PMJSON

PMJSON provides a pure-Swift strongly-typed JSON encoder/decoder as well as a set of convenience methods for converting to/from Foundation objects and for decoding JSON structures.

The entire JSON encoder/decoder can be used without Foundation, by removing the files ObjectiveC.swift and DecimalNumber.swift from the project. The only dependency the rest of the project has is on Darwin, for strtod() and strtoll(). The file ObjectiveC.swift adds convenience methods for translating between JSON values and Foundation objects as well as decoding from a Data, and DecimalNumber.swift adds convenience accessors for converting values into NSDecimalNumber.

Before diving into the details, here's a simple example of writing a decoder for a struct. There are a few different options for how to deal with malformed data (e.g. whether to ignore values of wrong types, and whether to try and coerce non-string values to strings or vice versa), but the following example will be fairly strict and throw an error for incorrectly-typed values:

struct Address {
    var streetLine1: String
    var streetLine2: String?
    var city: String
    var state: String?
    var postalCode: String
    var country: String?

    init(json: JSON) throws {
        streetLine1 = try json.getString("street_line1")
        streetLine2 = try json.getStringOrNil("street_line2")
        city = try json.getString("city")
        state = try json.getStringOrNil("state")
        postalCode = try json.toString("postal_code") // coerce numbers to strings
        country = try json.getStringOrNil("country")
    }
}

And here's an example of decoding a nested array of values:

struct Person {
    var firstName: String
    var lastName: String? // some people don't have last names
    var age: Int
    var addresses: [Address]

    init(json: JSON) throws {
        firstName = try json.getString("firstName")
        lastName = try json.getStringOrNil("lastName")
        age = try json.getInt("age")
        addresses = try json.mapArray("addresses", Address.init(json:))
    }
}

If you don't want to deal with errors and just want to handle optionals, you can do that too:

struct Config {
    var name: String?
    var doThatThing: Bool
    var maxRetries: Int
    
    init(json: JSON) {
        name = json["name"]?.string
        doThatThing = json["doThatThing"]?.bool ?? false
        maxRetries = json["maxRetries"]?.int ?? 10
    }
}

This library also provides support for Swift.Encoder and Swift.Decoder. See this section for details.

The JSON decoder is split into separate parser and decoder stages. The parser consums any sequence of unicode scalars, and produces a sequence of JSON "events" (similar to a SAX XML parser). The decoder accepts a sequence of JSON events and produces a JSON value. This architecture is designed such that you can use just the parser alone in order to decode directly to your own data structures and bypass the JSON representation entirely if desired. However, most clients are expected to use both components, and this is exposed via a simple method JSON.decode(_:options:).

Parsing a JSON string into a JSON value is as simple as:

let json = try JSON.decode(jsonString)

Any errors in the JSON parser are represented as JSONParserError values and are thrown from the decode() method. The error contains the precise line and column of the error, and a code that describes the problem.

A convenience method is also provided for decoding from a Data containing data encoded as UTF-8, UTF-16, or UTF-32:

let json = try JSON.decode(data)

Encoding a JSON value is also simple:

let jsonString = JSON.encodeAsString(json)

You can also encode directly to any TextOutputStream:

JSON.encode(json, toStream: &output)

And, again, a convenience method is provided for working with Data:

let data = JSON.encodeAsData(json)

PMJSON supports parsing JSON streams, which are multiple top-level JSON values with optional whitespace delimiters (such as {"a": 1}{"a": 2}). The easiest way to use this is with JSON.decodeStream(_:) which returns a lazy sequence of JSONStreamValues, which contain either a JSON value or a JSONParserError error. You can also use JSONParsers and JSONDecoders directly for more fine-grained control over streaming.

As mentioned above, the JSON decoder is split into separate parser and decoder stages. JSONParser is the parser stage, and it wraps any sequence of UnicodeScalars, and itself is a sequence of JSONEvents. A JSONEvent is a single step of JSON parsing, such as .objectStart when a { is encountered, or .stringValue(_) when a "string" is encountered. You can use JSONParser directly to emit a stream of events if you want to do any kind of lazy processing of JSON (such as if you're dealing with a single massive JSON blob and don't want to decode the whole thing into memory at once).

Similarly, JSONDecoder is the decoder stage. It wraps a sequence of JSONEvents, and decodes that sequence into a proper JSON value. The wrapped sequence must also conform to a separate protocol JSONEventIterator that provides line/column information, which are used when emitting errors. You can use JSONDecoder directly if you want to wrap a sequence of events other than JSONParser, or if you want a different interface to JSON stream decoding than JSONStreamDecoder provides.

Because of this split nature, you can easily provide your own event stream, or your own decoding stage. Or you can do things like wrap JSONParser in an adaptor that modfiies the events before passing them to the decoder (which may be more efficient than converting the resulting JSON value).

Besides encoding/decoding, this library also provides a comprehensive suite of accessors for getting data out of JSON values. There are 4 types of basic accessors provided:

1. Basic property accessors named after types such as .string. These accessors return the underlying value if it matches the type, or nil if the value is not the right type. For example, .string returns String?. These accessors do not convert between types, e.g. JSON.Int64(42).string returns nil.
2. Property accessors beginning with the word as, such as .asString. These accessors also return an optional value, but they convert between types if it makes sense to do so. For example, JSON.Int64(42).asString returns "42".
3. Methods beginnning with get, such as getString(). These methods return non-optional values, and throw JSONErrors if the value's type does not match. These methods do not convert between types, e.g. try JSON.Int64(42).getString() throws an error. For every method of this type, there's also a variant ending in OrNil, such as getStringOrNil(), which does return an optional. These methods only return nil if the value is null, otherwise they throw an error.
4. Methods beginning with to, such as toString(). These are just like the get methods except they convert between types when appropriate, using the same rules that the as methods do, e.g. try JSON.Int64(42).toString() returns "42". Like the get methods, there are also variants ending in OrNil.

JSON also provides both keyed and indexed subscript operators that return a JSON?, and are always safe to call (even with out-of-bounds indexes). And it provides 2 kinds of subscripting accessors:

1. For every basic get accessor, there's a variant that takes a key or an index. These are equivalent to subscripting the receiver and invoking the get accessor on the result, except they produce better errors (and they handle missing keys/out-of-bounds indexes properly). For example, getString("key") or getString(index). The OrNil variants also return nil if the key doesn't exist or the index is out-of-bounds.
2. Similarly, there are subscripting equivalents for the to accessors as well.

And finally, the getObject() and getArray() accessors provide variants that take a closure. These variants are recommended over the basic accessors as they produce better errors. For example, given the following JSON:

{
  "object": {
    "elements": [
      {
        "name": null
      }
    ]
  }
}

And the following code:

try json.getObject("object").getArray("elements").getObject(0).getString("name")

The error thrown by this code will have the description "name: expected string, found null".

But given the following equivalent code:

try json.getObject("object", { try $0.getArray("elements", { try $0.getObject(0, { try $0.getString("name") }) }) })

The error thrown by this code will have the description "object.elements[0].name: expected string, found null".

All of these accessors are also available on the JSONObject type (which is the type that represents an object).

The last code snippet above looks very verbose, but in practice you don't end up writing code like that. Instead you'll often end up just writing things like

try json.mapArray("elements", Element.init(json:))

The JSON type has static methods map(), flatMap(), and compactMap() for working with arrays (since PMJSON does not define its own array type). The benefit of using these methods over using the equivalent SequenceType methods is the PMJSON static methods produce better errors.

There are also helpers for converting to/from Foundation objects. JSON offers an initializer init(ns: Any) throws that converts from any JSON-compatible object to a JSON. JSON and JSONObject both offer the property .ns, which returns a Foundation object equivalent to the JSON, and .nsNoNull which does the same but omits any null values instead of using NSNull.

The JSON type conforms to Codable, so it can be encoded to a Swift.Encoder and decoded from a Swift.Decoder. This has been tested against the standard library-provided JSONEncoder and JSONDecoder. Due to limitations in the decoding protocol, decoding a JSON must attempt to decode multiple different types of values, so it's possible that a poorly-written Swift.Decoder may produce surprising results when decoding a JSON.

Encoding to a JSON.Encoder and decoding from a JSON.Decoder is optimized to avoid unnecessary work.

This library provides an implementation of Swift.Encoder called JSON.Encoder. This can encode any Encodable to a JSON, a String, or a Data. It's used similarly to Swift.JSONEncoder (except at this time it doesn't have options to control encoding of specific types).

This library provides an implementation of Swift.Decoder called JSON.Decoder. This can decode any Decodable from a JSON, a String, or a Data. It's used similar to Swift.JSONDecoder (except at this time it doesn't have options to control decoding of specific types).

The test suite includes some basic performance tests. Decoding ~70KiB of JSON using PMJSON takes about 2.5-3x the time that NSJSONSerialization does, though I haven't tested this with different distributions of inputs and it's possible this performance is specific to the characteristics of the test input. However, encoding the same JSON back to a Data is actually faster with PMJSON, taking around 75% of the time that NSJSONSerialization does. These benchmarks were performed with Swift 2.x and it's possible the numbers have changed since then.

Installing as a framework requires a minimum of iOS 8, OS X 10.9, watchOS 2.0, or tvOS 9.0.

After installing with any mechanism, you can use this by adding import PMJSON to your code.

The Swift Package Manager may be used to install PMJSON by adding it to your dependencies list:

let package = Package(
    name: "YourPackage",
    dependencies: [
        .package(url: "https://github.com/postmates/PMJSON.git", from: "3.0.1")
    ]
)

To install using Carthage, add the following to your Cartfile:

github "postmates/PMJSON" ~> 3.0

This release supports Swift 4. If you want Swift 3.x support, you can use

github "postmates/PMJSON" ~> 2.0

To install using CocoaPods, add the following to your Podfile:

pod 'PMJSON', '~> 3.0'

This release supports Swift 4. If you want Swift 3.x support, you can use

pod 'PMJSON', '~> 2.0'

Licensed under either of

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT) at your option.

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you shall be dual licensed as above, without any additional terms or conditions.

• Update to Swift 5.
• When encoding/decoding URLs with JSON.Encoder and JSON.Decoder, encode and decode their absolute string instead of relying on the native implementation which encodes them as an object. This matches the behavior of JSONEncoder and JSONDecoder.
• Fix availability attribute for JSON.Encoder.DateEncodingStrategy.iso8601WithFractionalSeconds.
• Bump JSON.Encoder.DateEncodingStrategy.iso8601WithFractionalSeconds and JSON.Encoder.DateEncodingStrategy.iso8601WithFractionalSeconds to iOS 11.2+ and tvOS 11.2+ as, despite the constant being marked as available earlier, it's not supported at runtime. (#33)
• Convert JSONObject.ns and JSONObject.nsNoNull to return a [String: Any] instead of an [AnyHashable: Any]. (#25)
• Split JSON.Encoder.encodeAs* and JSON.Decoder.decode methods into overload pairs where one takes options: and the other doesn't. This makes it easier to replace function references to JSONEncoder/JSONDecoder methods with the equivalents from PMJSON.
• Add conformance to Combine's TopLevelEncoder and TopLevelDecoder, using Data as the input/output type. This means that JSON.Encoder.encode(_:) is now marked as deprecated instead of unavailable.
• Rename JSON.flatMap* and JSONObject.flatMap* methods to .compactMap* instead when the transformation returns an optional. (#28)
• Mark a lot of methods as @inlinable.

• Add method JSONError.withPrefix(_:) that returns a new error by prepending a prefix onto the path. This can be used in custom parsing code to produce good errors if the existing convenience functions don't do what you want. (#26)

• Squelch Swift 4.1 warnings.

• Improve .pretty output for empty arrays/dictionaries.
• Speed up JSON.encodeAsData() pretty significantly. It's now very nearly as fast as JSON.encodeAsString().
• Speed up JSON.Encoder.encodeAsString() and JSON.Encoder.encodeAsData().
• Add a couple of convenience static methods to JSON that mimic the enum cases: JSON.int(_:) and JSON.cgFloat(_:). These can be used when JSON(_:) triggers too much type complexity. Also add a JSON(_:) override for CGFloat.
• Add JSON.Encoder.keyEncodingStrategy. This is very similar to Swift 4.1's JSONEncoder.keyEncodingStrategy, although by default it won't apply to any nested values of type JSON or JSONObject (there's another option applyKeyEncodingStrategyToJSONObject that controls this).
• Add JSON.Decoder.keyDecodingStrategy. This is very similar to Swift 4.1's JSONDecoder.keyDecodingStrategy, although by default it won't apply to decoding any values of type JSON or JSONObject (there's another option applyKeyDecodingStrategyToJSONObject that controls this).
• Add JSON.Encoder.dateEncodingStrategy. This is very similar to JSONEncoder.dateEncodingStrategy except it includes another case for encoding ISO8601-formatted dates with fractional seconds (on Apple platforms).
• Add JSON.Decoder.dateDecodingStrategy. This is very similar to JSONDecoder.dateDecodingStrategy except it includes another case for decoding ISO8601-formatted dates with fractional seconds (on Apple platforms).
• Add JSON.Encoder.dataEncodingStrategy. This is identical to JSONEncoder.dataEncodingStrategy.
• Add JSON.Decoder.dataDecodingStrategy. This is identical to JSONDecoder.dataDecodingStrategy.

• Add convenience property JSONError.path.
• Add method JSONError.withPrefixedCodingPath(_:) to make it easier to use JSONError-throwing methods in a Decodable implementation.

• Fix Swift Package Manager support.

• Convert to Swift 4.
• Implement Codable on JSON.
• Add a Swift.Decoder implementation called JSON.Decoder.
• Add a Swift.Encoder implementation called JSON.Encoder.

• Add Linux support for Decimal (on Swift 3.1 and later). NOTE: Decimal support is still buggy in Swift 3.1, and the workarounds we employ to get the correct values on Apple platforms don't work on Linux. You probably shouldn't rely on this working correctly on Linux until Swift fixes its Decimal implementation.
• Add Linux support for decoding from/encoding to Data.
• Add Linux support for LocalizedError on the Error types (only really applies to Swift 3.1 and later).
• Fix compilation on Linux using the release configuration.
• Support running the test suite with swift test.

• Fix Linux compatibility.

• Add method JSON.parser(for:options:) that returns a JSONParser<AnySequence<UnicodeScalar>> from a Data. Like JSON.decode(_:options:), this method automatically detects UTF-8, UTF-16, or UTF-32 input.
• Fix compatibility with Swift Package Manager.

• Add full support for decimal numbers (on supported platforms). This takes the form of a new JSON variant .decimal, any relevant accessors, and full parsing/decoding support with the new option .useDecimals. With this option, any number that would have been decoded as a Double will be decoded as a Decimal instead.
• Add a set of forEach accessors for working with arrays, similar to the existing map and flatMap accessors.

• Handle UTF-32 input.
• Detect UTF-16 and UTF-32 input without a BOM.
• Fix bug where we weren't passing decoder options through for UTF-16 input.

• Change how options are provided to the encoder/decoder/parser. All options are now provided in the form of a struct that uses array literal syntax (similar to OptionSets). The old methods that take strict/pretty flags are now marked as deprecated.
• Add a new depth limit option to the decoder, with a default of 10,000.
• Implement a new test suite based on JSONTestSuite.
• Fix a crash if the input stream contained a lone trail surrogate without a lead surrogate.
• Fix incorrect parsing of numbers of the form 1e-1 or 1e+1.
• When the strict option is specified, stop accepting numbers of the form 01 or -01.
• Add support for UTF-16 when decoding a Data that has a UTF-16 BOM.
• Skip a UTF-8 BOM if present when decoding a Data.

• Add Hashable to JSONEvent and JSONParserError.
• Make JSONParserError conform to CustomNSError for better Obj-C errors.
• Full JSON stream support. JSONParser and JSONDecoder can now both operate in streaming mode, a new type JSONStreamDecoder was added as a lazy sequence of JSON values, and a convenience method JSON.decodeStream(_:) was added.
• Rename JSONEventGenerator to JSONEventIterator and JSONParserGenerator to JSONParserIterator. The old names are available (but deprecated) for backwards compatibility.
• Add support for pattern matching with JSONParserError. It should now work just like any other error, allowing you to say e.g. if case JSONParserError.invalidSyntax = error { … }.

• Fix CocoaPods.

• Support Swift 3.0.
• Add setters for basic accessors so you can write code like json["foo"].object?["key"] = "bar".
• Provide a localized description for errors when bridged to NSError.
• Add support to JSONParser for streams of JSON values (e.g. "[1][2]").

• Add a set of convenience methods on JSON and JSONObject for mapping arrays returned by subscripting with a key or index: mapArray(_:_:), mapArrayOrNil(_:_:), flatMapArray(_:_:), and flatMapArrayOrNil(_:_:).
• Add new set of convenience JSON initializers.
• Change description and debugDescription for JSON and JSONObject to be more useful. description is now the JSON-encoded string.
• Implement CustomReflectable for JSON and JSONObject.

• CocoaPods support.

• Linux support.
• Swift Package Manager support.
• Rename instances of plist in the API to ns. The old names are still available but marked as deprecated.
• Support the latest Swift snapshot (2012-02-08).

Initial release.