Scroll down to the script below, click on any sentence (including terminal blocks!) to jump to that spot in the video!
SymfonyCon 2018 Presentation by Christopher Hertel & Christian Flothmann
With the popularization of DDD people started shifting from anemic models with only getters and setters to a rich model describing the state changes in specific methods. This way of designing models does not play well with Symfony forms. User provided input is inherently invalid while we want to maintain certain invariants in our domain model. A common approach to overcome these limitations is to create data transfer objects our forms are then bound to. This can lead to lots of mapping & glue code that might be cumbersome to write and maintain.
But couldn’t we do better? In this talk we will discuss the different aspects of a rich domain model that makes it hard to use it in conjunction with the Form component. We will then look at the possibilities to hook into the data flow of the form handling and discover how we can modify it to interact seamlessly with our model.
Welcome. Hi. Hello. The stage is huge, it's intimidating, but we're pretty happy that so many of you joined us. And we're talking about Symfony forms. So who have you used Symfony forms before? Oh, that's an easy one. Right? And we're talking about Symfony forms using rich domain models. Who's familiar with the term rich domain model. So quite a few and uh, but we are starting with introducing ourselves.
So let's do this. That is Christian, Christian is working for SensioLabs, in Germany and uh, besides being a software developer at SensioLabs he's working as a core member of Symfony software and Symfony docs.
And this guy next to me is Christopher, he's working at SensioLabs in Germany too. And he's also one of the organizers of Symfony user group in Berlin. So if you ever come to Berlin and you're lucky to, to attend the user group, you may meet him there too.
And Christopher will now guide you through a simple example application that we created to show you the concepts of our talk. So in order to make it easier to understand everything.
Yeah. Well, so we're doing it hands on with a simple use case and our use case is a simple product CRUD. CRUD is an abbreviation for create, read, update, delete. I think you're familiar with that principal, a principal and a, it looks like that. It's a simple interface. Looks a lot like bootstrap I guess. And there are two forms in it. So, one about the category and one about the product. And the domain is a product with a name, a category, and a price. Pretty simple. On the category itself has a name and an optional parent category pretty simple as well. And the price object has an amount tax rate and a currency.
So, uh, that's the workflow of the forms and we're talking just about the product form with a create and edit. So you're quite familiar with that. I guess it's a very simple action that handles, uh, both steps the create and edit. So let's see, at the very basic solution, that's pretty much a, like the best practice document in the documentation. So we're having a product controller with a form action and that's just creating the form type, handling the request, binding request to the form type and then the very typical, if it's submitted and it's valid structure with a redirect at the end or the rendering of the form optional with data and errors or plain.
The form type itself, is very straightforward as well. So we're having a product type with a name category, price amount, price tax and price currency. So we're not having a price object on this level, but three price attributes and the product type. And we're binding the product product class to the type.
Next up, the
Product class. So the model, which is the focus of our talk today, right? The model. And we're having a simple, a simple product entity with a private properties with assertion annotations on it and simple getters and setters, right? So pretty straightforward.
So, um, what's happening now when we are getting our forms. So, um, this is the form as you will see it in your browser and you will enter data and so on, and your browser will submit an HTTP request. You can see on the headers and then you have the body of the POST request with all the data coming from the form. And inside your your front controller, the incoming request will be transformed into a request object which is part of the http foundation component, which is an object oriented layout on top of the super global. It's basically what PHP offers you to handle HTTP requests and this is what it looks like if we dumped the structure. So you basically have an object around a structured array. And then inside your controller on this side, you have the request object on top, which is then passed down into the handle request method, which means that the request is handled by the form component and we can later check if form was submitted or not. Yes, right here. And here the form is taking into account that we are binding our form to the product class.
This was the final object, all the data that was coming into our controller was bound by the form and is now part of the form object.
So let's take a look at three columns of our three types of code that are involved to solve our problem. First, there was third party code that we just want to use that we don't want to maintain that we don't want to write ourselves, which means the front controller, which is just bootstrap code from a application. We have the HttpFoundation component as the object oriented layout on top of HTTP. We have the form component of course. And we have the validator component which we have seen in the product model than we have added assertions to our model to, to ensure that domain roots are taken into account. Then we have our domain layout, which means the core part of our business. Everything that's individual for our application, which is basically the product class the category class and the price class or the price price object.
And then we need to write some glue code to somehow wrap all these things together. And this is the controller, which means we are handling the request that we are dealing with a request and returning a response. We need to create a form type which describes how we map our request to our model and if we take a look at this from a graphical point of view, we have the data flowing in from, from the left and going through all the layouts in our model, our model is bound to the form which is then handling the request and putting data back into the model. So usually we want to focus on these yellow blocks, which is the core part of our business. And we usually want to reduce the green part. So the part that is just there for coupling our domain to the framework to the least minimum.
Okay. So that's the basic setup. Probably nothing new for you.
And the next step is that we look at our model, um, and we need to differentiate about anemic domain model and rich domain model. What we saw, the product type and the product is a dynamic domain model which is focused on data. So there's a structured way of having data in an object. It's very easy to implement and to maintain even you can generate it using helpers, like MakerBundle or a or the PhpStorm has some utilities as well. And it contains little or no logic at all, so maybe no constructor, no validation logic and setters, and it's not guaranteed that the data that is in their object is valid or is consistent. So that might be a problem.
Rich domain model combines data and logic and it is valid by design. So we have mandatory constructor arguments, for example. So we need to give a category name and category can't exist without a name. It's quite easy to test. Same is true for the anemic domain model, but some might argue it's not really sensible to test the anemic domain model. And we can now create state transitions by name and test them and um, do assertions on transitions.
And that's the last point, defined state transitions and we can, um, the concept is to be more explicit, explicit on the names of state transitions and not just have set name, but uh, maybe rename process.
Important small disclaimer, we don't want to lecture you about using rich domain models is the best way to do it. It's sensible to choose between them. Right? And that's, that's up to you. And uh, this guy, Martin Fowler wrote in 2003 that anemic domain models are anti-pattern and then stack overflow was a bit curious if it's an anti-pattern or not. And then there was discussions how they differ and how to avoid anemic domain models, how to find the right balance between using both maybe. Um, someone came up with this idea. It's not an anti-pattern anymore. Someone says it's SOLID design. Some said it's not SOLID, rich domain is SOLID and stack overflow was very confused. So that's the basic bottom line at stack overflow article about it and how not to agree about software design. And we just want to point out their differences and reasons to use one of them.
Anemic models are good for prototyping, very, very easy to use, easily generated and rich models are more about clean code and object oriented, um, testability and yeah, we can design our domain in it including using name transitions and process. Um, but it's your choice, right? But the next point is we want to show you how to use rich domain models with Symfony.
So if we take a look at our former model, we didn't have any defined state methods since we had just setter and getter methods so you would have to call them however you like and then you would have to run validation after. That's not how we want put that. So the first step is to, to, um, require every attribute to be passed when we are constructing the product. So we have to put the name, the category and the price. For category and price we can be sure that both are already valid because these routes apply there too. But for the name we need to, to perform some, some validation. We need to, to make sure that that the constraints that we defined are still applying here. And you can also see that we have removed the validation assertions, so this is something that is now happening inside our, our model. The same, happens here when we are renaming our product and you can also see now that we don't have a set named method anymore, but we have a rename method now so this is more like, like describing what we are doing with our model.
And here's our validateName method, which was just basically checking the length of our name. The price is a bit special because the price itself is not identified by an id or something like that. Um, price, just, well having a value. And this is what makes it so special because one price object can now be shared between different products. But if we are doing that now we want to compare them. So the value must not change. And this means that our price object has to be immutable. We will see that later, what that means for our forms after that. So we're doing all the validation again. And the same applies for the category. So if we are now using this model without making any other changes, we would run into many different exceptions and well that's not really funny. Um, so we need to do something about that to make our forms usable with our new model.
So the first solution that we came up with is using data transfer objects or DTOs. That's quite easy and looks like this. That's the DTO having public properties and the validation assertions on it. And then we are having some kind of mapping methods where we map the rich domain model to an DTO and back again to an entity. So right here we can check if the DTO, if there are changes and we can explicitly call those rename, moveToCategory or even the Costs method and apply those changes only when we need them. So that's pretty easy to do. And the code looks a bit like the model before, right? Um, the product type is actually the same besides the small change down there, changing the data class. The controller is a bit different because we have to map the DTO from entity and to get the DTO from the form and map to the entity again before persisting it. So quite simple steps, but we now can use rich domain models. Right?
Okay. From the user's point of view, that's not so much that is just changing the interface still looks the same as before the data that is sent is still the same, it's still transformed into the request object, as before. And, the interesting changes are happening here now because as Christopher already explained, the entity is first transformed into the DTO because we need to, to unbind the form to this DTO. We're handling the request here now and here we see the binding to the DTO again, we have the reside here this is the DTO and we need to um, transform this back into the entity. So we are catching the data from form and parking it here. Um, the entity so we can now persist this entity again.
This three column model that we have already seen before. And what changes here now is the right column, because we now have additional glue code. We have written DTOs and this is more code than before, um, which you can see here too. So before, we only had our model here and now we also have a green rectangle on the left and on the right and well basically we want to reduce this part. Now we have have added some, so um, maybe that's not the best solution we can come up with. But the solution enables us to use rich domain models by decoupling the model from the form type and it's very easy to implement and maintain as well. So the DTO was quite simple. Um, but it's additional glue code. Right? And I don't know if you saw that, but we have redundant validation rules, right? We have validation in our rich domain model and we need to have validation in our DTO. So this one has an impact on the maintainability. Next solution we came up with is harder.
Okay. Who of you has implemented a data mapper for forms before? Okay, one, two, three. Okay. 10 or 20 or so. So I'm not so many and yes, I think it's really the hardest solution, but we want to explain it nonetheless to show you. Um, and I think it's, if you've never used it before, it might be quite fast, but it's not the point to explain data mappers, we just want to show that it's possible to use data mappers as a solution to decouple and use rich domain models. Yes. So our controller now looks again the same as it was in our first solution when we were using the anemic domain, which basically means that we are now again binding the product entity to the form.
And the form is the part where things changed now. First we introduced a new price object which is just extracting the price related properties from the product and put it into a new model. And now we have a data mapper, so a custom data mapper. The form component ships with a default one, but we cannot use that. Um, so we add a new one here and we just say, okay, The data mapper that we want to use is implemented inside, this product type. You could also move it into another class that doesn't really matter here and a data mapper has two methods, methods. The first one is this one is called mapDataToForms and it's populating the form with the initial initial data. So we, basically here take the subtypes and set the data based on the model that was passed here. So the data variable is our product entity.
And after the form has been submitted, we must remap the data from the form into our entity and this was the part where things are a little bit more complicated now because we don't have any setter methods anymore. Um, so we are first getting the data from the form, comparing them to the data inside our entity, and then we call our state transitions, state transitioning methods. So for example, rename here or here, we are moving our product into a different category or the price changes and we are calling this method in this case.
When we are creating a new product, we need to make sure that the form is as populate a new product entity is populated with the submitted data and we are doing this here with the empty data option. Passing it a closure where we get the form and we can extract all the data from the form and pass it to the constructor of our product entity. We need to make sure here that we catch all the exceptions that could occur so that the user does not see an internal server error, but we can present them some kind of a nicer error message.
The data flow, it's still the same, or basically the same. Here's a little bit, a little difference that you can see because we now have the price as an additional nested element, but that's just a small detail here. Yeah, we moved the price values to a lower type, right, so we have a price type right now. Submitted data is still the same. The user does not change what they want to enter there and you can see here that the controller now is the same as in the first example. The request was handled again and here is the difference, we are now not not, the FormType still is bound to the product entity, but the important part now is that we have created our own data method that is used here. And the resulting Product, that's the important part is still maintained. We are basically changing what's happening when our form is handled, but we're not changing our model.
Here our three column model that we are familiar with now. Changes a bit. Now, we don't have DTOs, but we have data mappers and the validator component is gone too because in the process, step one, we introduce in the second step when we introduced our rich domain model, we are implementing our domain assertions by by doing the validation inside our methods and throwing exceptions and now that we don't have DTOs anymore, we also do not need the validator component anymore.
But the glue code part still makes a big part of our application. We have now data mappers that we need to maintain, that we need to change where we need to think about how can we test them, which is yeah, massively different story because testing data mappers really is not fun and um, but you probably want to do that because this is an important part of your application.
So the bottom line with data mappers enabled, enables us to use rich domain models and that requires advanced knowledge about the form component and the extension points of the form component. And this is additional glue code as well. Like it's not a with redundant validation rules we got rid of that but it's quite hard to test and it's quite hard to understand if you're getting in a new project and there's data mappers all over the place.
So, DTO's, redundant data mappers are hard.
Maybe we can, we can ask questions. So who have you after just having heard this, um, would like to use rich domain models with Symfony forms? Oh, okay. That's not so many. Would you prefer. So I think it was five or four people. Would you use DTOs? Or would you use data mappers? Oh, okay. Half and half I think. Okay. But uh, yeah, I mean it's clear that this is maybe not the best solution yet.
We have another one, RichModelFormsBundle. Yeah, we could, we could work on the name. Okay. But it's already out there. So it's RichModelFormsBundle. So, um, you can easily install it. It's still experimental. So we're working on the idea and uh, today is once again a kick off for a discussion, maybe. The installation is quite easy with a tool called Composer.
Um, and now we're talking about the features. So yeah, basically the idea behind this bundle is, and that's why we have shown you the third solution that the, the data mapper approach maybe could be generalized to something where you do not need to write the code yourself, but you would just configure this generic data mapper so that it does what you want based on your use cases. And um, yeah, now it's your turn. Thanks. Again.
First feature different write and read property path, there is an option already property paths, whether you can configure how the property accessor, um, talks to your model, but the new thing is that we can use, read and write property path, um, different. Because the problem with the initial, um, within default data mapper is that the getters and setters must really be named getName and setName.
In this case, we'll look at this example and um, yeah, we didn't want to have a setName method because in our domain it's called rename. So, um, we thought, okay, why don't we split the property path in, read and write. And yeah, that's the name, read_property_path and write_property_path. So in our case we are saying, okay, when you're reading the value, you just call getName and please call rename when we want to change the value. You could also pass a closure there if you need to to maybe call different methods based on the value of a checkbox, for example, because you maybe have, enable and disable and not something like set enabled with true and false.
A quick side note, the bundle implements a couple of form options, additional form options that you can use on your form type. So, exception handling. So our domain model throws exceptions on invalid input or on maybe in constructor, like this was the price object, right? Yes. And what you can now use is the expected_exception option. We are not really satisfied with this name because it reminds us more what we're used to when we are, um, using php unit, but in fact we don't want the exception to be thrown. We just want the bundle to intercept here. So yeah, if you have ideas for a better name, we would really welcome any suggestion here. Um, but I think the idea should, should be pretty clear so you can, can define which exceptions can occur here and which are valid to be thrown because they might be thrown by domain validations here and we want to show something to the user, some kind of of nice error message to show them what they did wrong.
Yeah, I guess the, the error handling of the domain is quite important part. We need to tackle because the exception message is not very good for the interface, so we need to, uh, to decouple that. And maybe there are some other validation concepts out there, not using exceptions, but using a message bags and something.
One thing we didn't mention yet is that all type errors that may occur because you passed the wrong data type to the method will automatically caught being caught too. So if you, um, have a model, you will often have realized that you will have to allow null values being there, and this will be intercepted as well.
Next one, mandatory constructor arguments. The form component, uh, creates objects, right? And, um, there's an empty data option and you can hand over already created objects. But in this case where we have an immutable value object or a value object, it's quite hard. This also applies for product entity, and yes, we could work around that by writing this closure ourselves, but that's um, 10 lines of code ourselves for every form type. So, um, let's define a new option which is called factory and factory can be the name of a class, which means, um, all the, um, values of this type would be passed to the constructor of this class. You could also, um, pass a closure on any PHP callable, um, so maybe if setting method on another class or so, and this would then return a new object. But only for cases where the form is populated without any initial data object. So when you maybe are creating a new product or so.
And the names of the properties, uh, the names of the fields are mapped to the names of the arguments so you can easily change them, um, to, to have a different order, something.
Next one, immutable value objects. So once again, the same code, um, but we need to create a new object on change because we can't um change the basic, uh, the base data object, right?
And um, okay, we already have the opportunity to define a factory. So let's just add another option and let's define it immutable because that's what we are dealing with here. Immutable objects and if we set this option to true every time you submit new data a new object will be created and the initial data object that was bound to the form is still the same as it was before we were handling the form.
So about the implementation, controller still looks the same as in the first solution. The form type is a bit different now because we have added all these options that we have shown you before, but that's the part where we now need to make adjustments. From the point of view of the model we are now still using the rich domain model, which is the main difference to the first solution where we were using the anemic domain model. In the data flow there's not so many differences here, here the request is coming in and behind the scenes the important part is now happening as part of a RichModelFormsBundle. And the code, um, we had our custom data mapper implementation before this is now gone because we have the new bundle. And the important part here is the switch or the movement from the right column, which is what we have to maintain to the left column, where there is code that hopefully just works as it is supposed to do and does not need to be maintained by us. So, um, we are here and we can now focus again on our core domain and we don't have to deal with stuff that is not part of our business. And well, we are not making money with it probably.
It generalizes the previous approach of the data mapper, so behind the scenes, um, the bundle implements a couple of data mappers and creepy stuff. Um, but we have additional data options form options and um, they are, the target is using rich domain models so we can improve, uh, the goal is to improve that and it's released, but we're still working on it. I think you tagged 0.1 yesterday or today. Yes.
Some say there are some installations on production already, but I wouldn't do that maybe. It's not feature complete. Rich domain models can be very different because they belong to your domain. And we just, we just targeted those cases we came up with. So maybe you have other targets and use cases and you could easily share them with us on GitHub. That's, that's your part to do maybe, um, yeah, that's a lie. But, uh, I don't know the project, maybe that's not that big. So it's your turn to help us give us input about, um, maybe the namings of the options and maybe your use cases. Just maybe just try it in your application yet, not in production maybe, but, but on your test server, or on your local development machine. Try it out and give us feedback, what works, what maybe does not work and be nice and where you need to write your own glue code once again. So we want to get rid of the glue code, right?
So, DTOs to sum it up, um, it's easy to implement, duplicated validation, it's additional glue code. Data mappers, hard to implement, hard to test, and additional glue code as well. New Bundle, no glue code on your side, it's experimental and maybe we can bring some of those ideas to the core in the future.
That's it. Thank you.
Three minutes for questions. Please don't ask hard stuff.
Oh sorry. Does this work? Hello? Can you go to a slide where you actually put the data array the exception in and stuff?
Sorry, I don't hear anything.
Can you go to the slide where you put your extra property paths and everything in there to the form type?
Um, maybe you can ask the question without the mic and we'll just repeat it. I think it's better to understand. Maybe we can. No, we cannot.
So the question is if you can use...You mean for this or for that? Oh yeah, that's how you use the form component. That's okay.
Can you repeat the question please?
That's, that's part of the interface that we cannot change here because that's coming from the, from the core. So the question was, do we need to pass an array here or can we pass an object? So yeah, that's not, that's not possible.
Yeah. But then we would have to work around much more stuff. Yeah, that's probably not what you want to do.
Alright. Christian and I will be around conference social as well, so see you there. Thank you.
"Houston: no signs of life"
Start the conversation!