DeepCodable: Encode and decode deeply-nested data into flat Swift objects

Have you ever gotten a response from an API that looked like this and wanted to pull out and flatten the values you care about? (This is a real response from the GitHub GraphQL API, with only the actual values changed)

{
	"data": {
		"node": {
			"content": {
				"__typename": "Example type",
				"title": "Example title"
			},
			"fieldValues": {
				"nodes": [
					{},
					{},
					{
						"name": "Example node name",
						"field": {
							"name": "Example field name"
						}
					}
				]
			}
		}
	}
}

DeepCodable lets you easily do so in Swift while maintaining type-safety, with the magic of result builders, key paths, and property wrappers:

import DeepCodable

struct GithubGraphqlResponse: DeepDecodable {
	static let codingTree = CodingTree {
		Key("data") {
			Key("node") {
				Key("content") {
					Key("__typename", containing: \._type)
					Key("title", containing: \._title)
				}

				Key("fieldValues") {
					Key("nodes", containing: \._nodes)
				}
			}
		}
	}


	struct Node: DeepDecodable {
		static let codingTree = CodingTree {
			Key("name", containing: \._name)

			Key("field", "name", containing: \._fieldName)
			/*
			The above is a "flattened" shortcut for:
			Key("field") {
				Key("name", containing: \._fieldName)
			}
			*/
		}


		@Value var name: String?
		@Value var fieldName: String?
	}

	enum TypeName: String, Decodable {
		case example = "Example type"
	}

	@Value var title: String
	@Value var nodes: [Node]
	@Value var type: TypeName
}

dump(try JSONDecoder().decode(GithubGraphqlResponse.self, from: jsonData))

Add to your Package.Swift:

...
	dependencies: [
		...
		.package(url: "https://github.com/MPLew-is/deep-codable", branch: "main"),
	],
	targets: [
		...
		.target(
			...
			dependencies: [
				...
				.product(name: "DeepCodable", package: "deep-codable"),
			]
		),
		...
	]
]

Conform a type you want to decode to DeepDecodable by defining a coding tree representing which nodes are bound to which values:

struct DeeplyNestedResponse: DeepDecodable {
	static let codingTree = CodingTree {
		Key("topLevel") {
			Key("secondLevel") {
				Key("thirdLevel", containing: \._property)
			}
		}
	}
	/*
	Also valid is the flattened form:
	static let codingTree = CodingTree {
		Key("topLevel", "secondLevel", "thirdLevel", containing: \._property)
	}
	*/

	@Value var property: String
}
/*
Corresponding JSON would look like:
{
	"topLevel": {
		"secondLevel": {
			"thirdLevel: "{some value}"
		}
	}
}
*/

Nodes in your codingTree are made of Keys initialized one of the following ways:

• Key("name") { /* More Keys */ }: node that don't capture values directly, but contain other nodes

  • This maps to a serialized representation like {"name": { ... } }

• Key("name", containing: \._value): node that should be decoded into the value property

All values to decode must be wrapped with the @Value property wrapper, and the \._{name} syntax refers directly to the wrapping instance (\.{name} without the underscore refers to the actual underlying value).

Decode a value into an instance of your type:

let instance = try JSONDecoder().decode(Response.self, from: jsonData)

DeepCodable is built on top of normal Codable, so any decoder (like the property list decoder in Foundation or the excellent third-party YAML decoder, Yams) can be used to decode values.

While decoding is probably the most common use-case for this type of nested decoding, this package also supports encoding a flat Swift struct into a deeply nested one with the same pattern:

struct DeeplyNestedRequest: DeepEncodable {
	static let codingTree = CodingTree {
		Key("topLevel") {
			Key("secondLevel") {
				Key("thirdLevel", containing: \.bareProperty)
			}

			Key("otherSecondLevel", containing: \._wrappedProperty)
		}
	}
	/*
	Also valid is the flattened form:
	static let codingTree = CodingTree {
		Key("topLevel") {
			Key("secondLevel", "thirdLevel", containing: \.bareProperty)

			Key("otherSecondLevel", containing: \._wrappedProperty)
		}
	}
	*/

	let bareProperty: String
	@Value var wrappedProperty: String
}
/*
Corresponding JSON would look like:
{
	"topLevel": {
		"secondLevel": {
			"thirdLevel: "{bareProperty}"
		},
		"otherSecondLevel": "{wrappedProperty}"
	}
}
*/

let instance: DeeplyNestedRequest = ...
let jsonData = try JSONEncoder().encode(instance)

With encoding, you don't have to use the @Value wrappers, though you can if you'd like to support decoding and encoding on the same type (in which case you can conform to DeepCodable as an alias for the two).

• Encoding and decoding a Swift object to/from an arbitrarily complex deeply nested serialized representation without manually writing Codable implementations

• Preservation of existing Codable behavior on the values being encoded/decoded, including custom types

  • Since DeepCodable is just a custom implementation of the Codable requirements, this also means you can nest DeepCodable objects like in the GithubGraphqlResponse example

• When conforming to DeepEncodable or DeepDecodable, don't interfere with the opposite normal Codable implementation (Decodable/Encodable, respectively)

  • You can declare something like struct Response: DeepDecodable, Encodable { ... } and decode from a deeply nested tree, and then re-encode back to a flat structure like normal Encodable objects

• No requirement for @Value property wrapper for types only conforming to DeepEncodable

• Omission of the corresponding tree sections when all values at the leaves are nil

  • This makes it so trying to encode an object with a nil value doesn't result in something like {"top": {"second": {"third": null} } }

• Flattened shortcuts using variadic parameters for long paths with no branching:

  • Key("topLevel", "secondLevel", containing: \._property) instead of Key("topLevel") { Key("secondLevel", containing: \._property) }