Wow this book is a goldmine for architecture patterns. I love how easy it is to get into a topic and quickly grasp it.
Having said that, from a practical and experience standpoint, using some of these patterns can really spiral out into an increased complexity and performance issues in Python, specially when you use already opinionated frameworks like Django which already uses the ActiveRecord pattern.
I’ve been in companies big and small using Python, both using and ignoring architectural patterns. Turns out all the big ones with strict architectural (n=3) pattern usage, although “clean”, the code is waaaay to complex and unnecessarily slow in tasks that at first glance should had been simple.
Whereas the big companies that didn’t care for these although the code was REALLY ugly in some places (huge if-else files/functions, huge Django models with all business logic implemented in them), I was most productive because although the code was ugly I could read it, understand it, and modify the 1000 lines of if-else statements.
Maybe this says something about me more than the code but I hate to admit I was more productive in the non clean code companies. And don’t get me started on the huge amount of discussions they avoided on what’s clean or not.
I think one of the biggest problems I encounter whenever I hear that a project follows strict architectural patterns essentially boils down to too many obfuscated abstractions that hide what is going on, or force you to jump through too many layers to accomplish tasks.
Many files/functions/classes need to be updated to accomplish even simple tasks because somebody made a decision that you aren't allowed to do X or Y thing without creating N other things.
But in those companies that didn't care about architectural patterns its very likely that while there was more ugly code in certain places, it resulted in code with less indirection and more contained to a single area/unit or the task at hand making it easier for people to jump in and understand. I see so many people who create function after function in file after file to abstract away functionality when I'd honestly rather have a 100 line function or method that I can easily jump around and edit/debug vs many tiny functions all in separate areas.
Not to say having some abstractions are bad but the more I work in this field the more I realize the less abstractions there are, the easier it is to reason about singular units/features in code. I've basically landed on just abstract away the really hard stuff, but stop abstracting out things that simple.
I've come to the similar conclusion - just write the damn logic inline, and only decouple the parts which would make the whole thing difficult to test. Test decoupled parts thoroughly but in isolation.
Strict architectural pattern usage requires understanding the domain, and understanding the patterns. If you have both, navigating the codebase will be intuitive. If you don't, you'll find 1000 LOC functions easier to parse.
That's the problem, if you are working in a compagny which have mostly junior (1 or two year of programming), it is better for you to not implement to complicate pattern otherwise your day will be fill of explaining what a Factory is.
I found the book's use of modeling how to pilot an alien starship to be a little misleading, because a starship is a highly engineered product that functions in large part as a control mechanism for software. It comes with a clean design model already available for you to discover and copy.
Domain modeling should not be about copying the existing model -- it should be about improving on it using all the advantages software has over the physical and social technologies the new software product is meant to replace.
People are smart, and in most projects, there are key aspects of the existing domain model that are excellent abstractions that can and should be part of the new model. It's important to understand what stakeholders are trying to achieve with their current system before attempting to replace it.
But the models used in the business and cultural world are often messy, outdated and unoptimized for code. They rely on a human to interpret the edge cases and underspecified parts. We should treat that as inspiration, not the end goal.
> I found the book's use of modeling how to pilot an alien starship to be a little misleading, because a starship is a highly engineered product that functions in large part as a control mechanism for software. It comes with a clean design model already available for you to discover and copy.
Doctor Who fans will note that TARDIS craft seem to follow a different design: they regularly reconfigure themselves to fit their pilot, don't have controls laid out in any sensible fashion, and there's at least one reference to how they're "grown, not built". Then again they were also meant to be piloted by a crew and are most likely sentient, so it's also possible that due to the adaptations, the Doctor's TARDIS is just as eccentric as he is.
It's not like Doctor Who is "hard" sci-fi tho, it's basically Peter Pan in Space.
I love this book but yes, you really need to understand when it makes sense to apply these patterns and when not to. I think of these kinds of architectural patterns like I think of project management. They both add an overhead, and both get a bad rap because if they are used indiscriminately, you will have many cases where the overhead completely dominates any value you get from applying them. However, when used judiciously they are critical to the success of the project.
For example, if I am standing up a straight-forward calendar rest api, I am not going to have a complicated architecture. However, these kinds of patterns, especially an adherence to a ports and adapters architecture, has been critical for me in building trading systems that are easy to switch between simulation and production modes seamlessly. In those cases I am really sure I will need to easily unplug simulators with real trading engines, or historical event feeds with real-time feeds, and its necessary that the business logic have not dual implementations to keep in sync.
>I’ve been in companies big and small using Python, both using and ignoring architectural patterns. Turns out all the big ones with strict architectural (n=3) pattern usage, although “clean”, the code is waaaay to complex and unnecessarily slow in tasks that at first glance should had been simple.
The problem with "strict architectural pattern usage" is that people think that a specific implementation, as listed in the reference, is "the pattern".
"The pattern" is the thought process behind what you're doing, and the plan for working with it, and the highest-level design of the API you want to offer to the rest of the code.
A state machine in Python, thanks to functions being objects, can often just be a group of functions that return each other, and an iteration of "f = f(x)". Sometimes people suggest using a Borg pattern in Python rather than a Singleton, but often what you really want is to just use the module. `sys` is making it a singleton for you already. "Dependency injection" is often just a fancy term for passing an argument (possibly another function) to a function. A Flyweight isn't a thing; it's just the technique of interning. The Command pattern described in TFA was half the point of Jack Diederich's famous rant (https://www.youtube.com/watch?v=o9pEzgHorH0); `functools.partial` is your friend.
> Maybe this says something about me more than the code but I hate to admit I was more productive in the non clean code companies.
I think you've come to draw a false dichotomy because you just haven't seen anything better. Short functions don't require complex class hierarchies to exist. They don't require classes to exist at all.
Object-oriented programming is about objects, not classes. If it were about classes, it would be called class-oriented programming.
My experience matches this. It's so liberating as well. I find it easier to internalise such code in my head compared to abstraction-soup. As you can imagine, I like golang.
Me three. I'm even happy to refactor code into a form where there's less repetition and perhaps more parametrised functions, etc.
Finding my way around a soup of ultra abstracted Matryoshka ravioli is my least favourite part of programming. Instead of simplifying things, now I need to consult 12 different objects spread over as many files before I can create a FactoryFactory.
This has been my experience in working with any kind of dogmatic structure or pattern in any language. It seems that the architecture astronauts have missed the point: making the code easier to understand for future developers without context, and provide some certainty that modifications behave as expected.
Here's an example of how things can go off the rails very quickly:
Rule 1: Functions should be short (no longer than 50 lines).
Rule 2: Public functions should be implemented with an interface (so they can be mocked).
Now as a developer who wants to follow the logic of the program, you have to constantly "go to definition" on function calls on interfaces, then "go to implementation" to find the behavior. This breaks your train of thought / flow state very quickly.
Now let's amp it up to another level of suck: replace the interface with a microservice API (gRPC). Now you have to tab between multiple completely different repos to follow the logic of the program. And when opening a new repo, which has its own architectural layers, you have to browse around just to find the implementation of the function you're looking for.
These aren't strawmen either... I've seen these patterns in place at multiple companies, and at this point I yearn for a 1000 line function with all of the behavior in 1 place.
> Turns out all the big ones with strict architectural (n=3) pattern usage, although “clean”, the code is waaaay to complex and unnecessarily slow in tasks that at first glance should had been simple.
My last job had a Python codebase just like this. Lots of patterns, implemented by people who wanted to do things "right," and it was a big slow mess. You can't get away with nearly as much in Python (pre-JIT, anyway) as you can in a natively compiled language or a JVM language. Every layer of indirection gets executed in the interpreter every single time.
What bothers me about this book and other books that are prescriptive about application architecture is that it pushes people towards baking in all the complexity right at the start, regardless of requirements, instead of adding complexity in response to real demands. You end up implementing both the complexity you need now and the complexity you don't need. You implement the complexity you'll need in two years if the product grows, and you place that complexity on the backs of the small team you have now, at the cost of functionality you need to make the product successful.
To me, that's architectural malpractice. Even worse, it affects how the programmers on your team think. They start thinking that it's always a good idea to make code more abstract. Your code gets bloated with ghosts of dreamed-of future functionality, layers that could hypothetically support future needs if those needs emerged. A culture of "more is better" can really take off with junior programmers who are eager to do good work, and they start implementing general frameworks on top of everything they do, making the codebase progressively more complex and harder to work in. And when a need they anticipated emerges in reality, the code they wrote to prepare for it usually turns out to be a liability.
Looking back on the large codebases I've worked with, they all have had areas where demands were simple and very little complexity was needed. The ones where the developers accepted their good luck and left those parts of the codebase simple were the ones that were relatively trouble-free and could evolve to meet new demands. The ones where the developers did things "right" and made every part of the codebase equally complex were overengineered messes that struggled under their own weight.
My preferred definition of architecture is the subset of design decisions that will be costly to change in the future. It follows that a goal of good design is minimizing architecture, avoiding choices that are costly to walk back. In software, the decision to ignore a problem you don't have is very rarely an expensive decision to undo. When a problem arises, it is almost always cheaper and easier to start from scratch than to adapt a solution that was created when the problem existed only in your head. The rare exceptions to this are extremely important, and from the point of view of optics, it always looks smarter and more responsible to have solved a problem incorrectly than not to have solved it at all, but we shouldn't make the mistake of identifying our worth and responsibility solely with those exceptions.
> What bothers me about this book and other books that are prescriptive about application architecture is that it pushes people towards baking in all the complexity right at the start, regardless of requirements, instead of adding complexity in response to real demands.
The trouble is if you strictly wait until it's time then basically everything requires some level of refactoring before you can implement it.
The dream is that new features is just new code, rather than refactoring and modifying existing code. Many people are already used to this idea. If you add a new "view" in a web app, you don't have to touch any other view, nor do you have to touch the URL routing logic. I just think more people are comfortable depending on frameworks for this kind of stuff rather than implementing it themselves.
The trouble is a framework can't know about your business. If you need pluggable validation layers or something you might have to implement it yourself.
The downside, of course, is we're not always great at seeing ahead of time where the application will need to be flexible and grow. So you could build this into everything, leading to unnecessarily complicated code, or nothing, leading to constant refactors which will get worse and worse as the codebase grows.
Your approach can work if developers actually spot what's happening early and actually do what's necessary when it actually is. Unfortunately in my experience people follow by example and the frog can boil for a long time before people start to realise that their time is spent mostly doing large refactors because the code just doesn't support the kind of flexibility and extensibility they need.
Patterns and Abstractions have a HUGE cost in python. They can be zero cost in C++ due to compiler, or very low cost due to JVM JIT, but in Python the cost is very significant, especially once you start adding I/O ops or network calls
Some parts of this book are extremely useful, especially when it's talking about concepts that are more general than Python or any other specific language -- such as event-driven architecture, commands, CQRS etc.
That being said, I have a number issues with other parts of it, and I have seen how dangerous it can be when inexperienced developers take it as a gospel and try to implement everything at once (which is a common problem with any collection of design patterns like this.
For example, repository is a helpful pattern in general; but in many cases, including the examples in the book itself, it is a huge overkill that adds complexity with very little benefit. Even more so as they're using SQLAlchemy, which is a "repository" in its own right (or, more precisely, a relational database abstraction layer with an ORM added on top).
Similarly, service layers and unit of work are useful when you have complex applications that cover multiple complex use cases; but in a system consisting of small services with narrow responsibilities they quickly become overly bloated using this pattern. And don't even get me started with dependency injection in Python.
The essential thing about design patterns is that they're tools like any other, and the developers should understand when to use them, and even more importantly when not to use them. This book has some advice in that direction, but in my opinion it should be more prominent and placed upfront rather at the end of each chapter.
Could you explain how repository pattern is a "huge overkill that adds complexity with very little benefit"? I find it a very light-weight pattern and would recommend to always use it when database access is needed, to clearly separate concerns.
In the end, it's just making sure that all database access for a specific entity all goes through one point (the repository for that entity). Inside the repository, you can do whatever you want (run queries yourself, use ORM, etc).
A lot of the stuff written in the article under the section Repository pattern has very little to do with the pattern, and much more to do with all sorts of Python, Django, and SQLAlchemy details.
In theory it's a nice abstraction, and the benefit is clear. In practice, your repository likely ends up forwarding its arguments one-for-one to SQLAlchemy's select() or session.query().
That's aside from their particular example of SQLAlchemy sessions, which is extra weird because a Session is already a repository, more or less.
I mean, sure, there's a difference between your repository for your things and types you might consider foreign, in theory, but how theoretical are we going to get? For what actual gain? How big of an app are we talking?
You could alias Repository = Session, or define a simple protocol with stubs for some of Session's methods, just for typing, and you'd get the same amount of theoretical decoupling with no extra layer. If you want to test without a database, don't bind your models to a session. If you want to use a session anyway but still not touch the database, replace your Session's scopefunc and your tested code will never know the difference.
It's not a convincing example.
Building your repository layer over theirs, admittedly you stop the Query type from leaking out. But then you implement essentially the Query interface in little bits for use in different layers, just probably worse, and lacking twenty years of testing.
Repository pattern is useful if you really feel like you're going to need to switch out your database layer for something else at some point in the future, but I've literally never seen this happen in my career ever. Otherwise, it's just duplicate code you have to write.
In my experience, both SQL and real-world database schema are each complex enough beasts that to ensure everything is fetched reasonably optimally, you either need tons of entity-specific (i.e. not easily interface-able) methods for every little use case, or you need to expose some sort of builder, at which point why not just use the query builder you're almost certainly already calling underneath?
Repository patterns are fine for CRUD but don't really stretch to those endpoints where you really need the query with the two CTEs and the four joins onto a query selecting from another query based on the output of a window function.
I had a former boss who strongly pushed my team to use the repository pattern for a microservice. The team wanted to try it out since it was new to us and, like the other commenters are saying, it worked but we never actually needed it. So it just sat there as another layer of abstraction, more code, more tests, and nothing benefited from it.
Anecdotally, the project was stopped after nine months because it took too long. The decision to use the repository pattern wasn't the straw that broke the camel's back, but I think using patterns that were more complicated than the usecase required was at the heart of it.
> And don't even get me started with dependency injection in Python.
Could I get you started? Or could you point me to a place to get myself started? I primarily code in Python and I've found dependency injection, by which I mean giving a function all the inputs it needs to calculate via parameters, is a principle worth designing projects around.
> I have seen how dangerous it can be when inexperienced developers take it as a gospel and try to implement everything at once
This book explicitly tells you not to do this.
> Similarly, service layers and unit of work are useful when you have complex applications that cover multiple complex use cases; but in a system consisting of small services with narrow responsibilities they quickly become overly bloated using this pattern. And don't even get me started with dependency injection in Python.
I have found service layers and DI really helpful for writing functional programs. I have some complex image-processing scripts in Python that I can use as plug-ins with a distributed image processing service in Celery. Service layer and DI just takes code from:
```python
dependency.do_thing(params)
```
To:
```python
do_thing(dependency, params)
```
Which ends up being a lot more testable. I can run image processing tasks in a live deployment with all of their I/O mocked, or I can run real image processing tasks on a mocked version of Celery. This lets me test all my different functions end-to-end before I ever do a full deploy. Also using the Result type with service layer has helped me propagate relevant error information back to the web client without crashing the program, since the failure modes are all handled in their specific service layer function.
This was my takeaway too. It’s interesting to see the patterns. It would be helpful for some guidance upfront around when the situations in which they are most useful to implement. If a pattern is a tool, then steering me towards when it’s used or best avoided would be helpful. I do appreciate that the pros and cons sections get to this point, so perhaps it’s just ordering and emphasis.
That said, having built a small web app to enable a new business, and learning python along the way to get there, this provided me with some ideas for patterns I could implement to simplify things (but others I think I’ll avoid).
> That being said, I have a number issues with other parts of it, and I have seen how dangerous it can be when inexperienced developers take it as a gospel and try to implement everything at once (which is a common problem with any collection of design patterns like this.
Robert Martin is one of those examples, he did billions in damages by brainwashing inexperienced developers with his gaslighting garbage like "Clean Code".
Software engineering is not a hard science so there is almost never a silver bullet, everything is trade-offs, so people that claim to know the one true way are subcriminal psychopaths or noobs
Clean code has lots of useful tips and techniques.
When people are criticizing it they pick a concept from one or two pages out the hundreds and use it to dismiss the whole book. This is a worse mistake than introducing concepts that may be foot guns in some situations.
Becoming an experienced engineer is learning how, when and where to apply tools from your toolkit.
I’m a Typescript dev but this book is one of my favorite architecture books, I reference it all the time. My favorite pattern is the fake unit of work/service patterns for testing, I use this religiously in all my projects for faking (not mocking!!) third party services. It also helped me with dilemmas around naming, eg it recommends naming events in a very domain specific way rather than infrastructure or pattern specific way (eg CART_ITEM_BECAME_UNAVAILABLE is better than USER_NOTIFICATION). Some of these things are obvious but tedious to explain to teammates, so the fact that cosmic python is fully online makes it really easy to link to. Overall, a fantastic and formative resource for me!
That book is in a similar place in my heart, I barely used Python in my professional life, yet it's a book I often come back to even if I'm using a different language. It's also great that book is available both online and in paper form.
I grew tired from the forced OOP mindset, where you have to enforce encapsulation and inheritance on everything, where you only have private fields which are set through methods.
I grew tired of SOLID, clean coding, clean architecture, GoF patterns and Uncle Bob.
I grew tired of the Kingdom of Nouns and of FizzBuzz Enterprise Editions.
I now follow imperative or functional flows with least OOP as possible.
In the rare cases I use Python (not because I don't want to, but because I mainly use .NET at work) I want the experience to be free of objects and patterns.
I am not trying to say that this book doesn't have a value. It does. It's useful to learn some patterns. But don't try to fit everything in real life programming. Don't make everything about patterns, objects and SOLID.
my favourite model is to write as many pure functions as possible, and then as many functions of 1-4 parameters that interact with the outside world, and only then create domain objects to wrap those - it keeps the unrelated complexity out of the domain and then I can also reuse those functions without having to create the entire object that I don't always need.
I am not convinced that domain driven design works. Objects doesn't model the real world well. Why we should think DDD model the real world or a business well? And why do we even need to model something?
Computers are different than humans.
I think we should be pragmatic and come with the best solution in terms of money/time/complexity. Not trying to mimick human thought using computers.
After all a truck isn't mimicking horse and carriage. A plane isn't mimicking a bird.
- Reimplement SQLAlchemy models (we'll call it a "repository")
- Reimplement SQLAlchemy sessions ("unit of work")
- Add a "service layer" that doesn't even use the models -- we unroll all the model attributes into separate function parameters because that's less coupled somehow
- Scatter everything across a message bus to remove any hope of debugging it
- AND THIS IS JUST FOR WRITES!
- For reads, we have a separate fucking denormalized table that we query using raw SQL. (Seriously, see Chapter 12)
Hey, let's see how much traffic MADE.com serves. 500k total visits from desktop + mobile last month works out to... 12 views per MINUTE.
Gee, I wish my job was cushy enough that I could spend all day writing about "DDD" with my thumb up my ass.
I've made it through about 75% of the book and have never gotten the sense that they think everything discussed in the book is something you should always do. Each pattern discussed has a summary of pros and cons. While they may be a bit lacking, they clearly articulate the fact that you should be thinking whether or not the pattern matches the application's needs.
I don't think there's many applications that will require everything in the book but there are certaintly many applications that could apply one or more patterns discussed.
OK so show us how to write software for a complex business properly. Oh, I see, it's a throwaway account. This is just drive-by negativity with zero value.
I started writing python professionally a few years ago. Coming from Kotlin and TypeScript, I found the language approachable but I was struggling to build things in an idiomatic fashion that achieved the loose coupling and testability that I was used to. I bought this book after a colleague recommended it and read it cover to cover. It really helped me get my head around ways to manage complexity in non trivial Python codebases. I don’t follow every pattern it recommends, but it opened my eyes to what’s possible and how to apply my experience in other paradigms to Python without it becoming “Java guy does Python”.
Truly one of the great python programming books. The one thing that I found missing was the lack of static typing in the code, but that was a deliberate decision by the authors.
Haven’t read the book, so I don’t know exactly what position they’re taking there, but type checking has done more to improve my Python than any amount of architectural advice. How hard it is to type hint your code is a very good gauge of how hard it will be to understand it later.
My experience is that once people have static typing to lean on they focus much less on the things that in my view are more crucial to building clean, readable code: good, consistent naming and small chunks.
Just the visual clutter of adding type annotations can make the code flow less immediately clear and then due to broken windows syndrome people naturally care less and less about visual clarity.
Yup! I'm also hopeful that the upcoming type-checker from Astral will be an improvement over Mypy. I've found that Mypy's error messages are sometimes hard to reason about.
Some examples use dataclasses, which force type annotations.
Python does not support static typing. Tooling based on type annotations doesn't affect the compilation process (unless you use metaprogramming, like dataclasses do) and cannot force Python to reject the code; it only offers diagnostics.
I have this on my shelf. It's a small volume, similar to K&R, and like that book mine is showing visible signs of wear as I've thumbed through it a lot.
Readit News
Having said that, from a practical and experience standpoint, using some of these patterns can really spiral out into an increased complexity and performance issues in Python, specially when you use already opinionated frameworks like Django which already uses the ActiveRecord pattern.
I’ve been in companies big and small using Python, both using and ignoring architectural patterns. Turns out all the big ones with strict architectural (n=3) pattern usage, although “clean”, the code is waaaay to complex and unnecessarily slow in tasks that at first glance should had been simple.
Whereas the big companies that didn’t care for these although the code was REALLY ugly in some places (huge if-else files/functions, huge Django models with all business logic implemented in them), I was most productive because although the code was ugly I could read it, understand it, and modify the 1000 lines of if-else statements.
Maybe this says something about me more than the code but I hate to admit I was more productive in the non clean code companies. And don’t get me started on the huge amount of discussions they avoided on what’s clean or not.
Many files/functions/classes need to be updated to accomplish even simple tasks because somebody made a decision that you aren't allowed to do X or Y thing without creating N other things.
But in those companies that didn't care about architectural patterns its very likely that while there was more ugly code in certain places, it resulted in code with less indirection and more contained to a single area/unit or the task at hand making it easier for people to jump in and understand. I see so many people who create function after function in file after file to abstract away functionality when I'd honestly rather have a 100 line function or method that I can easily jump around and edit/debug vs many tiny functions all in separate areas.
Not to say having some abstractions are bad but the more I work in this field the more I realize the less abstractions there are, the easier it is to reason about singular units/features in code. I've basically landed on just abstract away the really hard stuff, but stop abstracting out things that simple.
The problem is this takes years of on-site experience to attain this level of domain understanding.
Domain modeling should not be about copying the existing model -- it should be about improving on it using all the advantages software has over the physical and social technologies the new software product is meant to replace. People are smart, and in most projects, there are key aspects of the existing domain model that are excellent abstractions that can and should be part of the new model. It's important to understand what stakeholders are trying to achieve with their current system before attempting to replace it.
But the models used in the business and cultural world are often messy, outdated and unoptimized for code. They rely on a human to interpret the edge cases and underspecified parts. We should treat that as inspiration, not the end goal.
Doctor Who fans will note that TARDIS craft seem to follow a different design: they regularly reconfigure themselves to fit their pilot, don't have controls laid out in any sensible fashion, and there's at least one reference to how they're "grown, not built". Then again they were also meant to be piloted by a crew and are most likely sentient, so it's also possible that due to the adaptations, the Doctor's TARDIS is just as eccentric as he is.
It's not like Doctor Who is "hard" sci-fi tho, it's basically Peter Pan in Space.
For example, if I am standing up a straight-forward calendar rest api, I am not going to have a complicated architecture. However, these kinds of patterns, especially an adherence to a ports and adapters architecture, has been critical for me in building trading systems that are easy to switch between simulation and production modes seamlessly. In those cases I am really sure I will need to easily unplug simulators with real trading engines, or historical event feeds with real-time feeds, and its necessary that the business logic have not dual implementations to keep in sync.
The problem with "strict architectural pattern usage" is that people think that a specific implementation, as listed in the reference, is "the pattern".
"The pattern" is the thought process behind what you're doing, and the plan for working with it, and the highest-level design of the API you want to offer to the rest of the code.
A state machine in Python, thanks to functions being objects, can often just be a group of functions that return each other, and an iteration of "f = f(x)". Sometimes people suggest using a Borg pattern in Python rather than a Singleton, but often what you really want is to just use the module. `sys` is making it a singleton for you already. "Dependency injection" is often just a fancy term for passing an argument (possibly another function) to a function. A Flyweight isn't a thing; it's just the technique of interning. The Command pattern described in TFA was half the point of Jack Diederich's famous rant (https://www.youtube.com/watch?v=o9pEzgHorH0); `functools.partial` is your friend.
> Maybe this says something about me more than the code but I hate to admit I was more productive in the non clean code companies.
I think you've come to draw a false dichotomy because you just haven't seen anything better. Short functions don't require complex class hierarchies to exist. They don't require classes to exist at all.
Object-oriented programming is about objects, not classes. If it were about classes, it would be called class-oriented programming.
Finding my way around a soup of ultra abstracted Matryoshka ravioli is my least favourite part of programming. Instead of simplifying things, now I need to consult 12 different objects spread over as many files before I can create a FactoryFactory.
Here's an example of how things can go off the rails very quickly: Rule 1: Functions should be short (no longer than 50 lines). Rule 2: Public functions should be implemented with an interface (so they can be mocked).
Now as a developer who wants to follow the logic of the program, you have to constantly "go to definition" on function calls on interfaces, then "go to implementation" to find the behavior. This breaks your train of thought / flow state very quickly.
Now let's amp it up to another level of suck: replace the interface with a microservice API (gRPC). Now you have to tab between multiple completely different repos to follow the logic of the program. And when opening a new repo, which has its own architectural layers, you have to browse around just to find the implementation of the function you're looking for.
These aren't strawmen either... I've seen these patterns in place at multiple companies, and at this point I yearn for a 1000 line function with all of the behavior in 1 place.
My last job had a Python codebase just like this. Lots of patterns, implemented by people who wanted to do things "right," and it was a big slow mess. You can't get away with nearly as much in Python (pre-JIT, anyway) as you can in a natively compiled language or a JVM language. Every layer of indirection gets executed in the interpreter every single time.
What bothers me about this book and other books that are prescriptive about application architecture is that it pushes people towards baking in all the complexity right at the start, regardless of requirements, instead of adding complexity in response to real demands. You end up implementing both the complexity you need now and the complexity you don't need. You implement the complexity you'll need in two years if the product grows, and you place that complexity on the backs of the small team you have now, at the cost of functionality you need to make the product successful.
To me, that's architectural malpractice. Even worse, it affects how the programmers on your team think. They start thinking that it's always a good idea to make code more abstract. Your code gets bloated with ghosts of dreamed-of future functionality, layers that could hypothetically support future needs if those needs emerged. A culture of "more is better" can really take off with junior programmers who are eager to do good work, and they start implementing general frameworks on top of everything they do, making the codebase progressively more complex and harder to work in. And when a need they anticipated emerges in reality, the code they wrote to prepare for it usually turns out to be a liability.
Looking back on the large codebases I've worked with, they all have had areas where demands were simple and very little complexity was needed. The ones where the developers accepted their good luck and left those parts of the codebase simple were the ones that were relatively trouble-free and could evolve to meet new demands. The ones where the developers did things "right" and made every part of the codebase equally complex were overengineered messes that struggled under their own weight.
My preferred definition of architecture is the subset of design decisions that will be costly to change in the future. It follows that a goal of good design is minimizing architecture, avoiding choices that are costly to walk back. In software, the decision to ignore a problem you don't have is very rarely an expensive decision to undo. When a problem arises, it is almost always cheaper and easier to start from scratch than to adapt a solution that was created when the problem existed only in your head. The rare exceptions to this are extremely important, and from the point of view of optics, it always looks smarter and more responsible to have solved a problem incorrectly than not to have solved it at all, but we shouldn't make the mistake of identifying our worth and responsibility solely with those exceptions.
The trouble is if you strictly wait until it's time then basically everything requires some level of refactoring before you can implement it.
The dream is that new features is just new code, rather than refactoring and modifying existing code. Many people are already used to this idea. If you add a new "view" in a web app, you don't have to touch any other view, nor do you have to touch the URL routing logic. I just think more people are comfortable depending on frameworks for this kind of stuff rather than implementing it themselves.
The trouble is a framework can't know about your business. If you need pluggable validation layers or something you might have to implement it yourself.
The downside, of course, is we're not always great at seeing ahead of time where the application will need to be flexible and grow. So you could build this into everything, leading to unnecessarily complicated code, or nothing, leading to constant refactors which will get worse and worse as the codebase grows.
Your approach can work if developers actually spot what's happening early and actually do what's necessary when it actually is. Unfortunately in my experience people follow by example and the frog can boil for a long time before people start to realise that their time is spent mostly doing large refactors because the code just doesn't support the kind of flexibility and extensibility they need.
Patterns and Abstractions have a HUGE cost in python. They can be zero cost in C++ due to compiler, or very low cost due to JVM JIT, but in Python the cost is very significant, especially once you start adding I/O ops or network calls
That being said, I have a number issues with other parts of it, and I have seen how dangerous it can be when inexperienced developers take it as a gospel and try to implement everything at once (which is a common problem with any collection of design patterns like this.
For example, repository is a helpful pattern in general; but in many cases, including the examples in the book itself, it is a huge overkill that adds complexity with very little benefit. Even more so as they're using SQLAlchemy, which is a "repository" in its own right (or, more precisely, a relational database abstraction layer with an ORM added on top).
Similarly, service layers and unit of work are useful when you have complex applications that cover multiple complex use cases; but in a system consisting of small services with narrow responsibilities they quickly become overly bloated using this pattern. And don't even get me started with dependency injection in Python.
The essential thing about design patterns is that they're tools like any other, and the developers should understand when to use them, and even more importantly when not to use them. This book has some advice in that direction, but in my opinion it should be more prominent and placed upfront rather at the end of each chapter.
In the end, it's just making sure that all database access for a specific entity all goes through one point (the repository for that entity). Inside the repository, you can do whatever you want (run queries yourself, use ORM, etc).
A lot of the stuff written in the article under the section Repository pattern has very little to do with the pattern, and much more to do with all sorts of Python, Django, and SQLAlchemy details.
That's aside from their particular example of SQLAlchemy sessions, which is extra weird because a Session is already a repository, more or less.
I mean, sure, there's a difference between your repository for your things and types you might consider foreign, in theory, but how theoretical are we going to get? For what actual gain? How big of an app are we talking?
You could alias Repository = Session, or define a simple protocol with stubs for some of Session's methods, just for typing, and you'd get the same amount of theoretical decoupling with no extra layer. If you want to test without a database, don't bind your models to a session. If you want to use a session anyway but still not touch the database, replace your Session's scopefunc and your tested code will never know the difference.
It's not a convincing example.
Building your repository layer over theirs, admittedly you stop the Query type from leaking out. But then you implement essentially the Query interface in little bits for use in different layers, just probably worse, and lacking twenty years of testing.
Repository patterns are fine for CRUD but don't really stretch to those endpoints where you really need the query with the two CTEs and the four joins onto a query selecting from another query based on the output of a window function.
I had a former boss who strongly pushed my team to use the repository pattern for a microservice. The team wanted to try it out since it was new to us and, like the other commenters are saying, it worked but we never actually needed it. So it just sat there as another layer of abstraction, more code, more tests, and nothing benefited from it.
Anecdotally, the project was stopped after nine months because it took too long. The decision to use the repository pattern wasn't the straw that broke the camel's back, but I think using patterns that were more complicated than the usecase required was at the heart of it.
Could I get you started? Or could you point me to a place to get myself started? I primarily code in Python and I've found dependency injection, by which I mean giving a function all the inputs it needs to calculate via parameters, is a principle worth designing projects around.
Deleted Comment
This book explicitly tells you not to do this.
> Similarly, service layers and unit of work are useful when you have complex applications that cover multiple complex use cases; but in a system consisting of small services with narrow responsibilities they quickly become overly bloated using this pattern. And don't even get me started with dependency injection in Python.
I have found service layers and DI really helpful for writing functional programs. I have some complex image-processing scripts in Python that I can use as plug-ins with a distributed image processing service in Celery. Service layer and DI just takes code from:
```python
dependency.do_thing(params)
```
To:
```python
do_thing(dependency, params)
```
Which ends up being a lot more testable. I can run image processing tasks in a live deployment with all of their I/O mocked, or I can run real image processing tasks on a mocked version of Celery. This lets me test all my different functions end-to-end before I ever do a full deploy. Also using the Result type with service layer has helped me propagate relevant error information back to the web client without crashing the program, since the failure modes are all handled in their specific service layer function.
the two main methods I've seen are to run tasks eagerly, or test the underlying function and avoid test Celery .delay/etc at all
That said, having built a small web app to enable a new business, and learning python along the way to get there, this provided me with some ideas for patterns I could implement to simplify things (but others I think I’ll avoid).
Robert Martin is one of those examples, he did billions in damages by brainwashing inexperienced developers with his gaslighting garbage like "Clean Code".
Software engineering is not a hard science so there is almost never a silver bullet, everything is trade-offs, so people that claim to know the one true way are subcriminal psychopaths or noobs
When people are criticizing it they pick a concept from one or two pages out the hundreds and use it to dismiss the whole book. This is a worse mistake than introducing concepts that may be foot guns in some situations.
Becoming an experienced engineer is learning how, when and where to apply tools from your toolkit.
https://www.obeythetestinggoat.com/pages/book.html
That book is in a similar place in my heart, I barely used Python in my professional life, yet it's a book I often come back to even if I'm using a different language. It's also great that book is available both online and in paper form.
I'll definitely give this book a chance!
You might like this: https://martinfowler.com/bliki/TestDouble.html
I grew tired from the forced OOP mindset, where you have to enforce encapsulation and inheritance on everything, where you only have private fields which are set through methods.
I grew tired of SOLID, clean coding, clean architecture, GoF patterns and Uncle Bob.
I grew tired of the Kingdom of Nouns and of FizzBuzz Enterprise Editions.
I now follow imperative or functional flows with least OOP as possible.
In the rare cases I use Python (not because I don't want to, but because I mainly use .NET at work) I want the experience to be free of objects and patterns.
I am not trying to say that this book doesn't have a value. It does. It's useful to learn some patterns. But don't try to fit everything in real life programming. Don't make everything about patterns, objects and SOLID.
Computers are different than humans.
I think we should be pragmatic and come with the best solution in terms of money/time/complexity. Not trying to mimick human thought using computers.
After all a truck isn't mimicking horse and carriage. A plane isn't mimicking a bird.
- Reimplement SQLAlchemy models (we'll call it a "repository")
- Reimplement SQLAlchemy sessions ("unit of work")
- Add a "service layer" that doesn't even use the models -- we unroll all the model attributes into separate function parameters because that's less coupled somehow
- Scatter everything across a message bus to remove any hope of debugging it
- AND THIS IS JUST FOR WRITES!
- For reads, we have a separate fucking denormalized table that we query using raw SQL. (Seriously, see Chapter 12)
Hey, let's see how much traffic MADE.com serves. 500k total visits from desktop + mobile last month works out to... 12 views per MINUTE.
Gee, I wish my job was cushy enough that I could spend all day writing about "DDD" with my thumb up my ass.
I don't think there's many applications that will require everything in the book but there are certaintly many applications that could apply one or more patterns discussed.
I cannot recommend it enough. Worth every penny.
Just the visual clutter of adding type annotations can make the code flow less immediately clear and then due to broken windows syndrome people naturally care less and less about visual clarity.
[0]: https://x.com/charliermarsh/status/1884651482009477368
It has type hints, such as here: https://www.cosmicpython.com/book/chapter_08_events_and_mess...
Do you mean it's not strict enough? There are some parts of the book without them.
Python does not support static typing. Tooling based on type annotations doesn't affect the compilation process (unless you use metaprogramming, like dataclasses do) and cannot force Python to reject the code; it only offers diagnostics.