[dhh]

Ensuring that the Ticket is valid is obviously a domain model concern. Policy objects can be a fine idea when they're swappable and you need to allow for multiple different policies. This is not one of those cases.

Here's a much simpler approach that keeps the validation logic in the domain model and uses features that Rails has had since the dawn of the framework: https://gist.github.com/dhh/9672827

[dhh]

On a separate note, I feel like this article series might better be titled "The Missing Parts of Our Knowledge of Basic Rails Features".

Reinventing basic features doesn't make your Rails deployment "advanced", it just makes it convoluted. There's no bonus prize for introducing fancy patterns to situations that do not call for them.

Further more, here's the definition of the Active Record pattern, as described by Martin Fowler: "An object that wraps a row in a database table or view, encapsulates the database access, and adds domain logic on that data". The key part is that last sentence.

So a quote like what follows simply misunderstands the purpose the Active Record pattern: "A recurring theme in these posts is that ActiveRecord models should be very simple interfaces to a datastore – the User model should be responsible for validating and persisting attributes of a user, like name and date-of-birth, and not much else."

See the example diagram (which I actually drew for Martin back before even starting Rails!): http://www.martinfowler.com/eaaCatalog/activeRecord.html -- it includes domain logic methods like "getExemption", "isFlaggedForAudit", and so forth. Exactly like the pattern describes.

You're not a beautiful and unique snowflake.

[tomblomfield]

Thanks for taking a look - I certainly respect your viewpoint.

I guess where we diverge is the amount of domain logic which lives on an ActiveRecord model.

My experience is relatively limited, but in 4 or so years of professional Rails development, across many different codebases, the overwhelming majority of problems have been caused by bloated models with too many responsibilities.

You clearly have more experience with Rails, but your solution always seems to be "If you were smarter, you wouldn't have these problems with my framework". I've seen this problem recur again and again across multiple teams & codebases.

We're seeking to address that problem with these concepts that have been successfully used in other frameworks & languages.

[dhh]

Yes, when stumbling across bad code, the first instinct should be: how can I make this simpler. Not how can I wrap this bad code in more convoluted patterns. Don't use a big word when a small one will do.

Additionally, I find that the key problem with bloated models stems from missing domain models (often has-many-through models). Not from moving the logic of the existing models into more noun classes. In your example here, the purpose of the ticket is to tie a user to a grouper: that's exactly the place to put logic that governs that connection!

Finally, what gets my goat is this notion that these patterns are necessitated by "advanced deployments". As it was some law of nature that when your app hits a certain size, you have to introduce this litany of misfit patterns. That's like arguing that if your book is longer than 300 pages, it must also use really big words and complex sentence structures. What?

The solution to large applications is to double down on simplicity, not give up on it. It is even more important to get the basics right. Execute them beautifully before even contemplating to freewheel from there.

[nthj]

> Additionally, I find that the key problem with bloated models stems from missing domain models (often has-many-through models). Not from moving the logic of the existing models into more noun classes.

When I first started learning Rails 4 years ago, I watched a video of a talk you gave where you refactored a kludgy piece of code to use an additional resource/model. The code unraveled before our eyes.

I still remember that as a Neo moment.

[highpixels]

Don't suppose you have a link to that video? Or remember any more details so I could dig it up?

[nthj]

I think it was this one, but unfortunately I can't find the whole video.

https://www.youtube.com/watch?v=mp4z2eK1Avw

http://johannesbrodwall.com/2007/02/27/crud-rest-rails/

[jodosha]

> That's like arguing that if your book is longer than 300 pages, it must also use really big words and complex sentence structures. What?

I don't think this analogy applies. If my house is full of things, I prefer to put them in small, well recognizable boxes, instead of in one gigantic mess. This is simplicity too.

[pothibo]

Just food for thoughts, I forked DHH's implementation and modified it to support a bigger structure (so you can keep stuff in different files/class). I haven't tested the file so it may contains bugs, I apologize if it does.

The file name contains dash(-) replace them with slash(/) as github doesn't like path.

https://gist.github.com/pothibo/9673715

I hope you like it ;)

Edit: It had a shit loads of typos, I believe I have fixed most of them.

[dhh]

Love when people play code ping pong. There's too much talk and not enough play in these threads. But I have to ask you, do you really think that splitting out those two classes improved things for the example? Or was this just future coding for possibly-maybe extension points? Because from where I'm sitting, it made things less clear and harder to follow with no additional upside.

[pothibo]

In the context of your example, your implementation is the best, hands down.

My fork had two objectives. The first was to nail the argument of "What do I do if my model has 1.2k LOC". Look at it as an extension of your solution.

The second was to show that before using Policy, there is mechanism in rails that you can use to extract bit of code away from the model (concerns, validators, etc.)

So yeah, you could say this was a future coding for possibly-if moons are aligned- extension points ;)

To reiterate, I start with in-model validation 100% of the time and move from there as needs be.

[darrencauthon]

I don't think your model has to have 1200 lines of code to see the benefits. DHH's ticket class stars with 2 lines of model code, which is a very rare case in the Rails applications I've seen. Counting by hand, I see DHH's example has one, two, three...

Twenty-six (26) lines of model code that are dedicated to validating this specific "confirmation" validation.

Given that validation is just one of the concerns of models built with The Rails Way, imagine sprinkling all of the other code that a model will have in-and-around these 26 lines of code. The central purpose of those 26 lines of code vanish into the fog of the rest of the model, creating the mystical behaviors that are common in some Rails applications.

I totally understand the concern about preemptive architecture, but once the code count for a given behavior creeps from 4, to 8, to 12... it's time to organize it into some sort of central place that's isolated from the rest. I mean, gosh, we know how this works in the real world, right?

Dev A: Here's my 26 lines of validation code to the model, DHH says it's the right thing to do.

Dev B (hours later): I need to add a little bit of behavior here, but I'm only adding a little bit of code. I'm not going to try to deduce what all of that code does, so I'll just tack mine on the bottom.

Dev C (days later): I have to add a bit of code... hmm... this is sort of a mess, I see mostly validation code but I also see some behavior. I'm not touching any of it. I'll just patch my stuff here... and here!

[iterate over and over, you get a mess of a Rails app]

I've seen this happen to multiple Rails applications... Rails applications built by experienced developers who know Rails in-and-out.

Plopping a bunch of code in something as important as a Rails model and leaving the cleanup for the next dev isn't very polite.

[ollysb]

The theme I'm seeing in this series is a desire to add a layer between controllers and the domain. This new layer seems closer to the controller than the domain (it always acts upon the domain) so are you not simpler making the controller layer thicker? By removing logic from the domain you're creating a distance between the information and your business policies. For instance you now have to remember when to apply particular policy objects. If the logic lived in the domain then the objects would take care of this automatically.

Fundamentally I have an issue with this whole approach, if you have bloated domain objects then organise them better (as DHH mentions a common problem is not having enough classes in your domain). Also, given the choice between bloated objects and logic separated from it's information I would take bloated objects any day.

[AnneOminous]

But if you have a beef with how much of the domain logic is in the Model, don't blame Rails, blame MVC. I mean, come on.

Either you want to use MVC or you want to go in a different direction. Fine. But don't blame Rails for being a good MVC implementation. Instead, formulate your own, new "philosophy" about how this all should work.

While I don't generally acknowledge the validity of "Before you criticize, think of something better", in this case I think it's appropriate. You appear to be critical of the whole MVC philosophy. And that's FINE. No problem. But before you go criticizing Rails I think you should formulate your own philosophy of how it all should work instead, rather than blaming Rails for doing what it is supposed to do, and doing it well.

[pothibo]

>> On a separate note, I feel like this article series might better be titled "The Missing Parts of Our Knowledge of Basic Rails Features".

Agreed on so many level. I see so many new gems that does the exact same thing as what rails do by default. I've spent some times browsing Rails source. It took me a while and as a side effect, the number of gems I am using now is a fraction of what I used when I started.

You can't build libraries and abstraction on top of Rails if you don't understand rails to begin with.

[mnarayan01]

I (more or less) agree with the sentiment expressed by "The Missing Parts of Our Knowledge of Basic Rails Features". That said, looking at your example, grouper_cant_be_full would not allow me to provide their desired functionality unless I already knew how to do it. Simply passing a symbol rather than a string there, along with a fairly short comment about what to do with it (i.e. in the controller and config/locales) would easily help people expand their knowledge. In short, you might be right in the above sentiment, but posts like these make hard for me to fault people like the author.

Edit: I should probably note that I _don't_ think you should be "required" to help out like this at all. Simply developing/sharing Rails is certainly _way_ more than I'm doing for other people. I'm just saying if you _are_ going to comment on this stuff, it would be nice if you transferred some of your greater understanding with your comments.

[ritchiea]

Thanks for responding to these threads. While the "beautiful & unique snowflake" stuff is a bit much, I appreciate developers standing up for simplicity in design. The architecture astronauts try to take the intellectual high ground by providing a complex solution and it drives me nuts. We should be working to make our code less complex not more complex.

[jonahx]

Non-facetious question: In what sense do you consider your solution simpler than the one suggested by the article? The only additional complexity I see in their solution is creating one more class, but that seems like a surface complexity.

I respect your opinion but it seems that you aren't really addressing the advantages the article is claiming their design has.

The likelihood of change is much different for the policies than for the basic domain model attributes of a ticket, and this to me is the key reason to prefer the policy design that I'd like to hear you address objectively.

Based on your comment about swappable policies, I'd guess that your position is to wait until you need to pull it out before you do. But if you can see a likely change coming and there is no cost to designing for it, why not do it? Is your bone of contention with the idea that we can assign reasonable probabilities to what is likely to change; with the idea that there is no cost to pulling the policy object out now; with both of those things; or with something else entirely?

[VeejayRampay]

Well I was probably the only one, but I didn't even know you could use valid? with a custom context (which apparently calls the validations on that context).

The more you know...