[dhh]

The proof is always in the pudding. While there are good and reasonable times to introduce "interactors", this particular example is poor. The tests presented are anemic, and the code is absolutely not any clearer by being extracted and wrapped. I would indeed have kept all this in the controller.

The key point for an "interactor" extraction is imo when you have multiple models being created in symphony, like a Signup model. Or if you for some reason need to reuse the behavior.

But if all your controllers look like this, with one "interactor" model per action, you're doing it wrong.

Whatever floats your boat, though. If this is what you prefer, great. But please hold the "beginner's version" crap. Plenty of large apps are built with vanilla Rails. Basecamp is one.

[dhh]

I rewrote the code in this example to use the "Beginner's Version" of Rails (sigh). You judge which you like better: https://gist.github.com/dhh/9333694

[dhh]

Here's another version that doesn't even use private methods in the controller and uses a PORO for the email grouping: https://gist.github.com/dhh/9333991

[rubiquity]

I think a lot of blog posts use trivial and contrived examples for the point of explaining the concept. There's not a lot of people out there that are good at explaining complicated refactorings. Katrina Owen comes to mind as one of the few that is good at explaining such refactorings.

Perhaps if someone made a post like this with starting code that was more complex it would be a better example, and also harder for you to counter it with a couple gists ;)

[jlangenauer]

Isn't ConfirmedGrouperEmails here just an Interactor / Service Object anyway, but one that is derived from ActionMailer?

To my mind, this code is much clearer, which proves the OPs point.

[danso]

Thanks for writing this. While I often look for ways to take external-interaction-responsibility away from Rails objects, the OP didn't really make sense to me, at least in terms of saving time and making things more logical.

However, there's a distinction that has to be made in your example and the OP's. In the OP, the failure of the Interactor, including the delivery of emails, would cause the controller to enter the "failure" branch:

    if interactor.success?
      redirect_to home_path
    else
      flash[:error] = interactor.message
      render :new
    end

Whereas your example, the controller would take the successful branch if the data model was saved, regardless of whether email delivery failed:

    if @grouper.save
      ConfirmedGrouperEmails.new(@grouper).deliver
      AssignBarForGrouper.enqueue(@grouper.id)
 
      redirect_to home_path
    else
      render :new
    end

So we're not comparing apples to apples here. However, as a layperson, I'd have to agree with you: Why should the error be raised to the grouper-creating user, when it should be going to the part of the system that handles mailing? But maybe the actual details are more complicated than that...

[dhh]

Why would the delivery of the emails fail? Because your SMTP server is down? That's an exceptional state, handle it with exceptions -- not with conditions.

Or maybe because the email addresses are invalid? Handle that when they are captured. It's way too late for that here.

[Axsuul]

The failure of delivery of emails should be handled gracefully such as within a message queue that gets retried until it succeeds or dies.

[iamwil]

I can't seem to find it on Google. What's a PORO?

[cheshire137]

Plain Old Ruby Object, as opposed to an ActiveRecord model.

[dragonwriter]

Never seen the term used before, but by analogy to POJO and POCO it would have to be a "Plain Old Ruby Object".

[the_fury]

Plain Old Ruby Object

[tomblomfield]

I've spent the last few months re-writing a medium-sized code-base (several hundred thousand LoC) that looks like your version into code that looks like the blog version. Test suite run time has dramatically decreased, code is more loosely coupled and we see far fewer bugs.

> You obviously need to choose the patterns that fit the problem you’re trying to solve – it’s rarely one-size-fits-all, and some of these principles may be overkill in a very simple 15-minute blog application. Our objection is that Rails feels like it’s a Beginners’ Version, often suggesting that one size does fit all

This quote from the post (minus the "Beginners" jibe - sorry!) is my core objection. Your example works fine if you're working with a simple application. But the "Rails Way" feels like it starts to fall over when you get to medium or large codebases.

I feel like Rails could benefit hugely from a few core "Advanced" features that help you grow a small application into a larger, functioning business. Sure, you can add them yourself, but then you lose the advantage of shared standards and strong architectural conventions. Every new developer we hire has to learn what exactly we mean by "Interactors" or "Decorators" or "Policies".

[dhh]

Yes, rewriting a system after it's already settled and designed can indeed bring a cleaner code base about. But don't confuse that with "better architecture".

Don't even get me started on the hand-wavy "loosely coupled".

If this is a poor example, pick a good example. I'll be happy to code ping pong you whatever example you choose.

One way to spend hundreds of thousands of LOC on an application is to stuff it with needless abstractions. That doesn't make it "advanced", and it's not Rails that's falling over, it's probably just some shitty code.

Maybe that's a clue that your chosen paradigms weren't that helpful. Or rather, that they convoluted the code base instead of clarified it. It's certainly a red herring that you need to teach your abstractions to new developers and that it's an endeavor to do so.

POROs are great, though. Our app/models is full of them. We even added app/services too. The trouble I have is with people who, like you, fall in love with the flawed notion that their application is such a special snowflake that it deserves "advanced techniques". Bullshit. There are good techniques and bad techniques. Size of the application is rarely a factor, except in ballooning the ego of the programmer.

Anyway, the invitation stands. Present any piece of real code and we can ping pong on it. I enjoy talking about specifics and improving real code. I detest "oh that was just a poor example, but the general principle is..." kind of debates. If you can't produce a good example, you don't have a general principle.

[nthj]

> POROs are great, though. Our app/models is full of them. We even added app/services too. The trouble I have is with people who, like you, fall in love with the flawed notion that their application is such a special snowflake that it deserves "advanced techniques".

The main difference I see between the Rails philosophy and most of these "Use SRP and Interactors and things!" blog posts are that Rails is way more interested in using the correct tool for the job, and most of these blog posts are of the "use this for everything and every one!" variety.

I have one project right now that I decided I wanted a Command for one action. I'm accepting input from a webhook that I can perform asynchronously, it interacts with like 5 different models, and I wanted to beat up on the tests pretty thoroughly. So I did that.

Elsewhere I'm using concerns to make like 4 different models easily sortable.

And other places I'm using Module#concerning to keep models organized.

ONE TRUE WAY is overrated. I'd rather use whatever works given the context.

[gregmolnar]

"ONE TRUE WAY is overrated. I'd rather use whatever works given the context." That's it.

[andycroll]

This. This. This.

[AndyDavis3416]

I hope I am not sticking my head into the lion's mouth here, but here (https://gist.github.com/BiggerNoise/9334673) is a link to a gist of some code in one of our apps.

Cases have Tasks and Activities. Activities are always associated with a case, and may be associated with a task.

In the controller, when assigning a case, one would type: AssignCaseCommand.new(params).execute!

The command class implements the Command pattern (the pretty well established name for interactors) and is a Composite as well.

BTW - the same command that assigns the tasks of a case is also used to merely assign a task. And, if we're just recording an activity, we would just use the create activity command.

I think it would be insanity to try and capture the rules associated with re-assigning tasks in the controller. (I am pretty sure that you would agree.)

--Minor Edit for clarity--

[tinco]

I think it's great that you stick your head into the lion's mouth, even if he chewed it clean off it's a great contribution :)

I think the main problem with your command pattern approach is that you didn't implement the full command pattern. At least, I assume you didn't, you didn't actually show the controller. The command pattern has one extra class, the invoker. The invoker takes commands, and invokes them. A great example of this is the worker queue, it takes generic worker objects that have an 'execute' method.

It is never the class that creates the worker objects, that executes them. Yet that seems to be the way you're going to implement the controllers that use these commands.

What I think you've built instead is an adapter pattern. You have abstracted a class of problems in a way that they have a common interface, so that in your controllers you have a uniform way of invoking them.

This is where I think dhh sees the muddiness in your code. Why are you abstracting your interaction from the controller, which is supposed to be the exact thing that should be controlling the interaction.

The only reason for that would be that the same interaction is happening in different controllers in exactly the same way. And that's something that dhh claims does not happen in real applications, or at least he demands an example.

[AndyDavis3416]

Honestly, I was disappointed in his response. With all his harrumphing about real examples, I would have expected him to make an effort to understand a real example. But, as I said elsewhere, the downside of real examples is that they require quite a bit more effort to grasp than contrived ones.

To address your comments:

While it is true that one of the great features (and frequent motivations) for using the Command pattern is the separation of the creation and invocation, it is not the only motivation. The two big payoffs that we are getting are:

- Separation of invocation from the knowledge of how it is carried out

- Composition of complex commands from smaller, simpler commands

There's no requirement that the invoker is never the same as the creator. It's just that most often it isn't, particularly (as you noted) in a worker queue system. That's really just a feature of the implementation that the command readily enables rather than a fundamental property of the pattern. For us, sometimes it is, sometimes it isn't; that's not the primary payoff that we're going for.

I can't support the characterization of this as an Adapter. This is fundamentally behavioral. I have a class that represents an action that I want to have happen. All of the places that want this action do not need to know what is involved, they only need to create the command and hand it off/invoke it.

FWIW - I have five different places in my software that create an AssignCaseCommand. - One controller

- Three commands that can create this as a sub command

- A utility job

Plus that simple concept is available to any custom implementation code that we write.

[dhh]

I asked for specific code, you gave me specific code, so here's how I would have structured that without the (imo, needless) command pattern: https://gist.github.com/dhh/9348053

It includes a lot of more context than your example, so I was guessing liberally at the missing pieces. But hopefully this is still easily digestible.

[dhh]

If reassigning a case is part of the domain, it belongs in the model.

I don't know how to put this gently, but this concoction looks like a hot mess with no separation between modeling and controlling.

These commands have a ton of domain logic all tangled up with activity reporting, formatting, and parameter parsing. It looks like a big ball of mud to me :(

[gnaritas]

> It looks like a big ball of mud to me :(

That is of course the most popular of all architectural patterns. :)

[AndyDavis3416]

Perhaps that's why the OP used a simple example. Complex examples are pretty hard to get your head around at first glance; even sane systems can be perceived as a 'hot mess' upon a quick read without any context.

I disagree that re-assigning a case belongs in the case model. It is a complex interaction between several models. Extracting it to a command object allows all of the rules to be easily understood.

[tomblomfield]

My contention is that there often are times when the complexity of the business requires a fairly extensive set of operations to occur in concert, including the creation of several models, email-sending etc.

In this case, the Right Place (tm) to put these is in POROs / Interactors that can be tested in isolation and re-used across the code-base if necessary. This code should not live in controllers and certainly not in ActiveRecord models.

It sounds like you agree with that, since you're using them in your own codebase?

In that case, is there not an advantage to standardising the API for these POROs in Rails? Simple methods like fail!, success?, failure? and message

Another way of asking the question - why do so many Rails developers fail to realise that they can use these concepts? Why are so many Rails codebases packed full of enormous ActiveRecord models that are crippled with numerous before_save callbacks?

Sure, shitty coders write shitty code. But can't you lend them a hand?

[cgag]

I don't get why POROs would need some special API. I also don't get why you're pulling code out into things like ConfirmGrouper that still depend on the DB. Why not a true PORO and something like ConfirmGrouper.for(grouper)? Why not just write methods that return bools?

Though I don't know if I agree that the actual sending of emails belongs in a PORO either and not a controller. It seems like determining whether or not emails need to be sent is the kind of logic I would put in a PORO and do the actual email sending from a controller method as in DHH's example.

Edit: Just saw DHH's second example, that's essentially how I would write it.

[ajsharp]

> In this case, the Right Place (tm) to put these is in POROs / Interactors that can be tested in isolation and re-used across the code-base if necessary.

Key phrase: "if necessary".

It sounds like you're advocating for building abstractions before you actually have a reason to use them. Your reason seems to be "we may need to reuse them in the future" or, "one can see where we might move in a direction where we'd want want to re-use this code", or worse yet, "we're definitely going to build some functionality after this release when we'll want to re-use this stuff."

I've been there, really, I have. Build the abstraction when you actually need it, not when you think you're going to need it, or pretty close to needing it, etc. Because 98% of the time, YAGNI.

> In that case, is there not an advantage to standardising the API for these POROs in Rails?

Not a big enough that I've come across to warrant building domain-specific abstractions and special rules that, as @dhh mentioned, will serve only to raise the barrier to entry to your codebase.

[tomblomfield]

> It sounds like you're advocating for building abstractions before you actually have a reason to use them.

If it sounds that way, I've over-simplified the examples in the blog post in the interests of clarity. I'll take more care in future! The specific interactor in question is used in 4 separate places.

From the post:

> You obviously need to choose the patterns that fit the problem you’re trying to solve – it’s rarely one-size-fits-all, and some of these principles may be overkill in a very simple 15-minute blog application.

[al2o3cr]

"Sure, shitty coders write shitty code. But can't you lend them a hand?"

It's not clear to me that adding more APIs and conventions that said shitty devs won't read about, understand, or even be aware of is actually useful.

If shitty devs are your problem, don't fix the dev tools - FIRE the tools pretending to be devs...

[dablweb]

There are a lot of brilliant developers giving talks at every RubyConf advocating these very techniques...

[djur]

The standard API for POROs is Object (and Class, etc.). Standardizing a new API necessarily makes them no longer "plain".

[halostatue]

The current app I work on has a lot of warts, but only five `before_save` callbacks (there's maybe five times that many total callbacks, although there are lots of validations).

If you really want to standardize the interface, use the tools that already exist: ActiveModel and validations. Your “Interactors” are models; they just happen to be coordinating models that affect multiple ActiveRecord models simultaneously. (I did this for an account plan change system that affected billing and device assignment to a plan. It could easily affect a dozen or more records, all in a single transaction, but was itself easily testable.)

[Killswitch]

> "I detest "oh that was just a poor example, but the general principle is..." kind of debates. If you can't produce a good example, you don't have a general principle." - DHH, March 3rd 2014

That's going in my quote book. Thank you DHH.

[dablweb]

I think that labelling the use of POROs as an "advanced technique" is simply wrong. How could it be? In fact it's simpler than implementing an AR callback (with more complexity behind the scenes than anything!)

You are using app/services and POROs in your own app? So are we. But I guarantee they don't look like yours because there is no endorsed convention.

Why not? Because Rails is still clinging to such an outdated idea as MVC? Pure MVC at a large scale is bullshit.

"Size of the application is rarely a factor". I don't agree. Size of the application is always a factor. A big factor.

What I'd like to see is the community embrace conventions outside the fabled "MVC" paradise. Rails is "convention over configuration", and is content to hold your hand for your first thousand lines of code... but then what?

[jlebrech]

sometimes more abstraction makes for simpler code, it's just that there's now simple presentation code for the frontend and also simple code for persistence.

avoiding a new layer of abstraction because it can just about be done in vanilla rails isn't a solution, it's just throwing more lines of code at the problem.

recently i've found that everything i do at the frontend intermixed with the controller, models and helpers could be made simpler with two way databinding.

it's also simpler, if someone want's to see if it looks right you can plug in some fixtures and tweaking css can be done independantly instead of needing to write something in ruby to affect how something in html looks.

maybe for a project with a large codebase is to use lighter frameworks such as backbone rather than having to do full rewrites in emberjs or angularjs. and let the code rewrite itself, ie. just clean up the portions of code that were only there for presentation.

[nottombrown]

You can't possibly think that there's never a use case for the Interactor pattern. Please watch Uncle Bob's talk: http://www.confreaks.com/videos/759-rubymidwest2011-keynote-...

[dhh]

Let's just say I have philosophical differences with the views presented in that presentation.

Also, I never said never. I said this particular example was a poor example and the general principle derived from it was equally poor. Garbage in, garbage out.

[rhizome]

I can imagine people thinking that this is a strength, but my frustration with that talk is that it isn't presented with much to do with Rails.

[halostatue]

I can possibly think that there's never a reason to reach for a pattern when it's better to let a pattern emerge from your code and then refine it. As far as the video is concerned I personally won't bother to watch it—Bob Martin sells snake oil as “clean code”.

[jfaucett]

I agree 100 percent with the statements in the first two paragraphs on "Part 1 Inflectors". I have first hand experience with every statement you made from testing to managing the callback dependency mayhem. This post seems interesting to me mainly for the test suite benefits alone, being able to maintain independency in the test-suite base would definately save lots of bloated stubbing and/or db hits. Im wondering what kind of time improvements the suite got?

[joenas]

Say you wanted to do the same thing from a Rake task or the console. Wouldn't an interactor be a nice thing to have then?

[halostatue]

I've only been involved in a couple of Rails projects (at least since the wedding list management app I wrote for my own wedding back in the v1 days), I'm in complete agreement with David.

And honestly, Rails already has something that works beautifully for an “interactor” extraction. It's called ActiveModel. Add a bit of other code (include ActiveModel::ForbiddenAttributesProtection, ActiveModel::Validations::Callbacks, and ActiveModel::SerializerSupport; extend ActiveModel::Naming) and you've got an object that acts like an ActiveRecord object but orchestrates changes across multiple models in a clean, constructive way—without introducing an entirely different way of working with your code.

I've done this at both places I've been doing Rails and it works at both cleaning up the models it orchestrates and keeping the business logic in the places where it belongs. I'm working actively to get rid of some of the Rails Fads that have come through and wreaked havoc on the current codebase (we've got Presenters and Commands and all sorts of other stuff that is just making the code harder to understand with little value).

There's a million ways to skin this particular cat, and claiming that “adult” Rails needs this particular gem or that particular way of writing code says a lot more about the person claiming it than the people who aren't following that particular fad.

[dasil003]

> And honestly, Rails already has something that works beautifully for an “interactor” extraction. It's called ActiveModel.

The thing which Rails and DHH offer no guidance whatsoever is the separation of persistence from the domain model. The Active Record pattern is itself a conflation of those two things, so this is an opinionated choice, and I respect it as a sane default for a wide variety of applications. It works perfectly fine up to a certain scale but when the business logic reaches a certain complexity and the persistence logic reaches a certain complexity then it makes sense to separate them. A lot of proposed solutions might be overkill, but keep in mind that DHH and the company formerly known as 37signals specialize in minimalist software, so they combat this complexity from the UX down rather than conceiving architectures to support it. I agree with this philosophy insomuch as all else being equal, simpler is better, but the problem is that some applications are necessarily more complex than Basecamp, and sometimes we need more than what Rails provides out of the box. Convincing DHH of this is pointless because he doesn't have to deal with it and he has no incentive to understand anyone else's pain in this regard.

[halostatue]

Here's the thing, though: Rails purposefully doesn't provide more than the minimum. I disagree with DHH on many things (ActiveRecord’s persistence model being based on MySQL misfeatures has been one of my biggest frustrations with Rails since he introduced it at a RubyConf where we both presented), but he's right that this sort of thing doesn't belong in Rails.

Everyone[1] needs a persistence layer and domain objects. That ActiveRecord is both ActiveModel and persistence is neither here nor there—and Rails 3 did a lot to making it possible to separate ActiveModel (domain object behaviour) and ActiveRecord (persistence layer behaviour) in a logical perspective so you can choose to persist differently than ActiveRecord does. Everyone needs some kind of controller. Everyone needs some kind of presentation layer. Aside from providing sane defaults for all of this, Rails doesn't try to dictate how your application is built.

The “more complex” architecture required for your application is unlikely to be a good fit for my application’s “more complex” architecture. For Rails to anoint one particular “more complex” architecture…is to fail anyone else who doesn’t need that architecture or would need something different. If you want to promote a particular type of architecture, there are ways: write an Engine; include generators. Give people guidance. Tell people why they want to use your particular architecture, but also tell them why they shouldn’t use it. If a lot of people need it, they will adopt it. Like Rspec. Like cucumber. Like ActiveModel::Serializers.

And, honestly, consider whether you really need your “more complex“ architecture or whether you can combat it the same way that Basecamp does. We're doing that with our product team, trying to find features we can remove and reimplement later when we really need them. Ours is a fairly mature product, so it's hard—but it's happening.

[1] For some large subset of everyone. Anyone who is reaching for something like Rails, Django, etc. needs some sort of persistence layer. Anyone who needs to store data needs that. Everyone needs domain objects.

[dasil003]

Where did I say Rails should provide this? My point was primarily that attempting to convince DHH of your architectural choices is sisyphean and ultimately pointless since he has crafted a nice bubble for himself where he doesn't have to be exposed to architectures or complexities that he finds distasteful.

[halostatue]

> The thing which Rails and DHH offer no guidance whatsoever is the separation of persistence from the domain model.

This seems to suggest you were looking for guidance or sane defaults for this. If that's not what you intended it to mean, I'm sorry for misreading it.

[dasil003]

Your interpretation is not what I intended but neither is it completely unfair.

I am looking for guidance (or rather good discussion and ideas) about how to address this issue for larger Rails apps, just not from Rails or DHH. I don't think it's Rails' place to do this since I don't really think a one-size-fits-all approach is right and it sort of tramples over the core philosophy of ActiveRecord which I believe is a useful default. On the other hand I don't buy into DHH's philosophy that its all astronaut architecture either.

[tomblomfield]

I agree with you - Interactors are particularly useful when you're creating multiple models, and the example could have been better.

But they're useful for dealing with side-effects of a particular operation too (sending multiple emails, notifying admins). Much better than ActiveRecord callbacks.

If you put this logic in the controller, what happens when you want a separate API controller that does the same thing? Or some admin functionality elsewhere in the codebase? There's zero possibility of code re-use.

[dhh]

Coulda, woulda, damn shoulda. You're future coding with your "what ifs". Most controller actions are not reused. The time to extract for reuse is when you need to reuse. Not crystal balling about that you probably will be in the future.

Because when the reuse case actually arrives, you might find that you need to reuse some, but not all of the action. If you're literally doing the exact same thing for both web and api, why are you using a separate API controller? Just use respond_to with formats.

Second, yes, you shouldn't have side-effects like sending email in your models. But you don't need to! Just stick that logic in your controllers. That's what it's there for -- to render the views (of which emails is one of them -- see http://david.heinemeierhansson.com/2012/emails-are-views.htm...).

The AR callbacks are wonderful for coordinating the domain model. Often times you'll want to create auxiliary objects when something else is created. That's what's it's there for. Or to otherwise keep the integrity of the domain model in place.

[tomblomfield]

Sure, I think we're saying the same thing here.

This example is from a real-life codebase, and the interactor was extracted from 3 or 4 (bloated, repetitive) controllers. Perhaps this could have been made clearer in the post.

I'm not arguing at all that every controller needs an Interactor, or all Rails codebases should start out with Interactors present.

I am saying that Interactors are very useful concepts for re-using code, and that many medium and large Rails codebases could benefit from them.

[dhh]

Please do share the 3 or 4 controllers all sharing this logic. I'd be happy to play code pong with them.

[dablweb]

Regardless of reuse, controller actions can get awfully large if unchecked. What would you say is the maximum LOCs for a public controller method?

Callbacks in Controllers and AR::Models tend to make things worse IMO for anything other than authentication; and for intricate actions interactors seem like the best defence.

I have had the (mis)fortune to jump into a number of large Rails codebases and I can say with hand-on-heart that the only ones that made any sense off-the-bat were ones using a this-or-similar pattern.

The community is embracing policies, interactors and decorators for a reason. Real developers are having real problems when Rails apps get a certain size, and there is no endorsed method on how to handle these problems.

Surely a couple of asides on the Rails docs and some official endorsement could help point new developers in the direction of a possible solution? A solution the industry already seems to be taking; regardless of whether 37Signals deems it fit for their particular domain.

[halostatue]

> What would you say is the maximum LOCs for a public controller method?

As small as it needs to be in order to get the job done, and as large as it must be to get it done clearly. Sometimes that's zero lines of code; sometimes it's 500. You can usually factor down a 500 line method, but it may be worth asking what the breadth cost is should the code really be single-use.

If your metric is anything else, then you're playing Stupid Metrics Games, like that espoused by Code Climate.

[dablweb]

You're really against Code Climate huh? Just what did it do to you to garner such hatred? I'm not personally a fan but I can see how it may help large companies in certain situations.

Anyways, when I asked about a "maximum" I was not asking in absolute terms. What do you see as a reasonable max for the majority of use cases?

At which point do you start feeling bad about your method?

[halostatue]

I'm against Code Climate because it promulgates nonsensical metrics in ways that are actively harmful to good engineering. Just like Bob Martin and his concept of “clean code”. All of these things are at best sigil markers. They aren't solutions, and they can't provide any answers.

I don't have a max length to a method, because I'm not a prescriptivist. What I have is a point where I stop understanding what a method is doing, or I can't explain it cleanly. Sometimes I keep the method as long, but better comment it. Sometimes I factor it out. The point at which I do so varies based on the code, time, and many other factors.

I've worked on a C++ codebase that had a 3,000+ line function that was nearly impossible to factor out because of a non-trivial amount of state that would have either:

a. required passing much of that state as parameters to the factored out functions/methods (even if you put it in a wrapper object/struct); or

b. extracting all of the (working) behaviour into one or more objects whose sole purpose is to encapsulate the behaviour of this function.

I absolutely hated working on that function, but it was essentially the main loop of the program. You could step through those 3k lines and get a fairly decent feel for what the program was going to do, when, and how often it would repeat. The hardest part was where people before had extracted code out…that was called exactly once. We tried three times to extract the code—and failed three times, ending up leaving the code the way it was because our extractions were more complex and less understandable than the existing crappy code.

My point is that there's no such thing as a reasonable max for the majority of use cases. It depends on what you're doing and in what code.

[dablweb]

Do you have an example of a public controller method larger than even 25 lines that does its job clearly?

[halostatue]

Controller method that I can share? No.

This method in mime-types is reported by Code Climate as a code smell. They're wrong: it's the smallest it can possibly be while still correctly performing the necessary goal; any smaller, and you have to break it into multiple smaller methods that provide no value except keeping Code Climate happy. https://github.com/halostatue/mime-types/blob/master/lib/mim... The method is ~35 lines long. There are other cases I can provide from open source work (https://github.com/halostatue/mime-types/blob/master/lib/mim... is a good example: deprecated code, parser for a file type where splitting into multiple methods only complicates the logic flow).

It is true that the larger a method gets the less readable, understandable, and clear it gets—but that should never translate into the sort of nonsensical request you made, asking about maximum LOC. Asking that paints you as a prescriptivist who doesn't bother understanding the context the code requires.

I've been writing software for a long time, and while I try to write methods as short, clear, composable “paragraphs”, I sometimes will write something much longer than is readable because I can't figure out a meaningful way to break it down. That comes over time and reading and interaction with the code.

The best developers in my experience are pragmatic. They are aware of design patterns, but don't treat them like sewing patterns. They are aware of development practices, but don't treat them like holy writ. The GoF design patterns book is an excellent descriptivist treatise, but then people started treating it like a cookbook and looked for reasons to implement Patterns everywhere, rather than extracting the patterns from their code.

Blog posts like this one (that tell me that I should use an Interactor) pattern are actively dangerous, because they provide dicta without properly explaining the pain that the pattern evolved to solve, or the proper evolution of the pattern.

The current code base I work on shows a lot of evidence of Rails Fads just like this one, and I'm trying to get my team to step back and ask why we do things certain ways and to go back to first principles. Don't just reach for a Presenter because it's what was done before. Don't even reach for an ActiveModel::Serializer because it's what we're preferring now for API representations. Figure out your problem. State it clearly. Write the solution clearly. Find code that works similarly and figure out (a) if and (b) how they can be extracted into common code.

There is a cost to adopting things like Presenters and making smaller methods: your interface becomes larger. You can complain all you like of large files and functions, but code bases that have large numbers of classes whose purpose aren't clear…are harder to navigate and understand. (I have, in the current code base, unextracted code from external classes when that external class is used in one place and it makes the behaviour more understandable. It also provides a better place to understand where similar behaviour may appear later so that we can properly extract code if and when it is necessary later.)

[karmajunkie]

"The AR callbacks are wonderful for coordinating the domain model. Often times you'll want to create auxiliary objects when something else is created. That's what's it's there for. Or to otherwise keep the integrity of the domain model in place."

No, they're not. They're great for coordinating activities around your database. That's not a domain model, unless your domain is databases. Conflating domain with database through the use of active record doesn't mean that structural models are domain driven, and events based around database interactions aren't going to be any more domain driven than the active record objects themselves.

[tinco]

Please take a second to think about who you're saying "No, they're not." to. You'd think perhaps that he knows what he has AR callbacks in AR for, regardless of what you might think they're in there for.

The whole idea of ActiveRecord is that you to make your models with a clean interface to persistence. Among the features of that interface are the callbacks. They are there for your domain model to make use of, that's the whole story. If persistence is not part of your domain, why are you extending ActiveRecord?

[karmajunkie]

Please. I'm well aware of who the parent was. I'm well aware of what his intent was. And I'm well aware not only of the strengths of the active record pattern (which I implemented in at least two different languages five years before he put Rails out there—as did many, many others) but also its inherent weaknesses.

You keep using the term "domain model" to refer to active record objects, which makes it clear that you've bought into the Rails appropriation of the term. AR objects are not domain models. If they were, the persistence would be handled elsewhere. Compared to a real domain model, AR objects are just a short step above using a naked database driver itself (and not necessarily an improvement thereof.) So to answer your question ("why are you extending activerecord?") its because you're taking a shortcut to dumping data into your database, and failing to create a real domain model while you're at it. That's a path that works ok for plenty of apps—but you want to know why rails has a reputation as being great at prototypes and shitty at scaling complexity? Because ActiveRecord is shitty at scaling complexity, and that's a hell of a lot more relevant to the vast majority of applications than scaling performance.

[tinco]

Oh I insist.

"If they were, the persistence would be handled elsewhere."

Persistance is handled elsewhere, it's handled in the superclass. If you think that's not a good idea, you shouldn't be using ActiveRecord, because that's the way it works.

You could argue there's not enough abstraction from the DB in AR, but that's a different discussion. Here we discussed if the persistance callbacks were an appropriate tool for in the domain model, and I reckon they are.

Also, any fool knows that putting lots of code in a single file is not a good way of scaling complexity, so yeah at some point you're going to need to break your code out in concerns and what not. ActiveRecord most certainly does not stand in the way of that.

[rgbrgb]

>> If you put this logic in the controller, what happens when you want a separate API controller that does the same thing?

You refactor.

In my experience, anytime you're writing something for the sake of "possible code re-use", you're wasting time. Code should and does get refactored often. By adding levels of indirection from the outset, you add a barrier to refactoring and likely additional unnecessary code.

[malyk]

I bet dhh is going to say something like "then refactor your controller into a service/interactor/etc., but not before".

YAGNI, and all that.

[rafekett]

the example is poor, butI think you're missing the core point here -- slow tests. I don't think this really improves readability or organization much -- but my real world experience with a giant slow rails test suite gives me 100% confidence that patterns like this are the only way to not have a completely intractable giant test suite.

[javiercr]

Definitely agree. Would love to see DHH commenting on the slow tests problem.

[markdodwell]

How long does the Basecamp test suit (excluding any integration tests) take to run?

[evilgeenius1]

Large - yes, but with an extremely simple domain model (and better for it).