Migration from Vapor 1 to Vapor 2

Vapor 2 is released 🚀🎉 This is a big breaking change, but a good one. We have a small server-side project which is developed with Vapor. When Vapor 2 was released, we directly took a look at what changed. We were a bit 😍 and a bit😩. Vapor 2 has fewer lines of code, the performance is better, and has a lot of new cool features. But on the other side, there is too much change that we had to adopt. As a beginner to the server side and Vapor, it took quite some time. We took some notes while migrating to version 2. First, let me introduce the basic components that I’ll talk about. We’re using MongoDB as a database. We have two custom middlewares which are for verifying requests and setting up response headers. We use custom Routing with RouteCollection. That’s all.

Let’s start migrating! 💪🏽

We started with the Package.swift file. We were using Vapor 1.5.0 and VaporMongo 1.1.0

1
dependencies: [
2
  .Package(url: "https://github.com/vapor/vapor.git", majorVersion: 1, minor: 5),
3
  .Package(url: "https://github.com/vapor/mongo-provider.git", majorVersion: 1, minor: 1)
4
]

Now, we’re using Vapor 2.0 and MongoProvider 2.0.0 (was named VaporMongo before). MongoProvider comes with FluentProvider which is the new package. “The Fluent provider package adds Fluent to project and adds some additional, Vapor-specific conveniences like HTTP conformances.”

1
dependencies: [
2
  .Package(url: "https://github.com/vapor/vapor.git", majorVersion: 2),
3
  .Package(url: "https://github.com/vapor/mongo-provider.git", majorVersion: 2)
4
]

Then the moment came. Ran vapor clean && vapor build command and BAM! Tons of errors. 😩 There are a lot of things to do.

Model

The Model protocol has moved to the FluentProvider package. We had to import FluentProvider to our Model files. In Model, query() method was changed to makeQuery(). So, we were using User.query() and we changed it to User.makeQuery() to run custom queries on the User model. The other entity methods like all, find, chunk, count are still the same.

Fluent models now require Storable protocol conformance for database operations. This protocol requires only one property that handles id and exists properties’ jobs. We don’t need to handle these properties manually anymore. We put let storage = Storage() as a property of every model. Instead of Node struct and NodeConvertible protocol, we now have Row struct and RowConvertible protocol. It has two new methods one for initializing the model and the other for making a row. Even though I don’t get the idea of a Row in databases like MongoDB, the implementation looks better now. So, we had NodeConvertible implementations before like this:

1
init(node: Node, in context: Context) throws {
2
  id = try node.extract("_id") // Trick for MongoDB database
3
  name = try node.extract("name")
4
  score = try node.extract("score")
5
}
6

7
func makeNode(context: Context) throws -> Node {
8
  return try Node(node: [
9
    "_id": id, // Trick for MongoDB database
10
    "name": name,
11
    "score": score
12
    ])
13
}

We migrated to RowConvertible and changed our code to following:

1
init(row: Row) throws {
2
  name = try row.get("name"")"
3
  score = try row.get("score")
4
}
5

6
func makeRow() throws -> Row {
7
  var row = Row()
8
  try row.set("name", name)
9
  try row.set("score", score)
10
  return row
11
}

Lastly, we need to send ids in the response body. We had to conform JSONRepresentable protocol and create custom JSON for our models.

1
extension User: JSONRepresentable {
2
  func makeJSON() throws -> JSON {
3
    var json = JSON()
4
    try json.set("id", id?.string)
5
    try json.set("name", name)
6
    try json.set("score", score)
7
    return json
8
  }
9
}

Note: If you still need NodeConvertible, NodeRepresentable protocol’s function syntax has been changed.

func makeNode(context: Context) throws -> Node migrated to this: func makeNode(in context: Context?) throws -> Node

RouteCollection protocol

The only change in RouterCollection protocol is the method definition.

1
class OurCoolCollection: RouteCollection {
2
  typealias Wrapped = HTTP.Responder
3
  func build<Builder: RouteBuilder>(_ builder: Builder) where Builder.Value == Responder { ... }
4
}

migrated to

1
class OurCoolCollection: RouteCollection {
2
  func build(_ builder: RouteBuilder) throws { ... }
3
}

Also drop.collection(OurCoolCollection()) is now throwable.

Request Parameters

Because of technical limitations, we had a limit for the request parameters before. Now, there is no limit because the implementation is totally changed. There is a new Parameterizable protocol. Anything that conforms to Entity protocol, can conform Parameterizable without any extra work. So, every Fluent model can be Parameterizable easily with only conforming like extension User: Parameterizable { }

. In our custom collection, the implementation of endpoints was like this:

1
let users = builder.grouped("user")
2
let userController = UserController()
3
users.post("", User.self, handler: userController.create)

with only one change, it became like this

1
let users = builder.grouped("user")
2
let userController = UserController()
3
users.post("", handler: userController.create)

And the implementation of create function in the controller was like this:

1
func create(request: Request, user: User) throws -> ResponseRepresentable

Instead of the function parameters, we can get all request parameters directly from the request. So, the syntax of the function became like following

1
func create(request: Request) throws -> ResponseRepresentable {
2
  let user = try request.parameters.next(User.self)
3
}

String and Int data types automatically conform Parameterizable protocol. It means that we can use them as parameters. We also have an endpoint that only requires the id of the User as a String in the request. In Vapor 2, the new implementation is like this:

1
func someFunction(request: Request) throws -> ResponseRepresentable {
2
  let id = try request.parameters.next(String.self)
3
}

and in the get function, we just called .parameter property of Parameterizable protocol.

1
users.get("someEndpoint", String.parameter,
2
          handler: userController.someFunction)

Droplet & Configuration

All of the Droplet parameters are now constants. Changing them after initialization is not allowed anymore. Vapor 2 has a new Config concept in which we can set up our Droplet’s configurations like middlewares, providers, preparations, etc. Droplet has two initialization methods. We used init(_ config: Config). Creating the configuration and setting it up separately looks cleaner.

Our Droplet was written in Vapor 1 and it was something like that:

1
let drop = Droplet()
2
drop.middleware.insert(try CORSMiddleware(configuration: drop.config), at: 0)
3
drop.middleware.append(DummyMiddleware())
4
drop.middleware.append(SomeMiddleware())
5
drop.collection(OurCoolCollection())
6
try drop.addProvider(VaporMongo.Provider.self)
7
drop.preparations.append(User.self)

Now, we first created the config with

1
let config = try Config()

next, we added middleware to config

1
config.addConfigurable(middleware: DummyMiddleware(), name: "dummy")
2
config.addConfigurable(middleware: SomeMiddleware(), name: "some")

lastly, we added these middlewares to our droplet.json file. Vapor 2 puts the middlewares in order according to droplet.json. For example, if you want your AwesomeMiddleware to run at first, add it to the top of the middlewares array at the droplet.json file. After adding the middlewares to the droplet.json, it became like this

1
{
2
  "server": "engine",
3
  "client": "engine",
4
  "console": "terminal",
5
  "log": "console",
6
  "hash": "crypto",
7
  "cipher": "crypto",
8
  "middleware": [
9
    "cors",
10
    "abort",
11
    "type-safe",
12
    "date",
13
    "dummy",
14
    "some"
15
  ]
1 collapsed line
16
}

As mentioned before, we now have the new FluentProvider. We had to add fluent.json for under the Config folder. Otherwise, it crashes the app when we run it. Vapor allows you to define the idType for the database as UUID format in JSON file instead of creating manually in every model. Also, we have to define our driver in here. We’re using MongoProvider and because of that, our database driver is mongo. We added this provider with only one line

1
try config.addProvider(MongoProvider.Provider.self)

and we defined a fluent.json file like

1
{
2
  "idType": "uuid",
3
  "driver": "mongo"
4
}

Let’s look at preparations quickly. There is no extra work like Provider. We just added them to the config.

config.preparations.append(User.self)

Now, we can initialize our Droplet with the configuration in one line, that’s all.

let drop = try Droplet(config)

Our Config is not done yet. It seems hash and cipher have mandatory keys right now. So, we had to change our crypto.json file. Both of them now require three keys, method, encoding, and key. Our crypto.json file was like a dummy one before

1
{
2
  "hash": {
3
    "method": "sha256",
4
    "key": "password"
5
  },
6
  "cipher": {
7
    "method": "chacha20",
8
    "key": "passwordpasswordpasswordpassword",
9
    "iv": "password"
10
  }
11
}

Now, it looks smarter

1
{
2
  "hash": {
3
    "method": "sha256",
4
    "encoding": "hex",
5
    "key": "FjavUZaqfjjMg4q6722XXY3H4Yvmm39YorHQJ2uslPU"=
6
  },
7
  "cipher": {
8
    "method": "aes256",
9
    "key": "pHhMTJkxkykYn6cJu8fmtVUHkIx/61BTM5tYEwPsYa0=",
10
    "encoding": "base64",
11
  }
12
}

The last thing for the config was the url. It’s mandatory in mongo.json for MongoProvider configuration. Even if we gave host, port, username, password, and database parameters in the JSON file, it insistently asked for the url key. So, we just added url key to mongo.json file in the following format. The key point is, MongoKitten requires this url in a certain format. It has to start with mongodb://.

"url": "mongodb://user:password@127.0.0.1:27017/test_db"

Abort

Abort is not an enum anymore. It has converted to a struct with three parameters.

1
var status: Status { get }
2
var reason: String { get }
3
var metadata: Node? { get }

We were using the Abort like following

1
throw Abort.custom(status: .badRequest,
2
                   message: "Why is this error happening? Because: \(error)")
3
throw Abort.notFound

and it is migrated to cleaner implementation

1
throw Abort(.badRequest,
2
            reason: "Why is this error happening? Because: \(error)")
3
throw Abort(.notFound)

Update:

servers.json file is simplified and renamed to server.json. Basic server.json can be found in the documentation like this:

1
{
2
  "//": "The $PORT:8080 call tells the json file to see if there",
3
  "//": "is any value at the 'PORT' environment variable.",
4
  "//": "If there is no value there, it will fallback to '8080'",
5
  "port": "$PORT:8080",
6

7
  "host": "0.0.0.0",
8

9
  "//": "It's very rare that a server manages its own TLS.",
10
  "//": "More commonly, vapor is served behind a proxy like nginx.",
11
  "securityLayer": "none"
12
}

Last words

Adoption to the latest versions with breaking changes is generally hard. But when it comes to performance, clean code, and other new features, walking around with the old version becomes redundant. Some people declare this as living on the edge. Although I agree with this on some level, I generally try to use the latest versions when they came out. But I always give it a try with some trial projects, not directly on production code.

Further: