Login to bookmark this video
Buy Access to Course

Embedded Write


Share this awesome video!


Keep on Learning!

With a Subscription, click any sentence in the script to jump to that part of the video!

Login Subscribe

I'm going to try out the GET one treasure endpoint... using a real id. Perfect. Because of the changes we just made, the owner field is embedded.

What about changing the owner? Piece of crumb cake: as long as the field is writable... which ours is. Right now the owner is id 1. Use the PUT endpoint to update id 2. For the payload, set owner to /api/users/3.

And... execute! Bah! Syntax error. JSON is crabby. Remove the comma, try again and... yes! The owner comes back as the IRI /api/users/3.

Sending Embedded Data to Update

But now I want to do something wild! This treasure is owned by user 3. Let's go get their details. Open the GET one user endpoint, try it out, enter 3 and... there it is! The username is burnout400.

Here's the goal: while updating a DragonTreasure - so while using the PUT endpoint to /api/treasures/{id} - instead of changing from one owner to another, I want to change the existing owner's username. Something like this: instead of setting owner to the IRI string, set it to an object with username assigned to something new.

Would that work? Let's experiment! Hit Execute and it does not. It says:

Nested documents for attribute owner are not allowed, use IRI instead.

Allowing Writable Properties to be Embedded

So, at first glance, it looks like this isn't allowed: it looks like you can only use an IRI string here. But actually, this is allowed. The problem is that the username field is not writable via this operation.

Let's think about this. We're updating a DragonTreasure. This means that API Platform is using the treasure:write serialization group. That group is above the owner property, which is why we can change the owner.

215 lines | src/Entity/DragonTreasure.php
// ... lines 1 - 25
// ... lines 27 - 49
denormalizationContext: [
'groups' => ['treasure:write'],
// ... line 53
// ... line 55
class DragonTreasure
// ... lines 58 - 99
#[Groups(['treasure:read', 'treasure:write'])]
private ?User $owner = null;
// ... lines 102 - 213

But if we want to be able to change the owner's username, then we also need to go into User and add that group here.

172 lines | src/Entity/User.php
// ... lines 1 - 22
class User implements UserInterface, PasswordAuthenticatedUserInterface
// ... lines 25 - 46
#[Groups(['user:read', 'user:write', 'treasure:item:get', 'treasure:write'])]
// ... line 48
private ?string $username = null;
// ... lines 50 - 170

This works exactly like embedded fields when we read them. Basically, since at least one field in User has the treasure:write group, we are now allowed to send an object to the owner field.

New vs Existing Objects in Embedded Data

Watch: fire it up again. It works... almost. We get a 500 error:

A new entity was found through the relationship DragonTreasure.owner, but was not configured to cascade persist.

Woh. This means that the serializer saw our data, created a new User object and then set the username onto it. Doctrine failed because we never told it to persist the new User object.

Though... that's not the point: the point is that we don't want a new User! We want to grab the existing owner and update its username.

By the way, to make this example more realistic, let's also add a name to the payload so we can pretend that we're actually updating the treasure... and decide to also update the username of the owner while we're in the neighborhood.

Anyways: how do we tell the serializer to use the existing owner instead of creating a new one? By adding an @id field set to the IRI of the user: /api/users/3.

That's it! When the serializer sees an object, if it does not have an @id, it creates a new object. If it does have an @id, it finds that object and then sets any data onto it.

So, moment of truth. When we try it... of course, another syntax error. Get it together Ryan! After fixing that... perfect! A 200 status code! Though... we can't really see if it updated the username here... since it just shows the owner.

Use the GET one User endpoint... find user 3... and check that sweet data! It did change the username.

Ok, so I realize that this example may not have been the most realistic, but being able to update related objects does have plenty of real use-cases.

Cascading the Persist to Create a new Object

Looking back at that PUT request, what if we did want to allow a new User object to be created and saved? Is that possible? It is!

First, we would need to add a cascade: ['persist'] to the treasure.owner ORM\Column attribute. This is something we'll see later. And second, we would need to make sure to expose all of the required fields as writable. Right now only username is writable... so we couldn't send password or email.

The Valid Constraint

Before we keep going, we are missing one small, but important, detail. Let's try this update one more time with the @id. But set username to an empty string.

Remember, the username field has a NotBlank above it, so this should fail validation. And yet, when we try it, we get a 200 status code! And if we go to the GET one user endpoint... yeah, the username is now empty! That's... a problem.

How did that happen? Because of how Symfony's validation system works.

The top-level entity - the object that we're modifying directly - is DragonTreasure. So the validation system looks at DragonTreasure and it executes all of the validation constraints. However, when it gets to an object like the owner property, it stops. It does not continue to validate that object as well.

If you want that to happen, you need to add a constraint to this called Assert\Valid.

216 lines | src/Entity/DragonTreasure.php
// ... lines 1 - 55
class DragonTreasure
// ... lines 58 - 100
private ?User $owner = null;
// ... lines 103 - 214

Now... on our PUT endpoint... if we try this again, yep! 422: owner.username, this value should not be blank.

Being able to update an embedded object is really neat & powerful. But the cost of this is making your API more and more complex. So while you can choose to do this - and you should if it's what you want - you might also choose to force the API client to update the treasure first... and then make a second request to update the user's username... instead of allowing them to do it all fancy at the same time.

Next: let's look at this relationship from the other side. When we're updating a User, could we also update the treasures that belong to that user? Let's find out!