Testing, Updating Roles & Refreshing Data

This test is failing... which is great! It proves that our end goal - only returning the phoneNumber field to users authenticated as an admin - is totally not working yet. Before we get it working, let's finish the test: let's make the user an admin and assert that the phoneNumber field is returned.

First, "promote" this user to be an admin: call $user->setRoles() and pass this an array with ROLE_ADMIN inside.

Services are Recreated between Requests

Easy! But... we need to be careful with what we do next.

Let me explain: in the "real" world - when you're not inside a test - each request is handled by a separate PHP process where all the service objects - like the entity manager - are instantiated fresh. That's just... how PHP works: objects are created during the request and then trashed at the end of the request.

But in our test environment, that's not the case: we're really making many, sort of, "fake" requests into our app all within the same PHP process. This means that, in theory, if we made a request that, for some reason, changed a property on a service object, when we make the next request... that property would still be changed. In the test environment, one request can affect the next requests. That's not what we want because it's not how the real world works.

Not to worry: Symfony handles this automatically for us. Before each request with the client, the client "resets" the container and recreates it from scratch. That gives each request an "isolated" environment because each request will create new service objects.

But it also affects the service objects that we use inside our test... and it can be confusing. Stick with me through the explanation, and then I'll give you some rules to live by at the end.

Check out the entity manager here on top. Internally, the entity manager keeps track of all of the objects that it's fetched or persisted. This is called the identity map. Then, when we call flush(), it loops over all of those objects, finds the ones that have changed and runs any queries it needs.

Up here, this entity manager does have this User object in its identity map, because the createUserAndLogIn() method just used that entity manager to persist it. But when we make a request, two things happen. First, the old container - the one we've been working with inside the test class so far, is "reset". That means that the "state" of a lot of services is reset back to its initial state, back when the container was originally created. And second, a totally new container is created for the request itself.

This has two side effects. First, the "identity map" on this entity manager was just "reset" back to empty. It means that it has no idea that we ever persisted this User object. And second, if you called $this->getEnityManager() now, it would give you the EntityManager that was just used for the last request, which would be a different object than the $em variable. That detail is less important.

On a high level, you basically want to think of everything above $client->request() as code that runs on one page-load and everything after $client->request() as code that runs on a totally different page load. If the two parts of this method really were being executed by two different requests, I wouldn't expect the entity manager down here to be aware that I persisted this User object on some previous request.

Ok, I get it, I lost you. What's going on behind the scenes is technical - it confuses me too sometimes. But, here's what you need to know. After calling $user->setRoles(), if we just said $em->flush(), nothing would save. The $em variable was "reset" and so it doesn't know it's supposed to be "managing" this User object.

Here's the rule to live by: after each request, if you need to work with an entity - whether to read data or update data - do not re-use an entity object you were working with from before that request. Nope, query for a new one: $user = $em->getRepository(User::class)->find($user->getId()).

We would need to do the same thing if we wanted to read data. If we made a PUT request up here to edit the user and wanted to assert that a field was updated in the database, we should query for a new User object down here. If we used the old $user variable, it would hold the old data, even though the database was successfully updated.

Logging in as Admin

So I'll put a comment about this: we're refreshing the user and elevating them to an admin. Saying ->flush() is enough for this to save because we've just queried for this object.

Below, say $this->logIn() and pass this $client and... the same two arguments as before: the email & password.

Wait... why do we need to log in again? Weren't we already logged in... and didn't we just change this user's roles in the database? Yep! Unrelated to the test environment, in order for Symfony's security system to "notice" that a user's roles were updated in the database, that user needs to log back in. It's a quirk of the security system and hopefully one we'll fix soon. Heck, I personally have a two year old pull request open to do this! I gotta finish that!

Anyways, that's why we're logging back in: so that the security system sees the updated roles.

Finally, down here, we can do the same $client->request() as before. In fact, let's copy it from above, including the assertJsonContains() part. But this time, assert that there should be a phoneNumber field set to 555.123.4567.

Phew! Ok, we already know this will fail: when we make a GET request for a User, it is currently returning the phoneNumber field: the test is failing on assertArrayNotHasKey().

Dynamic Fields

So... now that we have this big, fancy test... how are we going to handle this? How can we make some fields conditionally available?

Via... dynamic normalization groups.

When you make a request for an operation, the normalization groups are determined on an operation-by-operation basis. In the case of a User, API Platform is using user:read for normalization and user:write for denormalization. In CheeseListing, we're customizing it even deeper: when you get a single CheeseListing, it uses the cheese_listing:read and cheese_listing:item:get groups.

That's great. But all of these groups are still static: we can't change them on a user-by-user basis... or via any dynamic information. You can't say:

Oh! This is an admin user! So when we normalize this resource, I want to include an extra normalization group.

But... doing this is possible. On User, above $phoneNumber, we're going to leave the user:write group so it's writable by anyone with access to a write operation. But instead of user:read, change this to admin:read.

That's a new group name that I... just invented. Nothing uses this group, so if we try the test now:

php bin/phpunit --filter=testGetUser

It fails... but gets further! It fails on UserResourceTest line 68... it's failing down here. We successfully made phoneNumber not return when we fetch a user.

Next, we're going to create something called a "context builder", which will allow us to dynamically add this admin:read group when an "admin" is trying to normalize a User resource.