Related: https://factoryfactoryfactory.net/
That’s actually where the title is coming from…
Why is the joke with Java always factories? Factories are really super useful in a dependency injection context.
I’ve been working primarily in Go for the past five years, including some extremely complex projects, and I have never once wished I had dependency injection. It has been wonderful. I have used dependency injection - previously I worked on a C# project for years, and that used DI - but I adore Go’s simplicity and I never want to use anything else (except for JS for UI, via Electron or Wails for desktop).
Edit: If we’re talking about dependency injection in the general sense (separation of concerns, modularization, loose coupling), then yeah I agree that’s kind of critical to writing good, maintainable software. When I hear “dependency injection” I think of frameworks such as Unity, and that is what I was specifically talking about - I am very happy with the fact that I have felt zero need to use any framework like that over the last five years.
Go programmer here: What does Go’s simplicity have to do with dependency injection? What does a language itself have to do with dependency injection?
Reading your post and not being personally familiar with your work, I do wonder, perhaps your “extremely complex projects” wouldn’t be so extremely complex if you practiced dependency injection?
How do you unit test your extremely complex projects if your business logic carries the additional responsibility of creating objects?
I say it’s all about data flow and composability, if it’s pretty much always in one direction (modular tree structure/architecture) then you just don’t need all these “patterns”…
What does Go’s simplicity have to do with dependency injection?
In my experience, following Go’s philosophy of simple solutions eliminates the need for complex solutions such as dependency injection.
How do you unit test your extremely complex projects if your business logic carries the additional responsibility of creating objects?
I write modular code that accepts interfaces so I can test the components I want to test. The vast majority of object creation happens at initialization time, not in the business logic. For the projects I’ve worked on, that would be true with or without DI - I don’t see how that’s relevant.
perhaps your “extremely complex projects” wouldn’t be so extremely complex if you practiced dependency injection?
When the CTO says, “Make it distributed and sharded,” I do what I’m told, but that is an intrinsically complex problem. The complexity is in the overall behavior of the system. If you zoom in to the individual execution units, the business logic is relatively simple. But the behavior of the system as a whole is rather complex, and DI isn’t going to change that.
Edit: I was interpreting “using DI” to mean using a DI framework such as Unity, and I would be happy to never need one of those frameworks ever again.
I write modular code that accepts interfaces so I can test the components I want to test.
basically dependency injection
Maybe I’m misunderstanding what “dependency injection” means. When I hear “dependency injection” I think of a DI framework such as Unity, so I thought “using DI” meant using one of those frameworks.
… until you’ve heard of Rust :)
(I think Go takes all mediocre language features together and makes an even more mediocre language TBH, take error handling for example, or generic programming (which I agree should be used sparingly, but is super useful if you need it))
Go does have generics nowadays, although they have some limitations (like pointer types being inefficient because reasons).
But yeah I’d tend to agree. Go’s strength and explicit design goal is that it’s relatively easy to learn and get started with, meaning it’s fast to onboard new devs. It’s very much a “get shit done” language.
Its weakness is – to paraphrase someone’s criticism of Go – the core dev team’s extreme unwillingness to adopt any programming language and tool chain designs and patterns invented after the 70’s. Pike’s rationale against generics was that in cases where you’d normally reach for generics you can either use interfaces (and especially
interface{}which is like Go’svoid*and throws all type information out the window and is slower than proper types because reasons), or just copy and paste code. Because what you as a developer want to do is reimplement something likeFind(needle SomeType, haystack []SomeType)for the Nth time in every project, or take a performance hit and lose all type information by usinginterface{}.And don’t even get me started on how long it took for Go to get proper dependency management and what the arguments against it were, Jesus fuck.
Go is currently the language I’m most fluent in after having written mostly Go for something like 8 years or even more (I remember when
errorwasError, pre 1.0 I think) and the one I’d be the most productive in, which is sort of unfortunate since I really don’t like it as a language 😁 Learning Julia at the moment though, since I’m going back to the university to study some more computer science and maybe get into evolutionary algorithm research, and Julia is a neat language for lots of different number crunch-y tasks (no I won’t touch Python, significant whitespace is a crime against common sense and fully dynamic typing gives me the heebie jeebies)I’ve heard of Rust. It sounds noisy and even more verbose than Go, which is already a fairly verbose language. I haven’t had any reason to learn Rust, so I haven’t done so. The error handling is annoying but at this point I don’t really notice it any more. And as interolivary said, Go has generics now.
Yeah this was my initial reaction way back when I first heard of Rust as well (sometime around 2015 or so I think). TBF it’s definitely not on the same level as e.g. Haskell. But it’s generally I would say less verbose than go (or at least has verboseness where it makes sense compared to go IMHO).
A good article about this: https://matklad.github.io/2023/01/26/rusts-ugly-syntax.html
The generic system is also (way) less powerful compared to Rusts (The trait type system/type-classes is really a nice Haskell-inspired thing, that I don’t want to miss anymore). Also the lack of sum types and proper pattern matching makes go more verbose IMHO.
*grumble*. I dabbled in Scala a few years back and I am really grumpy every time I remember that Go doesn’t have sum types, pattern matching, and the closed union of types construction you can create with an abstract final class in Scala. I loved that last one and used the heck out of it. I would love to have a compiler-enforced guarantee that a set of types was closed and could not be extended.
How do you write tests?
How does dependency injection have anything to do with writing tests? I write tests by writing a test function that executes the code I want to test…
I mean unit tests. I work on Spring Boot apps where there are distinct layers (controller -> service -> persistence), and you generally inject mocks into your object to isolate tests to the specific code you want under test. One benefit of this approach is that it’s pretty easy to get 90% coverage.
Do you write unit tests with objects mocked via interfaces? Or polymorphism via interfaces? Those are the main reasons to use DI.
I guess I have to start calling function invocation with generic parameters, fancy names (like “dependency injection” ^^)
Relevant:
I also prefer the simpler explanation.
https://www.jamesshore.com/v2/blog/2006/dependency-injection-demystified
As I read somewhere:
Dependency Injection is a 25-dollar term for a 5-cent concept.
Yea that works too.
I use various strategies depending on what seems appropriate, including the two you mention. I’ve never felt the lack of DI.
Btw. why don’t interfaces work + “as argument” (whether it’s a constructor in an OOP context or a function parameter)? I think interfaces are exactly built for such a use-case (without all the boilerplate that’s necessary with dependency injection as well as factories).
What are you talking about? At least in Java and PHP you can absolutely declare constructor and function parameters as interfaces. As you say that’s exactly what they’re for.
Yeah that’s my point, I don’t see a use for factories here, because it’s possible to just use generic parameters (whether it’s constructors or function arguments).
I’m somewhat confused by your statements, so perhaps I don’t understand.
Function/objects that allow changing their behavior by passing different objects into them, based on some interface, is called dependency injection. Some subset of behavior is determined by this passed behavior. E.g. To keep a logger class from having to understand how to write logs, you could create a WriteTo interface and various implementations like WriteToDatabase, WriteToFile, WriteToStdout, WriteToNull.
When you create this example logger, you’ll need to make a choice of what object to pass when you write the code. e.g.
new Logger(new WriteToDatabase(config))But maybe you don’t want to make that decision yet – you want to let a config file decide which writer(s) to create. The pattern to pick between dependencies at runtime is called a factory. In this case, you might make a WriterFactory to pick the right writer, or perhaps a LoggerFactory to hide the creation of both Writer and Factory objects.So, a factory is only really a facade to hide the runtime switching of how an object is created.
Also, the term dependency injection often gets confused with what you see in various Java / C# / and various frameworks in other languages – those usually use what’s called a “DI Container” or “IoC Container”. These manage and facilitate how dependency injection happens within the project, often with various annotations (e.g.
@Autowired). These containers are powerful, but sometimes complicated.However, you can absolutely still do DI without DI containers, and I think advocating for not using DI generally (and related patterns like factories) is rather misguided.
I still haven’t really understood the use (and use case) of “dependency injection” (and it feels to me I read now everything about dependency injection I could find), to me it seems to be yet another
ProblemFactory.Single responsibility principle: is your
GetData()function responsible for getting data? Or is it responsible for creating a new database connection and also using that to go get the data?Start naming your functions by what they really do. When you see the word “and” in your function name, you know your function is responsible for too much.
Dependency injection is the difference between
CreateDatabaseConnectionAndGetData()andGetData(connection ConnectionType).In the first example, that function will always connect to the specific db that you hard coded in it. It probably has to also read in a config file to get the connection details. Maybe you should name it
ReadConfigAndCreateDatabaseConnectionAndGetData()?In the second example, I can pass in a MySQL connection or PostgreSQL connection, or some dummy connection for testing.
Keep all that nasty dirty untestable code in one place and spare your business logic from owning all of that.
Thanks for the write up, but as I said, I know and I’ve read all about that already. I still cannot see, why a simple function argument and an interface isn’t enough (you can probably already call that “dependency injection” if you want to get fancy)
I guess I have just divorced with OOP and the “necessary” “design patterns”…
Things are more simple and less boilerplaty now for me :).
My brother in Christ, that is dependency injection. Just because you don’t want to call the spade a spade anymore doesn’t make it not so.
Yeah I guess you can call it like that, I’ll just call it function invocation…
You’re gonna have a tough time talking to others about your code if you don’t agree on common terminology. Function invocation is just function invocation, it doesn’t say anything about the form of the parameters or composition. Dependency injection is a well known and commonly understood method of facilitating decoupling and composition by passing a function’s dependencies as parameters. From your comments you’re talking about the second, but refusing the name, for… reasons?
I guess I’m a little bit too long already in the functional/data-driven world (after being a decade in OO languages (IMO too long…)). In OOP you may need a separate term for that.
But I think it’ just not really necessary in functional programming, it’s just another parameter when calling a function, that may be a somewhat type-constrained generic (for testing e.g. with a mock implementation).
I mean function parameters are somewhat dependencies anyway, so should I call all my parameters now dependencies and invocation “injection”?
“Dependency injection” is just a term for providing a function or method with its dependencies rather than making the function go and gather them itself.
It’s (typically) done through parameters, but it’s still more specific than just invoking a function. It describes how that function was written and the reasoning for certain parameters. To the other commenter’s point, you’ll have a hard time communicating about your code with other developers if you refuse to use the term dependency injection just because you don’t like OOP.
Di is just good functional practice. I’m not sure it’s a super importent idea to someone who knows how to write a good function.
Edit: “a function should do one thing and its operands should be passed as arguments” for the OO world.
I’m just lurking around, so don’t mind me. But I gotta say, it really does sound like y’all are just making shit up sometimes lol. Like a mechanic trying to charge you an extra 50 bucks because your jindle shaft needed an alignment
I’ve tried studying CSa d programming, both formally and informally, for long enough to know most of what y’all are talking about, but failed terribly enough that I still have no idea how people can actually build code. Syntax is easy enough, I always started a class like oh cool I can do this. And then they drop me off the deep end and I fail, every time. I work in IT infrastructure now and I’m happy to say I never have to worry about coding, though my foundational knowledge is useful for troubleshooting.
It’s all about experience IMHO, just start small projects (and in my experience ignore all these “design patterns”, I learned quite a lot of patterns, but I’m not really using any of them nowadays, maybe sometimes intuitively, (as this thread shows, “dependency injection” which can be a fancy term for generic function parameters).
Well maybe learn them, but take everything with a grain of salt, I think. Intuitively thinking is often better (with the drive to do it better). Try to write code such that it’s easily readable exactly focused on the problem that it’s trying to solve, not anything fancier (this is actually a very simple but effective Mantra). Otherwise it often leads to overengineering (all these “design patterns” for example…), or premature optimization. (E.g. something like the popular book “Clean code” is IMHO full of antipatterns, like the examples are definitely not something I would do, they are inefficient, boilerplate, and often make the code unreadable).
Generally speaking the use case is writing tests. If your tests just call all the dependencies and
newdirectly then it’s harder to write tests for your specific component while avoiding setting up a whole bunch of stuff (to make all those other classes work). By requiring all the dependencies to be provided to the class, you can swap them out at test time for something else that’s easier to work with.That said, IMO it’s a symptom of problems in language design. Using DI is only necessary because languages like C# don’t make it easy to mock out
newor classes used directly, so we resort to wrapping everything in interfaces and factories to avoid those features and replace them with ones that are easier to mock. If the language was designed such that those features were easy to replace during testing then DI probably wouldn’t be a thing.It’s probably a general symptom of what people call OOP nowadays, in a more functional composeable world (where I’m living in currently). You just use function parameters and interfaces (or as Rust calls it “Traits”). But I still think in OOP, this is enough as well and the dataflow is more clear.
Just make a SolutionStrategyManagerManagerFactory. Easy.
SolutionStrategyManagerManagerFactoryBuilderImpl
Can you make it an Enterprise Bean?
I’m sorry, but why is this not an AbstractToolBuildingFactoryFactoryFactory? How are you going to derive alternatives?
Btw. to reference the original article where this is from:
https://stevenheidel.medium.com/a-factoryfactoryfactory-in-production-822478b5afbd
