The Serializer
The key behind how API platform turns our objects into JSON... and also how it transforms JSON back into objects is Symfony's Serializer. symfony/serializer
is a standalone component that you can use outside of API platform and it's awesome. You give it any input - like an object or something else - and it transform that into any format, like JSON
, XML
or CSV
.
The Internals of the Serializer
As you can see in this fancy diagram, it goes through two steps. First, it takes your data and normalizes it into an array. Second, it encodes that into the final format. It can also do the same thing in reverse. If we're starting with JSON, like we're sending JSON to our API, it first decodes it to an array and then denormalizes it back into an object.
For all of this to happen, internally, there are many different normalizer objects that know how to work with different data. For example, there's a DateTimeNormalizer
that's really great at handling DateTime
objects. Check it out: our entity has a createdAt
field, which is a DateTime
object:
// ... lines 1 - 26 | |
class DragonTreasure | |
{ | |
// ... lines 29 - 48 | |
#[ORM\Column] | |
private ?\DateTimeImmutable $plunderedAt = null; | |
// ... lines 51 - 130 | |
} |
If you look at our API, when we try the GET
endpoint, this is returned as a special date time string. The DateTimeNormalizer
is responsible for doing that.
Figuring out Which Fields to Serialize
There's also another really important normalizer called the ObjectNormalizer
. Its job is to read properties off of an object so that those properties can be normalized. To do that, it uses another component called property-access
. That component is smart.
For example, looking at our API, when we make a GET request to the collection endpoint, one of the fields it returns is name
. But if we look at the class, name
is a private property:
// ... lines 1 - 26 | |
class DragonTreasure | |
{ | |
// ... lines 29 - 33 | |
#[ORM\Column(length: 255)] | |
private ?string $name = null; | |
// ... lines 36 - 130 | |
} |
So how the heck is it reading that?
That's where the PropertyAccess
component comes in. It first looks to see if the name
property is public. And if it's not, it then looks for a getName()
method:
// ... lines 1 - 26 | |
class DragonTreasure | |
{ | |
// ... lines 29 - 33 | |
#[ORM\Column(length: 255)] | |
private ?string $name = null; | |
// ... lines 36 - 59 | |
public function getName(): ?string | |
{ | |
return $this->name; | |
} | |
// ... lines 64 - 130 | |
} |
So that is what's actually called when building the JSON.
The same thing happens when we send JSON, like to create or update a DragonTreasure
. PropertyAccess looks at each field in the JSON and, if that field is settable, like via a setName()
method, it sets it:
// ... lines 1 - 26 | |
class DragonTreasure | |
{ | |
// ... lines 29 - 33 | |
#[ORM\Column(length: 255)] | |
private ?string $name = null; | |
// ... lines 36 - 59 | |
public function getName(): ?string | |
{ | |
return $this->name; | |
} | |
public function setName(string $name): self | |
{ | |
$this->name = $name; | |
return $this; | |
} | |
// ... lines 71 - 130 | |
} |
And, it's even a bit cooler than that: it will even look for getter or setter methods that don't correspond to any real property! You can use this to create "extra" fields in your API that don't exist as properties in your class.
Adding a Virtual "textDescription" Field
Let's try that! Pretend that, when we're creating or editing a treasure, instead of sending a description
field, we want to be able to send a textDescription
field that contains plaintext... but with line breaks. Then, in our code, we'll transform those lines breaks into HTML <br>
tags.
Let me show you what I mean. Copy the setDescription()
method. Then, below, paste and call this new method setTextDescription()
. It's basically going to set the description
property... but call nl2br()
on it first. That function literally transforms new lines into <br>
tags. If you've been around as long as I have, you remember when nl2br
was super cool:
// ... lines 1 - 26 | |
class DragonTreasure | |
{ | |
// ... lines 29 - 83 | |
public function setTextDescription(string $description): self | |
{ | |
$this->description = nl2br($description); | |
return $this; | |
} | |
// ... lines 90 - 137 | |
} |
Anyways, with just that change, refresh the documentation and open the POST or PUT endpoints. Woh! We have a new field called textDescription
! Yup! The serializer saw the setTextDescription()
method and determined that textDescription
is a "settable" virtual property!
However, we don't see this on the GET endpoint. And that's perfect! There is no getTextDescription()
method, so there will not be a new field here. The new field is writable, but not readable.
Let's take this endpoint for a spin! First... I need to execute the GET collection endpoint so I can see what ids we have in the database. Perfect: I have a Treasure with ID 1. Close this up. Let's try the PUT endpoint to do our first update. When you use the PUT endpoint, you don't need send every field: only the fields you want to change.
Tip
If you're starting a new API Platform project, this PUT
request will fail!
You can try a PATCH
request instead. This relates to a new standard_put
config in config/packages/api_platform.yaml
, which we talk about a bit
later in the tutorial.
Pass textDescription
... and I'll include \n
to represent some new lines in JSON.
When we try it, yes! 200 status code. And check it out: the description
field has those <br>
tags!
Removing Fields
Ok, so now that we have setTextDescription()
... maybe that's the only way that we want to allow that field to be set. To enforce that, eradicate the setDescription()
method.
Now when we refresh... and look at the PUT endpoint, we still have textDescription
, but the description
field is gone! The serializer realizes that it's no longer settable and removed it from our API. It would still be returned because it's something that we can read, but it's no longer writeable.
This is all really awesome. We simply worry about writing our class the way we want then API Platform builds our API accordingly.
Making the plunderedAt Field Readonly
Ok, what else? Well, it is a little weird that we can set the createdAt
field: that's usually set internally and automatically. Let's fix that.
Oh, but, ya know what? I meant to call this field plunderedAt
. I'll refactor and rename that property... then let PhpStorm also rename my getter and setter methods.
Cool! This will also cause the column in my database to change... so spin over to your console and run:
symfony console make:migration
I'll live dangerously and run that immediately:
symfony console doctrine:migrations:migrate
Done! Thanks to that rename... over in the API, excellent: the field is now plunderedAt
.
Ok, so forget about the API for a moment: let's just do a little cleanup. The purpose of this plunderedAt
field is for it to be set automatically whenever we create a new DragonTreasure
.
To do that, create a public function __construct()
and, inside, say this->plunderedAt = new DateTimeImmutable()
. And now we don't need the = null
on the property.
// ... lines 1 - 26 | |
class DragonTreasure | |
{ | |
// ... lines 29 - 48 | |
#[ORM\Column] | |
private \DateTimeImmutable $plunderedAt; | |
// ... lines 51 - 54 | |
public function __construct() | |
{ | |
$this->plunderedAt = new \DateTimeImmutable(); | |
} | |
// ... lines 59 - 128 | |
} |
And if we search for setPlunderedAt
, we don't really need that method anymore! Remove it!
This now means that the plunderedAt
property is readable but not writeable. So, no shocker, when we refresh and open up the PUT
or POST
endpoint, plunderedAt
is absent. But if we look at what the model would look like if we fetched a treasure, plunderedAt
is still there.
Adding a Fake "Date Ago" Field
All right, one more goal! Let's add a virtual field called plunderedAtAgo
that returns a human-readable version of the date, like "two months ago". To do this, we need to install a new package:
composer require nesbot/carbon
Once this finishes... find the getPlunderedAt()
method, copy it, paste below, it will return a string
and call it getPlunderedAtAgo()
. Inside, return Carbon::instance($this->getPlunderedAt()))
then ->diffForHumans()
.
// ... lines 1 - 11 | |
use Carbon\Carbon; | |
// ... lines 13 - 27 | |
class DragonTreasure | |
{ | |
// ... lines 30 - 118 | |
/** | |
* A human-readable representation of when this treasure was plundered. | |
*/ | |
public function getPlunderedAtAgo(): string | |
{ | |
return Carbon::instance($this->plunderedAt)->diffForHumans(); | |
} | |
// ... lines 126 - 137 | |
} |
So, as we now understand, there is no plunderedAtAgo
property... but the serializer
should see this as readable via its getter and expose it. Oh, and while I'm here, I'll add a little documentation above to describe the field's meaning.
Ok, let's try this. As soon as we refresh and open a GET
endpoint, we see the new field under the example! We can also see the fields we'll receive down in the Schemas section. Back up, let's try the GET
endpoint with ID one
. And... how cool is that?
Next: what if we do want to have certain getter or setter methods in our class, like setDescription()
, but we do not want that to be part of our API? The answer: serialization groups.
Since I am upgrading an app from 2.6 to 3 right now, I was wondering about a behavior of the serializer that I ovserved.
In API-Platform >= 3, the default setting in the api_platform.yaml is:
So far, so good. Properties, which are null, are no longer part of the response.
But in that specific app I work a lot with Doctrine Embeddables because I have giant forms, and it is a great way to put fields into logical blocks.
But when all properties of an embeddable are null, the endpoint returns an empty array.
I expected the embeddable not to be returned at all, or to be an empty object, but an empty array is returned.
Is that really the desired behavior?
I am right now wondering how to deal with that, since I have dozens of embeddable in use in that app.
I could modify all the getters and manually loop though all the properties and return null, if they are all null, or create some kind of custom normalizer.
Furthermore, I have two questions:
Is that really the wanted behavior?
What is the best workaround?