนก่อน
I do inheritance, but it's almost always just one level, very rarely two. I often need very slight variations on different objects, and majority of the functionality comes from the parent. If some of those variations are needed in multiple classes (but not in *every* of them), then they belong into interfaces and become composition.
A perspective that helps me understand why Dependency Injection has similar problems to inheritance: There is an intrinsic tension between code-bases that are flexible at runtime, and a code-base that is flexible to change by developers. Overgeneralizing slightly, the more runtime flexibility, the harder it is to reason about the code and thus the harder it is to change the code when your needs change. Both DI and inheritance are flavors of polymorphism, and polymorphism makes it hard to reason about code when some change needs to pierce the polymorphic abstraction. It can be fiendishly difficult to get a grasp of which code uses other code when it's all hidden behind polymorphism. So, I'm not a fan of inheritance nor DI. They have their place. But they're usually _too_ powerful. When you really have multiple implementations and need to mix em up for whatever reason, then encapsulating polymorphism is useful. But usually, it's unnecessary, and then it's simply a burden to comprehensibility. Being able to trace which code is used by what _exactly_ is very valuable; so giving that traceability up (by hiding behind polymorphism) needs to have a very concrete high-value upside to compensate for losing that ability.
I agree that using ?. without another condition is terrible. I think it has a ton of use though, as long as you follow it up with "?? someDefaultValue" or other handling. Hiding an error with it is terrible. Consolidating error handling with it, however, is beautiful.
10:05 A typical conversation at my job: I: Should we really change this thing to work totally different? It will be difficult to restore it. I'll remove all the stuff I don't need anymore. > Sure, just change it. We decided for the perfect design, it won't change anymore. A few months later: > I got an idea, for a new design, we should do it like this. I just think, that's almost exactly how we already had it.
Inheritance is more about modeling the domain and that subclasses ARE an instance of the superclass. Composition is about HAVING something. Those that argue they are solving the same problem are mistaken. Interfaces/traits are for a shared set of behavior with different implementation. They are all different and have their place. The problem is when a language forces a developer to only use one.
6:32 How about having a File class and an Image class. If a particular subclass of Image needs File I/O, then let that subclass extend the File class. If not, then the subclass of Image doesn’t have to extend the File class and hence doesn’t have to implement the File I/O methods
I've been viewing your back catalogue of videos and this is by far the best one so far. Started programming with COBOL and DB2 on IBM mainframes back in 1990. Now I just do Godot and Python for selling small games off Steam.
Another problem is that college tends to expain interfaces VERY POORLY and inheritance much more usably (at least as per what I saw) I remember I used abstract class somewhere where interface was fine just because I did not know interface as well
Exactly. The absolute best way to code is to fever dream a quick working example. Take a step back, go for a walk, think about what just happened. Then start again from scratch. I guarantee you won't need OOP if you really think about what code is necessary.
The ? in C# doesn't hide a problem, it just is a short hand for if(obj != null)... If you don't know what you are doing it doesn't matter in how many ways the language offers you to do it wrong in.
I love how CodeAesethic just came from nowhere, proceeded to release some bangers and now is easily one of my top 3 fav programming youtube channels lol
Nemean was even crazier, just dropped the Quake q_rsqrt video and got like a million views and tens of thousands of subscribers with only one video on his channel
@@animowany111 Not programming, but Bobby Fingers' view to upload ratio is insane. Guy comes out of nowhere and drops a 20 minute video that is currently sitting at half a million views 10 months later.
And mia for a while
I dislike quite a few of his videos. They feel like they were made by person that writes 100 lines of code and names it a project, and therefor it's not relatable to actual production code.
@@pixelpox11 I agree with you. Even the actual name of channel "code aesthetics" already sounds very weird, since code is not a book or poem, it's a machine instructions :)
Inheritance only works on paper because it forces you to go backwards.
Programming tends to happen bottom-up, you start with some core functionalities, and write more and more stuff on top. OOP requires you to work top-down, which necessitates knowledge of the full problem upfront. That’s why it works on paper (small exercises) but fails on the true stuff.
I’m not sure if you can build inheritance to work the other way round, but it’s clear Rusts traits are much closer to this, that’s why they feel nicer.
agreed
That's a really good way of describing it. Tbh, I prefer inheritance over composition, but it requires much more work to do it right.
I learned to program from a top down approach and I definitely find it very useful for working out problems. I barely ever use inheritance though
Underrated comment. I would expect if one were to poll pro-OOP vs anti-OOP programmers there would be a strong correlation between top down vs bottom up design mentality.
Depends on what your true stuff is. In embedded stuff, especially safety critical stuff you get a much tighter contract than any "big tech" company has ever written.
Not uncommen to get a close to full coverage unit test alongside a project.
But most of the time we use inheritance to fit some code into memory constraints. Something most desktop or server programmer don't need to bother with. At least not to the extend of, if we really need the bigger chip we lose millions of dollars.
I like Jonathan Blow's take on design paradigms: " The benefits are obvious because the thing got proposed because of its benefits and you'll hear about the benefits. The cost tend to be subtle and/or swept under the rug by the people that want to sell you on the benefits." - 'Jonathan Blow on unit testing and TDD' (7:25)
Amazing commentary
i'm a fan of that channel, the way he explains is quite like he's the 3blue1brown of programming.
And that should be emulated everywhere!
I would say the 3blue1brown of CS is reducible, he goes deep into CS topics
yeah code aesthetics is more like the 3blue1brown of medium articles
I'd say he has some of the same traits of 3blue1brown, don't want your comment to get outdated if 3blue1brown changes how he makes vids.
@@JohnZakaria I really don't think that calling codeAesthetic the 3blue1browne of CS is very accurate, programming is significantly more accurate. Like the topics he covers is:
- how to make code easier to read
- how to make code more extendible
- how to design code effectively (yes this is a mix of the above).
What he isn't covering is:
- how the computer works fundamentally
- complexity analysis
- math
What he's covering is significantly more accurate to call it programming than CS. And this isn't a bad thing btw. I find it funny when people think monkeybrain "CS good, programming bad". But saying that, they aren't the same topic, computer is understanding why we design code, how we can analyse code, and even lower how a computer works. It's even to find the limitations of code. Programming is a part of CS, and this is what CodeAesthetic is explaining (really well).
I definitely agree with that whole "when you're shown this it makes perfect sense". Like the idea of inheritance is smart, and practical, and obviously a decent way to do it.
Then you see a codebase that actually uses it and it's such a complete cluster fuck and it's so hard to keep track of what anything is supposed to be.
I think the issue is that inheritance is often taught with examples that describe existing properties or functionality. People will often use the example of having a square class that inherits from a polygon class.
The issue is that this works because you’re classifying existing functionality. When creating something new, the analogy falls apart.
Companies don’t have a Vehicle blueprint that they then extend to either get a Car blueprint or a Motorcycle blueprint to the build. They build the Car and Motorcycle completely separately.
Yeah I think people take it too far, I worked for a company which only employed "Senior Developers" and had 2 juniors to do really menial stuff. Some of the software those "Senior Devs" wrote had so much inheritance if you wanted to go down the rabbit hole it would take you hours to figure it out.
Thats really bad code architecture and I don't think you can call yourself a "Senior" anything if you don't think about how your code reads to a person who's not familiar with the structure..
At the same time though, these devs would be very critical of how you named your classes and methods, so you don't care about how your code reads but you really care about what a class is called? That doesnt make any sense to me.
doesn't it just depend on how it's implemented? For example, in my game, every item inherits from the base world object class. This class handles temperature and other things like that, so anything that affects temperature can easily do that. Then an example of an item is a tool. The tool has base things like, how it can hold items, how you can remove items from it, and whatnot. And then there's specific tools that have their own specific behaviors for items in the tool. I just want to know, how in god's name would you approach any of that without inheritance? I genuinely want to know. And I think if done well it's just common sense what things are supposed to be
@@moonashaDoes your game have weapons? If so, would a weapon be a subclass of item? What would happen if I wanted a hammer to act as both a tool and a weapon, or a baguette as a weapon and food etc. This flexibility might not be necessary for whatever you're building but it's something to consider
Yep, and honestly there’s some codebases that spoil us rotten and make it seem so good because the focus is so constrained that object orientation is good. Perfect example is game engines and really popular large web frameworks. Even then, there’s limits and people end up running into issues.
Want to see a HORRIBLE OOP library? Read LangChain, and try to figure out how Conversational Agents pass through chat_history, this nearly drove me mad. If you’re wondering why it’s that way, they release a new version of the library roughly every 1-2 days (absolute chaos of a commit history) and they prioritize maximum flexibility above LITERALLY ALL ELSE.
I love how Prime calls out CodeAesthetics for the next line curly braces and then rhetorically asks whether it's C# because C# standards infamously use next line curly braces. Then I double check the code and realize that it is actually C# code.
I know it's an old video but I'm so glad I found this channel. TH-cam addiction is my greatest struggle as a junior dev that just has been coasting in his work. I know this channel is primarily entertainment but replacing some of the political trash that I normally binge with your content is starting to improve my mindset and problem solving skills and I'm noticing improvement and growing competence at work as a full stack dev.
Coming from a game dev degree that I'm sometimes not proud of you're really helping me fill gaps in my knowledge base that I missed in my formal education. That said I'm happy I went to college for game dev, got me into coding and even if I was disillusioned by the industry during salary negotiation it got me started down this path and it's something I'll never take back.
I agree that it's better to entertain ourselves with this kinda videos instead of politics or sports lol
LMAO the curly brace thing at the beginning, finally someone else who HATES newline braces
it hurts me
and yet so many different style guides say not to >:(
Thats actually a best practice for # and Java
C#
@@NeoChromer not for c#
Inheritance has its place. It can definitely get out of hand. Composition has its issues, but he made the point that some of those issues might be due to putting too many responsibilities on the composes objects. Some of the issues with inheritance can surely be traced to the same.
Spot on!
Never heard of the triple method but thinking back, its seems to be categorically true in my experience.
I've seen lots of new products fail after months of effort because we didn't know all the things we learned in the first attempt but the 2nd pass usually ends up shipping and doing well. Whenever I've worked on proprietary framework level things it does seem like it takes at least two but often three passes for it to be fully baked and desirable to the org.
Holy moley I think it's been at least 10 years since I've heard anyone bring up Microsoft XNA.
yeah... i did it!
The file savers should really just be functions that take some sort of writable object, an image, and options. There really doesn't need to be a class, since that just creates an unnecessary alias to the image and options.
the OOP mind cannot comprehend this
Nooo you need to allocate a FileSaver object to the heap so the garbage collector doesn't get bored.
@@thesenamesaretaken ugh no you need a file saver factory to handle the different types of file you’ll encounter
"Avoid protected member variables with direct access" - when I was learning Android Development in Java, this KILLED my motivation entirely when trying to reason with the code, some view operations intensively use protected members from the parent class and I kept asking myself where those members were coming from...
Why not use the Python naming conventions to add _ to protected and __ to private (name mangled).
Just add correct code convention.
m_ - protected member variables.
p_ - private variables.
Well in any case you can just look into a codebase and find what this member variable is actually doing. I didn't really understand your concern.
Who needs inheritance when you have enums and traits?
these are facts
you just hold my heart in a lock, i swear
That makes no sense.
I love my next line squirly braces... especially in a language where things can occur between functions and their scope, such as declaring a function const, or member initializers, or metaprogramming(?) such as std::enable_if, or any other number of things.
Lines come free, readability doesn't.
MyClass::MyClass(int value_1, int value_2)
: member1(value_1), member2(value_2) {
//stuff
};
^^ isn't too bad until you have longer type names, or more parameters, or more going on with the class/function/etc.
MyClass::MyClass(int value_1,
int value_2)
:
member1(value_1),
member2(value_2)
{
//stuff
};
^^ is so much easier to read and is much easier to be consistent with in different situations.
I quite like these 2nd channel videos. The volume of clips and more of a focus on Actual Programming concepts, rather than the 'hot-takes' to just drive engagement like the main channel. I like seeing the M who just likes to help people get better.
the rust default impls of traits is available in C# as well (also C# allows Data Members in interface but not entirely since Properties are methods and not fields so it feels like it does when it actually doesn't)
Agreed.
Its also existed in c++ since forever.
Not sure why he was gushing so much about it in rust, its not a new concept
The biggest problem with the term NoSQL is that it's like "We run NotLinux on our servers"... well, what DO you run? Windows NT? FreeBSD? QNX? iOS?! FreeDOS?! "Everything except ..." is a worthless way to group things.
I did java for several years and i suffered at the hands of forced inheritance. Now i use literally anything other than java and i hate inheritance because it overcomplicates everything and forces you to fit the problem to the code rather than the other way around
i use inheritance only for derivatives, the child classes only add new functionality, they may or may not alter some functionality (i usually prefer not to), and maybe implement some virtual methods to fit an API (or use interfaces)
aside from that, i avoid it like the plague, if it's not needed, dont use it
C++, JavaScript, Python, C#, TypeScript and fucking COBOL has inheritance, and are fine, but not Java?
Do people just not know how to write Java or something?
@@HrHaakon I almost didn't reply because your comment severely missed the point, but I have a minute so I'll explain again: Java forces you to use classes and heavily encourages you to think about the problem in an object oriented way. (You don't have to of course if your classes are just function containers or bags of data but that is terrible Java. and as someone who knows how to write Java I'm sure you wouldn't approve.) This is not good because most problems are not cleanly represented in the pure OO inheritance world of Java. So while other languages may have inheritance, they also have more sensible ways of modeling code so that you never have to touch inheritance. Caveat with TypeScript: you should know about prototypical inheritance but you rarely have to actually use it. You can write a NodeJS application in TS that doesn't touch prototypes or explicit classes ever.
@@harleyspeedthrust4013
You don't need to use a lot of inheritance in Java. Stop trying to sneak in that assumption.
@@HrHaakon I didn't sneak it in. I made it pretty clear that you don't have to use inheritance in Java but if you don't then you're writing what Java people would call "bad Java." Maybe next time read the comment.
2:40 that is because they don't tell/show you how painful it is to change the structure of it after it has been coded . They show it from the perspective of a world where everything is easy, and is correct on the first try.
(I have not gone to college for programming, so I don't know if that is the case, but it sounds like that is why)
I have studied OOP in college a long time ago and this is exacyly how it is shown, OOP has its place but we have gone too far in making everything abstract and that in turn made things worse.
Schools teach you how, not why.
So when they introduce a new hammer, people want to try and hit everything that looks even remotely like a nail with it.
I really connect with Prime's idea of 3 stages of writing an application.
Coming from embedded development, I am very used to separating data and methods on data, and even now that I work on higher plane of abstraction I often always write my first version as a combination of data containers and methods on them. What I find is that often times on subsequent re-factors that pattern alone is sufficient without even going towards OOP way of structuring your code. I am not generally opposed to it, but in my experience it's far easier to first do the simplest thing first (data + methods on data) and if you spot patterns - refactor them however it makes sense to refactor them.
Using classes to structure your code from the beginning IMO is hard to get right even for experienced devs, inheritance or not, you are still effectively dealing with lots of local namespaces and are dealing with "this object owns this and that object owns that", if you then need to share some data, it's more work to make these changes compared to if they were simply passed in as a parameter to some method
I also feel that having a very clear way of identifying data within your program is often very beneficial and illuminating. I suppose this is more towards data oriented programming approach
My experience is that if you're having a clump of data, and then some operations on that data, it's pretty easy to get it done using OOP. Will you have to refactor a lot? Maybe. But it's pretty easy to think in terms of.
Then again, I model the database FIRST these days, as nailing that is pretty reasonable, and then the OOP layer on top handles messaging between the data model and the front end. So it's like the controller in the way MVC was originally conceived.
It works really well for enterprisey solutions.
That description of developing software in a 3ply-fashion is brilliant. Totally tracks sometimes you build it to understand what it does, then build it to make it ready for others, then build it to make for you and others in the future.
dependency injection was our bread and butter in high school robotics. We split the robot into subsystems and everything would be injected into relavent methods and classes. Was a massive pain
The solution to the image problem the video presents could be far more easily solved in practice with a single Image struct and some associated methods for editing + separate load/write functions to different image formats (maybe a ImgConverter interface if needed). The video is good but usually these OOP examples are overly complicated for no reason other than trying to fit a mental model of classes
That's the point, though. You have some class that works just fine, but then one day you want to extend it. Doesn't matter how cleverly your inheritance structure solves the current requirements. The moment a new requirement is added, the resulting refactor is huge.
Sure, you can solve the problem given here, with a drawable / non-save-and-load image just fine if you know that that's a requirement from the get-go. But then something new comes along.
Actually the method I proposed is way more flexible, as the image manipulation logic is separate from the conversion logic. Want more formats? Simply write a new save/load function pair and pass an Image struct to it.
@@marcs9451 How is writing more functions different from having them as methods in the classes?
@@MasterHigure The problem with inheritance when I used it (maybe bad at OOP) is mutation/global state is order dependent, and if you inherit something the child classes may sometimes want to call the super method first before the body sometimes last and it could lead to subtle bugs because of state, you sometimes have to have the super method be a higher order function so it can be called around your code. Also sometimes you have to have the parent call the parent version of a method other times you want the parent to call the child method...
@@MasterHigure extension methods 😎
Data duplication is good when it leads to a more decoupled codebase. You don't always want a giant users table with 100 columns because every part of your system needs parts of the table. Adding an extra column suddenly breaks everything because other parts of the project are now trying to insert null into the new column.
I believe pretty strongly inheritance is only good for composing orthogonal behaviors. Inheritance for specialization opens the door to overriding, calls to super and having to understand multiple versions of the same artifacts but jumping back and forth between modules to understand what each does. My brain doesn't compute that.
Code Aesthetics messed up with the inheritance diagram right before moving on to composition. The object on the right started with only blue and red square but ended up as a child of the green and blue object.
That's kinda the point. If your system is too messed up with ideas that are more complex than they need to be, then you'll make mistakes
Composition and Inheritance aren't opposites. They work together nicely, and someone with a good idea of both can write the best of both worlds
There is inheritance. A person inherits. Composition. Replace the person heart with a pig heart.
Vertically aligned Squirrely Braces is 1000x more readable. This isn't the 1970s where newlines are expensive.
So true. When learning programming, inheritance seems so logical and endlessly useful. Yet much of the more advanced programming literature/videos points out that composition is often better.
I agree with you on smooth scrolling. There's a reason why scrolling was originally in steps and it's because it causes less motion sickness. I am not particularly prone to motion sickness, but I find it much harder to read stuff when they're smooth scrolling, even on a 144hz monitor.
The greatest demonstration of this are youtube live comments vs twitch, where the former is smooth scroll and the latter is bulked into steps. Twitch chat is legible when it's being -SPAMMED-, I'm talking 100 messages a second spammed. TH-cam chat is barely legible with 5 messages per second, because it's constnatly scrolling up, there's no rest for your eye to actually read the words.
An example of where inheritance makes sense is Pokémon: you have the definition of a pokemon(name, evolution, moves it can learn, abilities it can have, base stats).
Then you have a class to express a pokemon in your team (has ivs, evs, only four moves out of whatever it can learn, a personality, only one ability) and from the pokemon class it helps to validate if it can learn the move, calculate actual stats based ob ivs and evs and so on.
But then I wouldn't do a battle class that inherits from pokemon, items and so on
I got to work on an old project where all the important bits happened in business objects that were pure spaghetti of inheritance. EVERYTHING happened through protected variables and methods.
You would see a 5-line function and think that that should be easy enough to understand... but then it ends up triggering 500 lines of code across 5 levels of inheritance. And it would go up and down, up and down...
Line 1 out 5 executes something in its parent, but that is passed right through to the grand parent. So the grand parent does some things, then passes it to the GRAND-GRAND-parent, which executes like 10 lines before it goes back to the grand-parent. The grand parent does a few more lines before calling something that passes through to the GRAND-GRAND-GRAND parent... which calls some completely other module... and from there it falls back to the original child object I was working on.
One line done, get ready for 4 more...
as a silver haired coder from the 1980s OOP & inheritance was the bees knees. on paper. awesome. in reality - i watched programmers inherit the crap out of stuff, till it was just crap. So deep that even a Matryoshka doll would have gasped. also kudos for your music tastes there with miss monique.
your advice on capital D Design and your 2-level way of writing production code is really REALLY helpful. Thank you.
hey dude- plz make a video explain dependency injection and inversion of control for the masses
Now I understand why Rust traits work in such a way. It has the advantages suggested in this video, without the disadvantages that comes from inheritance.
yep
trait system is so good I wish it existed everywhere else.
@@betterinbooks I agree. Once I was coding a server backend. I created the struct for the internal state that is going to be accessed by the endpoints. I then created a trait with the methods for accessing the state.
For each endpoint, I created a unique trait for it with the behavior of that endpoint. The trait applies to any struct that implement the original trait type.
Thanks to that, each endpoint is its own trait. We just have to implement these in the state type. Instead of one file with all the business logic, I was able to divide each into its own trait. Extremely easy to scale. Loved it so much.
Using that structure, we could have several people working and pushing to the same repository at the same time. Each developer needs to simply work on their trait.
"What is this, C#?" he says, while staring at a snippet of C# code. Yes, in fact, it is C# :p
This video so purely speaks to an idea I've been thinking about for a long time. I keep seeing these videos floating around about why OOP is a lie, and why it doesn't do what it was made to do. Which was solve the lack of modularity in imperative programming.
To be fair, it doesn't solve that problem, we have modules now. However, this is exactly why OOP has evolved over time. It's not about trying to make infinitely reusable code snakes, it's about breaking down large frameworks into easily definable objects that can be instanced to serve a higher purpose.
I clang-format on save with whatever defaults it considers the best. So because IDGAF enough to create a .clang-format file it does my braces on new lines. At first it would cause me geniune aggrievement but honestly you learn to accept it and move on. I work with C++, my life is shitty enough.
This is the first reaction video I liked - ever. You complement each other so well. It is like you inherit his evil genius and implement an emotional amplifier of profanity into it.
How about people think long and hard about what their classes look like before writing them. Why the hell would an Image have any concept of save and load when the sole purpose of an image should be to represent an image as data? Having anything extra is over abstracting, class should be the bare minimum and nothing more. I'd keep image the way it was (without save and load, and maybe even the transformations) and have the abstract class for saving and loading. Interfaces to me are for functionality which isn't restricted to any concept. For example if you wanted to have a collection of things which are all unrelated but all have save and load then I'd use an interface, that way I can just create a collection of what ever interface, ex: ISerializable, IEnumerable. If I want to save and load strictly for images, I'd have an abstract class inherit from Image and put abstract save and load methods in there. Now I can have a collection of images which can contain all image types as well as a collection of all "FileImage" types if I wanted only images that have save and load. This also doesn't break implicit conversions from Image to any FileImage.
And for those saying null-conditional operator is hiding a problem, I'd argue that it exposes a problem that was previously hiding. And null in itself is not an issue, its like saying instead 0 as a guid is an issue, its just a state. It is your responsibility to handle what state its in. Asking to get rid of null entirely is like asking to get rid of 0 entirely.
9:15, there should be quotes around the word perfect in "change is the enemy of perfect design". A truly perfect design enables and embraces change and provides seamless extendability without needing breaking changes.
His curly braces are symmetrical. That's real esthetics - real beauty. 💯
I great thing I heard to use was "Is" vs "Has". If something "Is" something else, then you use inheritance. If something "Has" something else, you use composition.
For example, "Is" a Png an image? No. Png "Is" a format for saving images. As are the other formats. So we can separate these from image and have them inherit from some ImageFormat class which contains methods to take in an image and save it and return a saved image.
"Is" a DrawableImage an Image? No. But a DrawableImage "Has" an image that it can draw to. So we can separate DrawableImage to its own class, and give it an image to perform drawing operations on.
I like your point at 10:00.
First build is discovering what you don't know.
Second build is learning the things you didn't know.
Third build is starting from the beginning with full knowledge.
24:06 This describes my hatred of DI *frameworks*. I love dependency injection and I use it a lot in Go, but the dependencies are created manually and passed around in code in such a way that's easy to trace and debug. I hated DI in other languages (C#, TS [Angular], PHP) because it almost always involves some sort of DI Framework that over complicates everything and turns DI into a magic global black box that's a pain in the ass to debug and manage configurations.
The video doesn't present a problem with inheritance. If anything, all it does is reveal the necessary foresight and experience required to construct a feature-rich, well-designed API. It just so happens that the lack of such things are naturally inherent in humans, and this by default.
Also, inheriting from an interface *is* inheritance.
6:25, just use the virtual keyword instead of abstract. This will allow you to define and write code in the function inside the abstract class.
This allows you the ability to override it in inherited classes while not being forced to.
1 for agree
2 for disagree?
Only 12 zeroes
My biggest problem with composition in Rust, especially those default implementations for functions of a trait... is that when these default implementations rely on state, you always have to provide accessor functions within the trait... and in rust you often need two for the same value, one for immutable and one for mutable state...
this forces the struct that implements said trait, to implement these accessors even if they just point to a single field in the struct... depending on how many such traits you have this can cause a lot of boilerplate code
it's C#, the opening brace on a separate line is the default and the norm there.
the methods that start with a small letter, though... A SIN! AN UNFORGIVABLE SIN! the only things that start with small letters in C# are local variables, parameters and private fields! (private fields also typically start with _ but then a small letter follows)
(I'm talking identifiers, of course keywords (int is also a keyword) start with small letters)
i put squiggly braces on the next line in protest; if a language is forcing me to define blocks with braces, then I am going to make them as prominent and annoying as possible
Those curly braces on the same line are standard practice in C#. Any linting tool would complain if you did anything different to that. They are called automatic properties and they are awesome.
Meme, not standard practice. Entirely due to historical VS defaults.
@@Ormaaj You are wrong.
🤣No I have superior taste.
@@Ormaaj Wrong about that, as well. That's the thing with you. Always wrong.
Whatever newline-curlybracer. At least I'm not wasting all the world's supply of vertical whitespace
The one thing I've never understood is why full on complex class objects is considered better code reuse than separate functions for common tasks.
Like classes make sense if the class covers every single use you have, but that's rarely the case in any moderately complex system.
I guess this is what they mean by composition over inheritance, but inheritance seems more like a convenience and organization advantage in certain cases, not a code reuse advantage.
Java Swing goes full in on inheritance, and it's the only one case that I found actually useful and easy to hook into. Building UI, quickly changing a draw method here, or making a table behave like a tree. Once I got it, it was really easy. But designing that stuff myself to be flexible like that, no thanks.
I also made my first game with xna, a tower defense, and I also had like 4 levels of inheritance for only the attack effects lmao
Oop obsesión was so bad when I was young
You have no idea how much I can relate to this
How to execute:
1. Dicks out for Harambe
2. Shit code
3. Full rewrite no matter what "statistics" tells you
Took me over 10 years to understand how true this is
25:03 Yeah, I've thought about this a bit and have been curious if it's not a bad idea to have an `embodies`, `compose`, or a similar keyword. The idea would be that within your class you would say that it embodies another object, which you can then access it's methods as if it's your own, but they would be private. Mostly a convenience feature, it would benefit from the IDE informing you which component you are fetching the method from, but I haven't thought of a scenario where it would be dangerous (yet.)
People will say "inheritance sucks" and "interfacing is great" in the same breath as if interfacing isn't just a form of inheritance.
Is it though? In a dynamically typed language, if I implement a common set of operations on some classes, then I have implicitly implemented an interface. That Java requires implementors to be explicit doesn't suddenly make it inheritance. There's also the question of what is being inherited: besides default methods, there isn't any behavior being shared by an interface to its implementors. I don't think there's much inheritance at all.
Imo the only good way to have this conversation is by looking at what the compiler does.
For example each time the compiler sees a dereference (dot operand) it performs an extra instruction for null checking, while local variables are accessed directly.
And if we get even more exquisite... Local assignment(dereferencing a field to the method stack is supposedly even faster... But the assignment tradeoff needs to be matched against the number of times the reference will be accessed.
The performance hit is probably insanely negligible since many instructions are pipelined and executed partially in parallel. By this logic, all OOP languages would be slow, but they arent
Inheritance is not bad. The code reuse alone is much better than any type of convoluted dispatching argument. The best, most bug free code is code that is never written or re-written. Without inheritance, you are guaranteed to write code over and over for no reason.
People just really dont understand inheritance in my opinion.
And if you are so concerned about performance, write everything in assembly.
@@jshowao I like OOP, but I won't deny it's drawbacks, specially in such limited scope tests like a simple loop.
JIT is better at optimizing code than a well written C, maybe even a very good C.
But the time it takes for the optimization to kick in... Takes some time.
Another thing about VM languages is that their compilers are really not that smart... Maybe because of exceptions and extensibility but devirtualization should happen a lot more... yet is something that rarely happens. Each virtual call adds a whole instruction... which eventually adds up...
Each extra instruction call is ONE LESS op code of instruction sequence, which is one less op code for the speculation sequence.
The worst offender is that each virtual function call forces a memory order boundary, which means that the compiler is not allowed to interleave beyond the virtual function call... this is the worst offender.
@@jshowao I love OOP, I just wish compilers were more confident when devirtualizing and parallelizing up front, maybe even add keywords to hint devirtualization.
Also I'd love for compilers to not implement 'freezing' so strictly, instead make a flow graph and ignore the constructor memory barrier if the logic allows it... do it up front instead of JIT.
This would allow swaths of lengthy op code sequences... Making massive and faster speculations uninterrupted.
I keep being told that if full de-virtualization would be applied, some pattern would break when dynamic dispatch is not obeyed, yet I keep getting some weird/nonsensical and incomprehensible examples of this...
in React, if you making heavily nested reusable "functional" components - they're effectively classes with multi layers of inheritance and you run into same problem of everything being too interwoven
I always cringe deeply when I see inheritance being used in Angular
~8:00 OOP inheritance can easily go out of hand when doing games, classic orc merchant npc issue. It's not nearly as bad when you have hard defined scope, although thats probably the crux that you tend to expand that scope over time and now suddenly your child classes bring along garbage thats only relevant to some children and you need to override stuff
This feels like almost an exact replica of a Raymond hettinger talk, but there was obviously a lot of care and love put into visualization and video development.
I love new line curly braces looks cleaner and more symmetrical.
It just depends how many "branches" of your core class there end up being. If say there are 30 variations.... then you'll probably have a hierarchy of inheritance with 3 or 4 "generations" to stop objects from inheriting too much. That can require a lot of pre-planning and can easily end up a mess. I'd usually prefer to use the modular approach of composition to form each sub type, but it really just depends on how many variations we plan on using and how worried we are about memory.
I use both new line and same line curly braces at work depending on the language and their relative best practices, so yeah you don't die from it
These design patterns help a lot when you need change the code with minimum impact.. sometimes can be over engineering, but most pf cases not!!
Prime continue with these videos because your lives most of time is in my work time and sometimes I can't watch and the Vod is to long!!!
awesome! i love to hear that dalton :)
.dmg -> a damage
.rpm -> a rampage
.AppImage -> Appalachian sized Imagination
.exe -> use Linux please
100%
The problem with "dependency injection" is that ppl get carried away with it, and you end up in a situation where inheritance is actually better. The best way is to get rid of classes all together - in c# you're rooted though cos everything is a class.
> A [pee enn gee] is called a ping
I'll update my pronunciation in a giffy.
That guy in the comments "grepvyne" made a great point:
"This is my fear writing the piece of shit version, that my manager will just say yep off to prod it goes and it'll never get re-written properly"
My tech lead told me this is exactly what happened years ago to the software we're currently maintaining and why the code base is a mess
Also, from hence day fourth, I will now pronounce NoSQL as "no squeal". Thank you
I already watched this video, but watching again along with you was so fun…
my current strategy is, "Can I do it with an interface?" If the answer is yes, I don't use inheritance. There are some cases though where inheritance is super useful, but I live mostly in the game coding world so it probably doesn't apply elsewhere. As an example, a game would have a base object, that keeps track of temperature, who owns it, etc. Then more specific objects with certain functionalities can inherit from it, without having to duplicate a bunch of code. Basically for me it comes down to how much code you're duplicating
You create classes for each of your properties, like temperature, and add them to the class of the object you are creating. Then, you make a factory with methods for instantiating your object with valid combinations of your property classes for creating different types of objects.
companies should be interviewing people by having them configure the linter for the language they’ll have to write. nevermind if they can cook up a linked list problem in 40min, find out of they want all code to be wrapped to exactly 80 columns and convert tabs to 5 paces
Did XNA in college senior year, I loved it. Glad to hear I wasn't the only person who used it. I made a pacman and tetris clone in it and was so excited to deploy it to the first generation of windows 7 phone.
RIP windows 7 phone, never got flash support and was too late to the market
What do you think of query builders? I honestly can't stand query builders for sql. I'd rather look at a basic string with some replacement keywords or embedded variables.
they always seem so neet
then... they become less neet
@@ThePrimeTimeagen just... Why should I be learning a new structure/language with your query builder? If you just keep it as sql I can read it without having to figure out your custom representation of a query.
String replacement can take a long walk of a short pier, but SQL is better than 90% of everything else the SECOND you're no longer doing basic CRUD that's 1-1 with a single table with no joins.
When implementing a compiler inheritance is elegantly straight forward. Functionality is presumed from the way objects are structured in memory. Where as composition there has to be an active mechanism in determining if such particular object contains a particular functionality. Maybe inheritance is a house of cards that may apart due to design issues. But when it works it should be more efficient then composition
Inheritance is an O(0) solution where as composition is O(1) at best.
The triple method - harambe moment at 10:05 is pure gold
Oh baby a triple!
It's indeed the best way to produce quality software. Developers who think they can do it right the first time just haven't learned that lesson yet. RIP Harambe
What happened there is that the parent class Image was actually transformed into a mutable isolated class for handling the operations; a better name from this image would have been ImageBuffer (which is what it actually represents). The error at the beginning was that thinking having PngImage and JpegImage import from ImageBuffer means it could be a good idea to place save and load there. There was a problem of cohesion: save and rotate shouldn't have been placed on the same class. If you want to solve this issue using inheritance, one possible solution is to declare save and load in an interface, not in the abstract Image class.
I'd rather prefer what haskell does with GADTs and type classes, so that the cohesion problem disappears, you just add instance methods in the correct module without having them to be in the same file as in OOP.
The semantic meaning for inheritance has always been taught as “X is a Y” is X inherited Y.
The reality is that the true semantics is “X is only a Y”.
Reasoning being X is/can be many things, however inheritance only allows for a single “is” and therefore “is only”
I love this channel and the long video format, for me that can't manage to watch the streams, this is gold
@2:14 -- yes, it makes perfect sense on paper ...just like the waterfall model for development made perfect sense on paper.
So...I don't see why the proposed solution isn't functionally equivalent to multiple inheritance. The entire idea of interfaces being "lightweight" but (in some languages) allowing you to give default implementations that reference other methods of the interface... isn't that just a smaller parent class?
I don't see how inheritance precludes any of the solutions to the given problems. "ImageFile" could also be an abstract parent class, as could "Image." Then your JPEG, PNG, and Bitmap all inherit from both parents, while drawable only inherits from "Image." Am I missing something?
I do inheritance, but it's almost always just one level, very rarely two. I often need very slight variations on different objects, and majority of the functionality comes from the parent. If some of those variations are needed in multiple classes (but not in *every* of them), then they belong into interfaces and become composition.
A perspective that helps me understand why Dependency Injection has similar problems to inheritance: There is an intrinsic tension between code-bases that are flexible at runtime, and a code-base that is flexible to change by developers. Overgeneralizing slightly, the more runtime flexibility, the harder it is to reason about the code and thus the harder it is to change the code when your needs change.
Both DI and inheritance are flavors of polymorphism, and polymorphism makes it hard to reason about code when some change needs to pierce the polymorphic abstraction. It can be fiendishly difficult to get a grasp of which code uses other code when it's all hidden behind polymorphism.
So, I'm not a fan of inheritance nor DI. They have their place. But they're usually _too_ powerful. When you really have multiple implementations and need to mix em up for whatever reason, then encapsulating polymorphism is useful. But usually, it's unnecessary, and then it's simply a burden to comprehensibility. Being able to trace which code is used by what _exactly_ is very valuable; so giving that traceability up (by hiding behind polymorphism) needs to have a very concrete high-value upside to compensate for losing that ability.
I agree that using ?. without another condition is terrible. I think it has a ton of use though, as long as you follow it up with "?? someDefaultValue" or other handling. Hiding an error with it is terrible. Consolidating error handling with it, however, is beautiful.
You joke, but the next-line curly braces is the main and probably only reason why I refuse to use Black for Python.
I just can't. I can't.
I'll put my squirly braces where I damn well please.
C# is partially written using C#, so you can pass in its own compiler. My Lecturer did his pHd on AI optimising and writing code.
10:05 A typical conversation at my job:
I: Should we really change this thing to work totally different? It will be difficult to restore it. I'll remove all the stuff I don't need anymore.
> Sure, just change it. We decided for the perfect design, it won't change anymore.
A few months later:
> I got an idea, for a new design, we should do it like this.
I just think, that's almost exactly how we already had it.
Inheritance is more about modeling the domain and that subclasses ARE an instance of the superclass. Composition is about HAVING something. Those that argue they are solving the same problem are mistaken. Interfaces/traits are for a shared set of behavior with different implementation. They are all different and have their place. The problem is when a language forces a developer to only use one.
Changes to the composed object can still force extensive refactoring of all the users of the composed object. Geez.
The point is that you can compose an entirly different object and leave the existing compsed objects alone.
I use next-line braces. It puts a little extra vertical space in exactly the right places.
6:32 How about having a File class and an Image class. If a particular subclass of Image needs File I/O, then let that subclass extend the File class. If not, then the subclass of Image doesn’t have to extend the File class and hence doesn’t have to implement the File I/O methods
I've been viewing your back catalogue of videos and this is by far the best one so far. Started programming with COBOL and DB2 on IBM mainframes back in 1990. Now I just do Godot and Python for selling small games off Steam.
Another problem is that college tends to expain interfaces VERY POORLY and inheritance much more usably (at least as per what I saw) I remember I used abstract class somewhere where interface was fine just because I did not know interface as well
Exactly. The absolute best way to code is to fever dream a quick working example. Take a step back, go for a walk, think about what just happened. Then start again from scratch. I guarantee you won't need OOP if you really think about what code is necessary.
The ? in C# doesn't hide a problem, it just is a short hand for if(obj != null)...
If you don't know what you are doing it doesn't matter in how many ways the language offers you to do it wrong in.