but no metaprogramming, no decorators/ directory, no new concepts, only standard plain Rails. As a bonus it's even an instance of functional programming.
Well, until you need different `colour_coded_availability` methods for different contexts. Helpers get a lot done, but there are absolutely use cases for a separate presentation layer.
Doesn't this thrash the ruby VM's virtual method cache? (It's been ~7 years since I was working on production ruby code, but I remember it being a really painful performance issue for us when using dynamic method missing routing. I might be getting it confused with class extensions though.)
Not certain what you mean by "virtual method cache", the Ruby VM has multiple layers of method cache (or call cache) but I've never heard any of them referred to as the "virtual method cache".
For the inline call caches in the interpreter loop, they are monomorphic, so if you call the same codepath with the decorator and the actual object, they will indeed flip flop.
The second layer of cache is class based, so after the inline cache is defeated you will end up doing a single hash-table lookup on the class.
As for YJIT, IIRC it does handle polymorphic call caches, so it won't mind such situation at all unless you have more than a handful of different implementations of that method being called at a given callsite.
What annoys me (only slightly) in the Ruby community, is it's loose and often "wrong" usage of design patterns.
IMHO, the whole idea of "design patterns" is that they are a standardized way of doing stuff. So that when someone, regardless of language, says "use an abstract factory" we know what it does, how it can be used, what it doesn't do, it's pitfalls etc.
What ruby calls "decorators" aren't "decorators". That's for this DIY implementation, but even more so for "Draper".
It's not only that an actual Decorator should not introduce new methods, but also how it's set-up and used. And therefore, what it's pitfalls and downsides are.
In the case of Draper or this DIY: the views now rely on a concrete and specific subclass - losing both LSP and DIP. Worse: the layer using it -views- aren't the one setting it up or controlling it - the controllers. So it introduces very hard dependencies between different domains - the views/partials/serializers now depend on whether or not you've remembered to actually decorate the model. Something that in my experience will go wrong and will crash in production - or if you're lucky and have lots of E2E tests, in CI.
The same "imprecise use of design patterns" happen with MVC, Models, ActiveRecord, "interfaces" and so forth. Often because of limitations of Ruby, but rather often because the Ruby community just uses the term "wrong".
Ruby doesn't have "interfaces" - obviously, because it doesn't have types or a type-enforcement.
So any design pattern, architecture or concept that relies on types, or interfaces, are "limited" in that sense. Ports, Adapters, Strategy, for example "
require" interfaces in their definition. Their benefits rely on interfaces, so if a language lacks this, you really only get the downsides. Factory, Observer, Decorators, etc mention them, and use them, but can be implemented without them.
Maybe "limitations" isn't the best word, because e.g. "an interface" is a deliberate limitation, imposed and designed by the developer.
> IMHO, the whole idea of "design patterns" is that they are a standardized way of doing stuff
To disagree - Design patterns are a communication language. You use them to talk about code; if you use them as strict recipes they become dogma, with all the problems that entails.
Beyond that - most of the canonical design patterns are coping mechanisms for the limitations of pre-modern Java / strict OOP. You just plain don’t need them in a language like Ruby - IMO, mostly because of ‘yeild’.
The “imprecise use” is a consequence of either trying to use an unnecessary design pattern, or trying too dogmatically to adhere to one.
But that means when we both say "decorator pattern", we must mean the same. The thing that in the Ruby community is a "Decorator" isn't a "decorator pattern".
It's not following the details, but also not even meant for the use-cases or to solve limitations of ruby or strict OOP. It's simply not a "decorator" that we both think of (or would google, or read in e.g. GoF). It's something entirely different. In ruby/Rails (draper gem, most notably), they happen to use the word "Decorator" for a system that makes "view helpers" OOP rather than the normal rails way of a large bucket of random global helper-functions.
I've had this conversation that shows the problem:
- Me: If we'd use the decorator pattern, we could easily tame that large clumsy tree of subclasses in our authorization logic and make it much easier testable and understandable.
- Wat? Decorators? But our authn has nothing to do with views! Why would we want to pull views-logic into our authn layer?
So, by using the term wrong, at least N=1 shows how the "communication language" of Design PAtterns fails.
Couldn’t get this idea out of my mind so I went back to the GoF text:
> Decorator subclasses are free to add operations for specific functionality. For example, ScrollDecorator’s ScrollTo operation lets other objects scroll the interface if they know there happens to be a ScrollDecorator object in the interface.
> ...The important aspect of this pattern is that it lets decorators appear anywhere a VisualComponent can. That way clients generally can't tell the difference between a decorated component and an undecorated one, and so they don't depend at all on the decoration.
and further on:
> The decorator conforms to the interface of the component it decorates so that its presence is transparent to the component's clients.
> It's not only that an actual Decorator should not introduce new methods
Maybe I'm misreading you, but I've never seen a definition of the Decorator pattern that doesn't involve new methods on the Decorator. If that's what you mean, can you say more? If not, what do you mean?
> Note that decorators and the original class object share a common set of features. In the previous diagram, the operation() method was available in both the decorated and undecorated versions.
and
>
GoF Book, "Design Patterns: Elements of Reusable Object-Oriented Software", p.166:
> "A decorator object conforms to the interface of the component it decorates so that its presence is transparent to clients."
OH WAIT I SEE IT I'M THINKING OF THE FACADE PATTERN. Yes, the Ruby community does tend to conflate Decorator, Facade, and Presenter in very confusing ways.
Might've been wise at the time, but nowadays I think the recommendation is to use View Components to package up view logic.
(I'm personally so-so on VCs; I think the core idea is pretty good but I'm not super sold on some of the implementation details - too OOP, not enough `yield`)
I started writing a comment wanting to defend it, because the expressions themselves are not that unreadable. <=> is a comparator (and used in many languages), clamp sets min and max sizes (and that can be guessed from the english meaning of the word), and 0.. is a range that goes from 0 to infinity (admittedly not that common an expression, but completely logical and intuitive way to express a range once you understood once that it is about ranges).
But then I realized that's nonsensical, and not what the code is supposed to do given the usage in the template. I assume something got mangled there when changing the code for the blog post.
Or I'm just understanding the ruby code wrong despite checking just now in irb, in that case that's a point for the intention of your comment, and a vote for "worse".
Readit News
For the inline call caches in the interpreter loop, they are monomorphic, so if you call the same codepath with the decorator and the actual object, they will indeed flip flop.
The second layer of cache is class based, so after the inline cache is defeated you will end up doing a single hash-table lookup on the class.
As for YJIT, IIRC it does handle polymorphic call caches, so it won't mind such situation at all unless you have more than a handful of different implementations of that method being called at a given callsite.
IMHO, the whole idea of "design patterns" is that they are a standardized way of doing stuff. So that when someone, regardless of language, says "use an abstract factory" we know what it does, how it can be used, what it doesn't do, it's pitfalls etc.
What ruby calls "decorators" aren't "decorators". That's for this DIY implementation, but even more so for "Draper".
It's not only that an actual Decorator should not introduce new methods, but also how it's set-up and used. And therefore, what it's pitfalls and downsides are.
In the case of Draper or this DIY: the views now rely on a concrete and specific subclass - losing both LSP and DIP. Worse: the layer using it -views- aren't the one setting it up or controlling it - the controllers. So it introduces very hard dependencies between different domains - the views/partials/serializers now depend on whether or not you've remembered to actually decorate the model. Something that in my experience will go wrong and will crash in production - or if you're lucky and have lots of E2E tests, in CI.
The same "imprecise use of design patterns" happen with MVC, Models, ActiveRecord, "interfaces" and so forth. Often because of limitations of Ruby, but rather often because the Ruby community just uses the term "wrong".
wat.
If anything I think Rubys lack of limitations could be the issue. There's a thousand different ways to do anything, and people do.
So any design pattern, architecture or concept that relies on types, or interfaces, are "limited" in that sense. Ports, Adapters, Strategy, for example " require" interfaces in their definition. Their benefits rely on interfaces, so if a language lacks this, you really only get the downsides. Factory, Observer, Decorators, etc mention them, and use them, but can be implemented without them.
Maybe "limitations" isn't the best word, because e.g. "an interface" is a deliberate limitation, imposed and designed by the developer.
Deleted Comment
To disagree - Design patterns are a communication language. You use them to talk about code; if you use them as strict recipes they become dogma, with all the problems that entails.
Beyond that - most of the canonical design patterns are coping mechanisms for the limitations of pre-modern Java / strict OOP. You just plain don’t need them in a language like Ruby - IMO, mostly because of ‘yeild’.
The “imprecise use” is a consequence of either trying to use an unnecessary design pattern, or trying too dogmatically to adhere to one.
I very much agree.
But that means when we both say "decorator pattern", we must mean the same. The thing that in the Ruby community is a "Decorator" isn't a "decorator pattern".
It's not following the details, but also not even meant for the use-cases or to solve limitations of ruby or strict OOP. It's simply not a "decorator" that we both think of (or would google, or read in e.g. GoF). It's something entirely different. In ruby/Rails (draper gem, most notably), they happen to use the word "Decorator" for a system that makes "view helpers" OOP rather than the normal rails way of a large bucket of random global helper-functions.
I've had this conversation that shows the problem:
- Me: If we'd use the decorator pattern, we could easily tame that large clumsy tree of subclasses in our authorization logic and make it much easier testable and understandable.
- Wat? Decorators? But our authn has nothing to do with views! Why would we want to pull views-logic into our authn layer?
So, by using the term wrong, at least N=1 shows how the "communication language" of Design PAtterns fails.
> Decorator subclasses are free to add operations for specific functionality. For example, ScrollDecorator’s ScrollTo operation lets other objects scroll the interface if they know there happens to be a ScrollDecorator object in the interface.
> ...The important aspect of this pattern is that it lets decorators appear anywhere a VisualComponent can. That way clients generally can't tell the difference between a decorated component and an undecorated one, and so they don't depend at all on the decoration.
and further on:
> The decorator conforms to the interface of the component it decorates so that its presence is transparent to the component's clients.
Maybe I'm misreading you, but I've never seen a definition of the Decorator pattern that doesn't involve new methods on the Decorator. If that's what you mean, can you say more? If not, what do you mean?
> Note that decorators and the original class object share a common set of features. In the previous diagram, the operation() method was available in both the decorated and undecorated versions.
and
>
GoF Book, "Design Patterns: Elements of Reusable Object-Oriented Software", p.166:
> "A decorator object conforms to the interface of the component it decorates so that its presence is transparent to clients."
I found Decorators just obscured display logic away in an inconvenient way.
(I'm personally so-so on VCs; I think the core idea is pretty good but I'm not super sold on some of the implementation details - too OOP, not enough `yield`)
1. spend 30 minutes beating around the bush, reinventing the wheel etc for your own education and interest; then
2. throw that away and use SimpleDelegator from the standard library
and to be clear, I don’t mean this negatively at all.
But then I realized that's nonsensical, and not what the code is supposed to do given the usage in the template. I assume something got mangled there when changing the code for the blog post.
Or I'm just understanding the ruby code wrong despite checking just now in irb, in that case that's a point for the intention of your comment, and a vote for "worse".
Deleted Comment