An attempt to gather all that is in flux in Swift.
Latest commit 9a9d7e4 Dec 4, 2015 @jklausa jklausa Merge pull request #102 from ksm/obsoletion
Obsoletion notice and link to Swift Evolution.
Permalink
Failed to load latest commit information.
README.md Obsoletion notice and link to Swift Evolution. Dec 3, 2015

README.md

This document is no longer maintained.

With Swift becoming open-source - and the Swift Evolution document being public, this project was made mostly obsolete. However, it still serves as a historical account of Swift's initial evolution.

Swift InFlux

The community is creating some incredible analyses and writing about Swift. What I keep asking myself whenever learning and reading about Swift is: how likely is this to change soon?

This document is an attempt to gather the Swift features that are still in flux and likely to change.

Contributing

To contribute: fork this project, add a section below (don't forget to update the Table of Contents!), and create a pull request.

Credits

Swift InFlux was created by Karol S. Mazur during SwiftCrunch. It is maintained by the creator, and two great contributors: Radek Pietruszewski and Jan Klausa.

Table of Contents





Abstract methods

FWIW, we already have many bugs tracking the idea of adding abstract methods to swift. We'll consider it in future releases, we understand the value :-)

— Chris Lattner

Source: https://devforums.apple.com/message/1006592#1006592

Limitations of current access control design

Swift added access control in Xcode 6.0 Beta 4, but there are limitations to its design. Entities marked as internal (which is the default) are not visible to other build targets, which means that unit tests (which traditionally live in a separate target) can't access them.

We're aware that our access control design isn't great for unit testing (and this was in the release notes), we're evaluating the situation to see what we can do.

— Chris Lattner

A limitation of the access control system is that unit tests cannot interact with the classes and methods in an application unless they are marked public. This is because the unit test target is not part of the application module.

— Xcode 6.0 beta 4 release notes

At the moment, one workaround is to mark all tested entities as public, another is to move tests to the same target as the application code. However, the former ruins the benefits of access control and the latter — of code modularization.

Sources: https://devforums.apple.com/message/1010766#1010766 Xcode 6.0 Beta 4 release notes

ABI stability

Swift currently offers no ABI stability for any code and provides no ABI non-fragility for framework authors.

— Greg Parker

gparker [Greg Parker] is right, and this is important.

Swift 1.0 is not guaranteed to be binary compatible with "swift 2.0" (or whatever). We guarantee that compiled and distributed apps and their frameworks are binary compatible with the OS and will continue to work well into the future, but a Swift framework today will not work in the future.

We expect to lock this down, but not by the final release of swift this Fall. This means that all swift code in an app (including frameworks it uses) should be built by the same version of Xcode.

— Chris Lattner

Sources: https://devforums.apple.com/message/986618#986618 https://devforums.apple.com/message/989931#989931

Better error handling features (possibly exceptions)

We're aware of the opportunity and also desire better error handling features in Swift, but they didn't make it in time for 1.0.

— Chris Lattner

Source: https://devforums.apple.com/thread/228324?start=50&tstart=0

Class variables

At the moment, stored class properties are not supported in Swift. They will come in a later release, but not in 1.0. Possibly, the main reason for this is that Swift largely depends on Objective-C runtime and implementation details, which don't support class variables.

Class variables not yet supported.

— Swift compiler error

The feature set for 1.0 is nearly final. 'yet' should not be taken to mean Swift 1.0.

— Chris Lattner

Swift in Xcode 6 will not support class variables.

— Greg Parker

Sources: https://devforums.apple.com/message/1022374#1022374 https://devforums.apple.com/message/1030167#1030167

C++ support

This is another obviously desirable feature, it is just a lot of work and didn't make it in 1.0 either.

— Chris Lattner

Source: https://devforums.apple.com/thread/228324?start=50&tstart=0

Dynamic dispatch of operators

FWIW, we're not happy with this either. Among other things, we've seen confusion where people define something like:

 func ==(lhs : MyBaseClass, rhs : MyBaseClass) {...}
 func ==(lhs : MyDerivedClass, rhs : MyDerivedClass) {...}

... and are surprised when they don't get dynamic dispatch.

We have not had a chance to fully revisit this, but it is very likely that we'll allow operators to be defined inside of types (i.e. as methods) as well as at global scope (necessary for mixed type operators). When defined as a non-final member of a class, these operators would be dynamically dispatched just like any other method.

— Chris Lattner

Source: https://devforums.apple.com/message/1074064#1074064

Enumerating enum types

Does anyone else think this would be fundamentally useful? Or is their a good way of apporaoching this in Swift currently that I'm missing?

Yes. All of this would be super useful. We have a large number of radars asking for similar functionality, thanks!

— Chris Lattner

Source: https://devforums.apple.com/message/1003674#1003674

Expanding scope of imported macros

Swift currently only imports simple constant-like macros from C and Objective-C. This leaves many key macros unaccessible to Swift, requiring redefinition. For example, the CPU_TYPE constant type cast macros in mach/machine.h.

#define CPU_TYPE_X86          ((cpu_type_t) 7)

It's hard to nail Apple down as to what we will or won't do, but I think it's safe to say that improving the way that Swift imports C / Objective-C headers is an important goal.

— eskimo1

Indeed. Improving the experience using imported C/Objective-C APIs is a strong goal for ongoing swift evolution.

— Chris Lattner

Sources:

Flow-sensitive optional unwrapping

It has been suggested that optional types could be implicitly unwrapped in the context of an if-statement checking if an optional has a value, for example:

let x: Type?

if exists x {
   x.doSomething() // works without ? or !
}

We're definitely aware of the advantages of control-flow sensitive type refinement for optionals (and also for other subtype relationships). This would be particularly handy in ternary operators, but isn't something on the books for 1.0. We have several radars requesting that and tracking it for consideration in a future release.

— Chris Lattner

Flow-sensitive type refinement like this is something that may happen in a future release of Swift

— CFM

Source: https://devforums.apple.com/message/1005148#1005148 https://devforums.apple.com/message/1066436#1066436

Generic subscripts

Lack of generic subscripts is a known limitation. We'll look at improving this at some point when it bubbles up in the priority list.

— Chris Lattner

Currently, generic subscripts are allowed only for generic types (e.g. Array, Dictionary).

Source: https://devforums.apple.com/message/1100335#1100335

Imported constant macros carry explicit type

Since Swift does not allow implicit type conversion, imported constant C and Objective-C macros lose a bit of flexibility in use as they carry an explicit type (Swift has no preprocessor). For example, M_PI from math.h is imported as a double.

FWIW, we consider it to be a bug that M_PI (and a variety of other imported constants) get an arbitrary fixed type assigned to them. This affects integer constants just as much as floating point ones.

In principle, there could be a way to provide "typeless named literals" in the language, and constants imported from C macros could be imported like that. I don't know if that's the approach we'll take, but it is one of several different options we'll evaluate down the road to improve this situation.

— Chris Lattner

Source: https://devforums.apple.com/message/1032523#1032523

Moving functionality from global functions to methods

We'll have to evaluate this in more detail when we have time to explore the design space, but here are some thoughts:

We'd prefer to build out the language to allow more things to be methods, instead of more things as global functions. In general, methods are better than global functions because they work better for code completion, and scoped better, and more consistent with other methods.

We currently have limitations in the type system and implementation that force some things (e.g. countElements and many others) to be global functions instead of methods, but we consider this a deficiency, not a feature.

— Chris Lattner

The prevalence of generic free functions is more about current language limitations than an intentional stylistic direction

— CFM

Source: https://devforums.apple.com/message/1074064#1074064 https://devforums.apple.com/message/1094312#1094312

Open source

Swift will be going open source late 2015 once Apple has shipped the 2.0 GM version of the language (Fall). The details that Apple has provided on this so far are:

  • Source code for the Swift compiler and standard library will be released under an OSI-approved permissive license
  • Ports for OS X, iOS, and Linux with be provided
  • Contributions will be accepted

Guys, feel free to make up your own dragons if you want, but your speculation is just that: speculation. We literally have not even discussed this yet, because we have a ton of work to do to respond to the huge volume of feedback we're getting, and have to get a huge number of things (e.g. access control!) done before the 1.0 release this fall. You can imagine that many of us want it to be open source and part of llvm, but the discussion hasn't happened yet, and won't for some time.

Sorry to leave you all hanging, but there is just far too much to deal with right now.

— Chris Lattner

Sources:

Optionals in imported Objective-C frameworks

As of Xcode 6.3, relatively few APIs have been audited for optional conformance. There has been considerable progress since early Xcode 6 betas and more improvements are expected to come in the future.

Since Xcode 6.3, it is possible for users to annotate nullability (optionality) of Objective-C APIs.

A large number of AppKit APIs have been audited for optional conformance in addition to WebKit, Foundation, UIKit, CoreData, SceneKit, SpriteKit, and Metal APIs. As a result, a significant number of implicitly unwrapped optionals have been removed from their interfaces. These changes clarify the nullability of properties, arguments, and return values in the APIs. The audit effort is ongoing.

The API changes replace T! with either T? or T depending on whether or not the value can be null, respectively.

— Xcode 6.1 release notes

Sources: Xcode 6.1 release notes

Optional methods in pure-Swift protocols

Optional methods in protocols are limited to @objc protocols only because we haven't implemented them in native protocols yet. This is something we plan to support. We've gotten a number of requests for abstract/pure virtual classes and methods too.

— Joe Groff

Source: https://devforums.apple.com/message/1051431#1051431

Overriding declarations from extensions

At the moment, it's not possible to override entities declared in extension by subclassing, like so:

class Base { }

extension Base {
    var foo: String { return "foo" }
}

class Sub: Base {
    override var foo: String { return "FOO" } // This is an error
}

Note: it is possible to override entities from extensions if they are marked as @objc, but it's likely to be an accidental side-effect of Objective-C runtime implementation and it might break in the future.

Declarations from extensions cannot be overriden yet.

— Swift compiler error

The feature set for 1.0 is nearly final. 'yet' should not be taken to mean Swift 1.0.

— Chris Lattner

Source: https://devforums.apple.com/message/1022374#1022374

Passing initializers as functions

[Is there any] specific reason why you can't reference initializers like a function?

No particular reason, it's just not implemented.

— Joe Groff

Referencing initializers to pass as an argument would be useful for currying.

Source: https://twitter.com/jckarter/status/542032426298392576

Reflection

At the moment, Swift has only very limited reflection capabilities. In addition to built-in syntax for checking variable types, there's a reflect() function that can tell the names an types of object's properties. It is, however, poorly documented. It's unclear whether or not reflection capabilities will be expanded in future releases.

Further reading: Simple Reflection in Swift

Though it’s not documented in the Swift Standard Library Reference — and is subject to change, and could disappear entirely — Swift has a reflection API.

— Brent Simmons

Sources: http://inessential.com/2014/07/13/swift_reflection https://gist.github.com/peebsjs/9288f79322ed3119ece4

Runtime dynamic libraries

Products that use Swift need to come with their own copies of the swift runtime. This is currently necessary for a variety of reasons. Xcode directly supports this for the following scenarios:

  • When you build an application that uses Swift, Xcode will automatically embeds the swift runtime dynamic libraries inside the app wrapper in the Frameworks folder. Application targets should also be automatically configured with rpath entries that will locate them there.
  • When you build a command line tool that uses Swift, Xcode will statically link the swift runtime into the command line tool's binary
  • When you build other product types (e.g. frameworks or app extensions) that use Swift, they are set up to expect to find the swift libraries embedded in the application that they are themselves embedded in

— mferris

This is done for several reasons:

Swift's runtime and libraries are not (yet) included in the operating system, so must be included in each app.

— CFM

The runtime is bundled with your app in order to prevent incompatible language changes from affecting deployed apps. Changes to the language, runtime implementation, or how frameworks get imported won't affect your already-compiled app — any ABI-level dependencies are between the runtime dylibs in your bundle and the main executable.

— Joe Groff

Given this, you will hit issues if your are creating an OS X command line tool that uses a framework written in Swift.

In your case, the swift standard libraries are being statically linked into your command line tool. The problem is that the framework cannot see that, and it expects to find them in an app wrapper that it is inside of. Xcode currently has no direct support for a standalone framework using swift that is not embedded in an application.

— mferris

Sources:

Static libraries

Xcode does not support building static libraries that include Swift code.

— Xcode 6.0 beta 5 release notes

The current runtime doesn't ship with the OS, so static libs would lead to multiple runtimes in the final executable. A statically linked runtime would be much more difficult to patch for compatibility with newer OS or Swift.

— Joe Groff

Sources: Xcode 6.0 Beta 5 release notes https://twitter.com/jckarter/status/555061575623507969

switch and if as expressions

We're aware that 'switch' and 'if' are commonly expressions in functional languages, and that this brings a lot of convenience. We'll consider expanding this in future releases, but it isn't a short term priority for Swift 1.1.

-Chris

Source: https://devforums.apple.com/message/1040049#1040049

Systems Programming Features

The focus of Swift 1.0 is definitely on improving general app development, but we do expect Swift to grow capabilities (e.g. perhaps even the ability to write inline assembly code) that allow it to fully span the gamut of programming: from writing the lowest level firmware up to the highest level application programming. We prefer to do this carefully and deliberately over time, rather than attempting to solve all the world's problems at once.

— Chris Lattner

Source: https://devforums.apple.com/message/1007178#1007178


Changed in Xcode 6.3 Beta 3

Changed in Xcode 6.3 Beta 2

Further enhancements to if let

Following the changes in Beta 1, the if let syntax has been extended to allow a boolean condition preceding a let clause. For example:

if loggingEnabled, let samples = getLogSamples() where samples.count > 10 {
    sendLogSamples(samples)
}

Changed in Xcode 6.3 Beta 1

Xcode 6.3 brings a large number of changes, bug fixes and new features to Swift, which is now version 1.2.

More information:

Compiler improvements

According to the Swift team, the compiler got a lot of under-the-hood improvements, which give us:

  • Incremental builds — files that haven't changed won't have to be recompiled every time, speeding up the process
  • Faster executables — Swift binary code is now better optimized, both in Release and Debug modes.
  • Better diagnostics — Clearer warning and error messages, as well as new fix-its
  • Stability improvements — less compilation and SourceKit crashes.
    • according to swift-compiler-crashes, 83% of crashing bugs (4200 crashes) have been fixed in this beta

Nullability annotations in Objective-C

You can now mark method and function parameters, return types, properties and variables in Objective-C as non-nullable (imported to Swift as T) or nullable (imported as T?). In Objective-C declarations, you can use new nonnull and nullable qualifiers before the type, for example:

-(nullable UITableViewCell *)cellForRowAtIndexPath:(nonnull NSIndexPath)indexPath;

Same keywords can be used with Objective-C property configuration:

@property (nonatomic, readwrite, retain, nullable) UIView *backgroundView;

Arbitrary C pointers, block pointers and C++ member pointers can be marked using __nonnull and __nullable, like so:

void enumerateStrings(__nonnull CFStringRef (^ __nullable callback)(void));

This will be imported as:

func enumerateStrings(callback: (() -> CFString)?)

Enhancements to if let

The if let construct can now be used to optionally unwrap multiple optionals at once. You can also add a guarding condition with where:

if let a = foo(), b = bar() where a < b,
    let c = baz() {
}

Constants no longer require immediate initialization

In Xcode 6.2 and before, constants defined using let had to be immediately assigned a value. The new rule is that a constant must be initialized before use (and, of course, it must not be reassigned or mutated later).

This allows patterns like:

let x: SomeThing
if condition {
    x = foo()
} else {
    x = bar()
}
use(x)

In previous versions this would require using var, even though no mutation is taking place.

Importing Swift enums into Objective-C

Simple enums (not using generics or associated values) can now be exported to Objective-C by marking them as @objc and setting Int as the raw value. For instance:

@objc enum Bear: Int {
    case Black, Grizzly, Polar
}

imports into Objective-C as:

typedef NS_ENUM(NSInteger, Bear) {
    BearBlack, BearGrizzly, BearPolar
};

First-class Set type

The Swift standard library now includes a fully generic Set type that bridges to NSSet and has value semantics.

Xcode drops Mavericks support

Xcode 6.3 does require Yosemite, as mentioned in the release notes. This will not be changed in a later beta.

— Chris Lattner

Swift can still target Mavericks (10.9), but Xcode itself will no longer support it.

Source: https://devforums.apple.com/message/1101275#1101275


Changed in Xcode 6.2

Xcode 6.2 doesn't include any significant changes to Swift except for nullability audit of the WatchKit framework.

Yeah, 6.2's (almost) all about WatchKit.

— Joe Groff

Sources: https://devforums.apple.com/message/1074584#1074584 https://devforums.apple.com/message/1082641#1082641 https://twitter.com/jckarter/status/555059438944387072

Changed in Xcode 6.1.1

Xcode 6.1.1 was a maintenance release with only minor improvements and bug fixes to Swift, including:

  • Passing class objects for pure Swift class to AnyObject values no longer causes a crash
  • Class methods and initializers that satisfy protocol requirements now properly invoke subclass overrides when called in generic contexts
  • Some causes of SourceKit crashes have been fixed
  • Minor changes to nullability in Objective-C frameworks.

There have been no changes to the standard library header and no significant compiler crash fixes.

There have been no significant differences between the 6.1.1 GM seed and the final version.

Source: Xcode 6.1.1 Release Notes

Changed in Xcode 6.1 GM Seed 2 / 6.1

GM Seed 2 doesn't appear to have any developer-facing changes (no changes to the standard library), but it brings a number of bug fixes to the Swift compiler.

The final version of Xcode 6.1 doesn't seem to have any changes in Swift compiler or standard library since GM Seed 2, however there have been minor updates to optional conformance in iOS and OS X SDKs.

Changed in Xcode 6.1 Beta 3 / GM Seed 1

Beta 3 brought a number of standard library changes and compiler bug fixes, but most importantly, a huge amount of new documentation comments was added to the stdlib header.

Beta 3 was renamed to "GM seed" shortly after release to indicate that the OS X SDK is GM (and you can use it to ship Yosemite apps). It is not the final seed of Xcode 6.1.

The nomenclature is admittedly confusing here. I'd recommend that you ignore the label, and work with a functional description:

What you need to know:

  • Xcode 6.1b3 has been renamed on the website to GM Seed, to indicate that you can use it to build and submit apps to the Mac App Store, using the Yosemite SDK.
  • This beta includes the iOS 8.1 beta SDK, but you cannot submit apps to the iOS app store with this Xcode build.
  • Further updates of 6.1 are expected before the "final GM seed" is released on the Mac App Store.

Chris Lattner

LiteralConvertible protocols use constructor

Objects implementing the various LiteralConvertible protocols need to implement an init method instead of a class method.

For example:

protocol BooleanLiteralConvertible {
    typealias BooleanLiteralType
    init(booleanLiteral value: BooleanLiteralType)
}

Other changes in Xcode 6.1 Beta 3 / GM Seed 1

  • Any can now refer to functions.

Changed in Xcode 6.1 Beta 2

Failable initializers in Objective-C frameworks

Objective-C init and factory methods are now imported as failable initializers to explicitly signal that they might return nil. Most initializers are automatically imported as init! (subject to auditing for optional conformance in later releases), however methods that take a NSError** parameter are always imported as init?.

For example:

init?(contentsOfFile path: String, encoding: NSStringEncoding, error: NSErrorPointer)

Source: Xcode 6.1 Beta 2 release notes

Redefinition of private entities

It's now possible to redefine private entities (functions, classes, global variables…) with the same name and type in different files. Previously, the compilation would fail because the linker couldn't tell them apart.

Source: Xcode 6.1 Beta 2 release notes

Other changes in Xcode 6.1 Beta 2

  • Mac apps can now apply @NSApplicationMain attribute to their app delegate class to generate an implicit main function for the app. (This works like @UIApplicationMain on iOS)
  • The fromRaw() static method in enums that have raw values has been replaced with a failble init?(rawValue:) initializer. Also, toRaw() method has been replaced with rawValue property. Example:

    enum Foo: Int { case A = 0, B = 1, C = 2 }
    let foo = Foo(rawValue: 2)! // formerly 'Foo.fromRaw(2)!'
    println(foo.rawValue) // formerly 'foo.toRaw()'
  • Nested functions that recursively reference themselves or other functions inside the same outer function will no longer crash the compiler.

Source: Xcode 6.1 Beta 2 release notes

Changed in Xcode 6.1 Beta 1

Failable initializers

Initializers can now fail by returning nil. A failable initializer is declared with init? to return an explicit optional or init! to return an implicitly unwrapped optional. For example, you could implement String.toInt as a failable initializer of Int like this:

extension Int {
  init?(fromString: String) {
    if let i = fromString.toInt() {
      // Initialize 
      self = i 
    } else { 
      // Discard self and return 'nil'. 
      return nil 
    } 
  }
}

The result of constructing a value using a failable initializer then becomes optional:

if let twentytwo = Int(fromString: "22") {
  println("the number is \(twentytwo)”)
} else {
  println("not a number”)
}

In the current implementation, struct and enum initializers can return nil at any point inside the initializer, but class initializers can only return nil after all of the stored properties of the object have been initialized and self.init or super.init has been called. If self.init or super.init is used to delegate to a failable initializer, then the nil return is implicitly propagated through the current initializer if the called initializer fails.

FWIW, the most common use-case for failable initializers will be with imported Cocoa APIs (e.g. loading a UIImage that can fail). Beta 1 doesn't import Cocoa APIs to use failable initializers, but that will be addressed "soon" in the next 6.1 beta.

— Chris Lattner

Source: Xcode 6.1 Beta 1 release notes, https://devforums.apple.com/message/1042776#1042776

Other changes in Xcode 6.1 Beta 1

Beta 1 has very few developer-facing changes in the standard library:

  • compare method on String is removed
  • StaticString's start was renamed to utf8String, there's also a new withUTF8Buffer method and unicodeScalar property
  • StrideTo and StrideThrough are now Reflectable
  • AssertString and StaticString are now Printable and DebugPrintable
  • New unsafeAddressOf function that can be used to identify objects

Further reading:


Changed in Xcode 6.0 GM / 6.0.1

Xcode 6 GM had no developer-facing Swift changes compared to Beta 7. The golden master was then officially released as 6.0.1 (not 6.0) with no changes to the Swift compiler

Changed in Xcode 6.0 Beta 7

This section needs expanding. Please contribute.

Changed in Xcode 6.0 Beta 6

Refinements to nil coalescing operator

Xcode 6.0 Beta 6 improves on the nil coalescing operator introduced last Beta. It's now possible to pass an optional as the righthand side operand — if both sides evaluate to nil, the whole expression evaluates to nil. This makes it possible to chain expressions using nil coalescing, like so:

let a: Int? = nil
let b: Int? = nil

a ?? b ?? 0

In the example above, the first chained expression that doesn't evaluate to nil will be used as the value of the entire operation.

Previously, passing a non-optional value as the second operand to ?? was technically valid, but its semantics were very confusing

Sources: Xcode 6.0 Beta 6 release notes, Airspeed Velocity

Optionals in Foundation

A large number of Foundation APIs have been audited for optional conformance, removing a significant number of implicitly unwrapped optionals from their interfaces. This clarifies the nullability of their properties and arguments / return values of their methods. This is an ongoing effort since beta 5.

These changes replace T! with either T? or T depending on whether the value can be null (or not) respectively.

Source: Xcode 6.0 Beta 6 release notes

Boolean semantics of types

Implicitly unwrapped optionals no longer conform to BooleanType, which means that they now have to be explicitly compared to nil in if statements. (This follows an equivalent change to Optional last beta)

Meanwhile, non-optional types may no longer be compared to nil.

Source: Xcode 6.0 Beta 6 release notes

Other changes in Xcode 6.0 Beta 6

  • The + operator can no longer append a Character to String, clarifying that + is only for concatenation. (This is analogous to appending an element to an array which was removed in Xcode 6.0 Beta 5)
  • Optional.hasValue was removed
  • RawOptionSetType (used by imported NS_OPTIONS) now supports bitwise assignment operators
  • One-element tuples can no longer have a label. In practice, that means that an enum case that stores one value cannot have a label (this will be fixed).
  • Messages in assert calls can now use string interpolation
  • New precondition() function. It works similarly to assert() (takes a condition and stops the program if it's false), however unlike asserts, preconditions aren't disabled in Release mode builds. (They will be stripped, however, if the application is compiled with -Ounchecked setting)
  • Arbitrary implicit conversions between types using __conversion() -> T was removed. It's now only possible to add implicit conversions from value literals using language-defined protocols.

Further reading:

Changed in Xcode 6.0 Beta 5

dynamic declaration modifier

dynamic is a new attribute that can be applied to properties, methods, subscripts and initializers to make all references to them dynamically dispatched (like message passing in Objective-C). This enables KVO, proxying, swizzling and other advanced Cocoa features to work with Swift.

Before Xcode 6.0 Beta 5, classes marked with @objc (or inheriting from NSObject) got the benefits of dynamic "for free", while non-@objc classes couldn't access dynamic dispatch at all. Now the two concepts are separate:

This change also clarifies the @objc attribute. Now it only makes a declaration visible to Objective-C (for examples, to objc_msgSend), instead of conflating exposure to the Objective-C runtime with guaranteed lack of devirtualization. This is a cleaner conceptual semantic model that enables performance improvements for a wide range of NSObject subclasses by allowing the compiler to use more efficient dispatch and access techniques for declarations marked @objc (which is usually implicit).

Though the feature is independent of the @objc attribute, the implementation still currently relies on Objective-C runtime details, so dynamic currently can only be applied to declarations with @objc-compatible types.

Source: Xcode 6.0 Beta 5 release notes

Mutable optional value types

From Xcode 6.0 Beta 5 release notes:

The optional unwrapping operator x! can now be assigned through. Mutating methods and operators can be applied through it.

var x: Int! = 0
x! = 2
x!++
// Nested dictionaries can now be mutated directly:
var sequences = ["fibonacci": [1, 1, 2, 3, 0]]
sequences["fibonacci"]![4] = 5
sequences["fibonacci"]!.append(8)

The ? chaining operator can also be used to conditionally assign through an optional if it has a ! value:

sequences["fibonacci"]?.append(0)
sequences["fibonacci"]?[6] = 13
sequences["perfect"]?[0] = 6 // Does nothing because the sequence has no 'perfect' key

Previously:

The issue here is that optional forcing and binding operators (postfix ! and ?) return an immutable rvalue, even when the operand is a mutable lvalue. This means that you cannot perform mutating operations on the result, which is why optional arrays, dictionaries and other value types are pretty useless right now. Unfortunately there isn't a great solution or workaround right now: one approach is to wrap the value in a class and use the optional on the class wrapper:

class StringArray {
    var elts : String[]
}
var myArray: StringArray?

We consider this a significant problem and are investigating various solutions to incorporate in a later Beta.

— Chris Lattner

Sources: https://devforums.apple.com/message/998882#998882 Xcode 6.0 Beta 5 release notes

Nil coalescing operator

A new operator, ??, has been introduced to help working with optionals. ?? takes an optional as its left operand, and a non-optional value or expression on the right. If the optional has a value, the whole expression evaluates to the value of the optional (the expression on the right is not evaluated). If the optional is nil, the right hand side expression is evaluated and passed as the result. You can think of the nil coalescing operator like the short-circuiting || operator, but for optionals.

For example:

var myArray: [Int] = []
print(myArray.first ?? 0) // produces 0, because myArray.first is nil
myArray.append(22)
print(myArray.first ?? 0) // produces 22, the value of myArray.first

Sources: Xcode 6.0 Beta 5 release notes

Revised attributes

Following the changes in Xcode 6.0 Beta 4, most of the @-attributes have been changed to declaration modifiers (shedding the @ prefix) in Xcode 6.0 Beta 5.

The @assignment attribute has been removed from operator implementations.

The @prefix, @infix, and @postfix attributes have been changed to declaration modifiers, so they are no longer spelled with an @ sign (now, prefix func (...)). Operator declarations have been rearranged from operator prefix - {} to prefix operator - {} for consistency.

The @class_protocol attribute has been removed; the new syntax for declaring that only protocol conformance is limited to classes is 'protocol P : class { ... }’.

The @auto_closure attribute has been renamed to @autoclosure.

— Xcode 6.0 Beta 5 release notes

Source: Xcode 6.0 Beta 5 release notes

Boolean semantics of optionals

Optional values no longer conform to BooleanType (formerly LogicValue) protocol, which means that:

(...) they may no longer be used in place of boolean expressions (they must be explicitly compared with v != nil).

— Xcode 6.0 Beta 5 release notes

Before this change, the boolean semantics of optionals were confusing when the optional wrapped a value that was a BooleanType itself:

var foo: Bool? = false
// This would print bar
if foo {
    println("bar")
}

Source: Xcode 6.0 Beta 5 release notes

Ranges, Intervals, Striding

Following minor changes to ranges in Xcode 6.0 Beta 3 and Xcode 6.0 Beta 4, Xcode 6.0 Beta 5 brings a major rework of the entire area:

The idea of a Range has been split into three separate concepts:

  • Ranges, which are Collections of consecutive discrete ForwardIndexType values. Ranges are used for slicing and iteration.
  • Intervals over Comparable values, which can efficiently check for containment. Intervals are used for pattern matching in switch statements and by the ~= operator.
  • Striding over Strideable values, which are Comparable and can be advanced an arbitrary distance in O(1).

Some of the types most commonly used with the range operators ..< and ... – for example, Int — conform to both Comparable and ForwardIndexType. When used in a context that requires pattern matching (such as a switch case), the range operators create Intervals. Otherwise they create Ranges. Therefore, in a context without type constraint such as let x = 3..<10, the result is a Range.

It is considered an error to form a Range whose endIndex is not reachable from its startIndex by incrementation, or an Interval whose end is less than its start. In these cases, Interval formation always traps and Range formation traps when a violation is detectable, that is, when the indices are Comparable.

1> 1...0
fatal error: Can't form Range with end < start

Intervals are represented by two generic types, HalfOpenInterval<T>, created by the ..< operator, and ClosedInterval<T>, created by the ... operator:

1> 3.14..<12
$R0: HalfOpenInterval<Double> = {
  _start = 3.1400000000000001
  _end = 12
}
2> 22...99.1
$R1: ClosedInterval<Double> = {
  _start = 22
  _end = 99.099999999999994
}

A range x..<y always has startIndex == x. Therefore, x is the first valid subscript, and this applies even when the Index type is Int. In other words, the first valid subscript of 5..<10 is 5, not 0. To prevent surprise, it is a compilation error to subscript a range over an Integer type outside a generic context (for example, expressions like (5..<10)[0]).

All Ranges are represented by instances of a single generic type, Range<T>, whose representation is always half-open (and thus always print in the REPL and Playgrounds as a half-open range). Currently an inclusive range cannot include the last value in a sequence (for example, 4...Int.max doesn’t work) unless the context requires an Interval (like a case pattern matching specification).

Source: Xcode 6.0 Beta 5 release notes

Required and designated initializers in subclasses

Swift compiler now strictly enforces the presence of required initializers in subclasses. If an ancestor of a class conforms to a protocol requiring a specific initializer and the class doesn't inherit that initializer automatically, it must define it by itself.

What it means, most commonly, is that if you subclass a Cocoa class that conforms to NSCoding (e.g. UIView) and add your own designated initializer, you must also define init(coder:). If you don't want to actually implement it, you can simply make it fail at runtime, like so:

required init(coder: NSCoder) {
    fatalError("Does not implement coding")
}

Also, the compiler now requires overrides of designated initializers to be explicitly marked with override and implementations of required initializers — with required.

Source: Xcode 6.0 Beta 5 release notes

Other changes

Xcode 6.0 Beta 5 has seen a lot of symbols being renamed:

  • Protocols were renamed so that they all end with able, ible or Type. For example, Integer protocol was renamed to IntegerType
  • LogicValue was renamed to BooleanType
  • getLogicValue() became a boolValue property and Optional<T> additionally has a hasValue property
  • UnsafeArray and UnsafeMutableArray were renamed to UnsafeBufferPointer and UnsafeMutableBufferPointer
  • UnsafeConstPointer and UnsafePointer were renamed to UnsafePointer and UnsafeMutablePointer for consistency and to encourage immutability
  • reinterpretCast() was renamed to unsafeBitCast()

Other changes in the standard library:

  • += operator on arrays can no longer append a single item to the array (you have to wrap it into an array)
  • String now has a constructor that takes an integer (you can even supply a radix)
  • New first, last and isEmpty functions

Finally:

  • Together with improvements in earlier betas, the Swift compiler can now produce far faster, better optimized code (on some benchmarks, Swift went from being two orders of magnitude slower than Objective-C to being an order of magnitude faster than Objective-C)
  • Meanwhile, Xcode can now recompile a single changed file instead of recompiling the whole project.
  • You can now import frameworks in Playgrounds
  • println() in Playgrounds now prints next to the line where it's defined (not just printed in the console output)

Further reading:

Changed in Xcode 6.0 Beta 4

Access control

Xcode 6.0 Beta 4 adds three levels of access control to user-defined entities: public (available anywhere), internal (available within the target where they're defined) and private (available only within the file where they're defined).

By default, most entities in a source file have internal access. This allows application developers to largely ignore access control while allowing framework developers full control over a framework's API.

It's also possible to define attributes with public getters but private setters using the private(set) syntax.

It has been noted that the current access control design makes unit testing a bit unwieldy.

Source: Xcode 6.0 Beta 4 release notes

Unicode string improvements

Character was changed in Xcode 6.0 Beta 4 to hold a full grapheme cluster instead of a single code point.

Certain accented characters (like é) can be represented either as a single code point or as a sequence of two or more code points (e + ́)

Before Xcode 6.0 Beta 4, é achieved using "e" and a combining mark would be treated as two Character instances. Now, every character is a single Character. The change helps avoid a class of bugs when dealing with complex Unicode strings.

In addition to the above, Xcode 6.0 Beta 4 removes \x, \u and \U escape sequences for Unicode characters and replaces them with a single, less error-prone \u{1234} syntax

Sources: http://oleb.net/blog/2014/07/swift-strings/ https://devforums.apple.com/message/1007773#1007773

Numerical data type conversion, e.g. CGFloat and Swift Double/Swift Float

From Xcode 6.0 Beta 4 Release Notes:

CGFloat is now a distinct floating-point type that wraps either a Float on 32-bit architectures or a Double on 64-bit architectures.

Sources: Xcode 6.0 Beta 4 release notes

What is happening here is that CGFloat is a typealias for either Float or Double depending on whether you're building for 32 or 64-bits. This is exactly how Objective-C works, but is problematic in Swift because Swift doesn't allow implicit conversions.

We're aware of this problem and consider it to be serious: we are evaluating several different solutions right now and will roll one out in a later beta. As you notice, you can cope with this today by casting to Double. This is inelegant but effective :-)

— Chris Lattner

Sources: https://devforums.apple.com/message/998222#998222

Revised declaration modifiers

The @final, @lazy, @optional, and @required attributes have been converted to declaration modifiers, specified without an @ sign.

Source: Xcode 6.0 Beta 4 release notes

New stride() functions

The .by() method for ranges has been replaced with general stride() functions. To adopt stride(), use stride(from: to: by:) for exclusive ranges and stride(from: through: by:) for inclusive ranges.

For example, you can now do:

stride(from: x, to: y, by: z)      // was: (x..<y).by(z)
stride(from: x, through: y, by: z) // was: (x...y).by(z)

Source: Xcode 6.0 Beta 4 release notes

Set of legal operator characters

With release of Xcode 6.0 Beta 4, the full grammar of operators was specified.

https://developer.apple.com/library/prerelease/mac/documentation/Swift/Conceptual/Swift_Programming_Language/LexicalStructure.html#//apple_ref/doc/uid/TP40014097-CH30-XID_927

The set of characters is in flux, but yes, most unicode symbol characters in the BMP that are classified as 'symbol' and 'math' are available as operator characters.

— Joe Groff

It's not documented yet, but the set of allowed operator characters includes 'math' and 'symbol' characters in the Unicode BMP, and operator characters can be augmented with combining characters. The full set of supported characters will be documented in one of the following seeds.

— Joe Groff

Sources: https://devforums.apple.com/thread/231723?tstart=450 https://devforums.apple.com/message/1000934#1000934

@IBOutlet

Before Xcode 6.0 Beta 4, marking a property with @IBOutlet implicitly made it a weak variable and an implicitly unwrapped optional. Now, the attribute merely makes a property visible to Interface Builder.

Previously:

In Beta 3 (and earlier) the @IBOutlet attribute implicitly makes the variable weak, and implicitly makes it an implicitly unwrapped optional (unless it's explicitly marked with ?). We added the 'strong' modifier in Beta 3.

This is super confusing, too magic, leads to problems (like this) where "retains" are lost for types like arrays because the only reference is weak, and isn't even best practice on iOS where most outlets should be strong. For all of these reasons, in a future Beta, @IBOutlet will become "just" an annotation for IB, without any implicit behavior.

— Chris Lattner

Source: https://devforums.apple.com/message/1002722#1002722

Fixed: Structs with both @lazy and non-lazy properties crashes compiler

structs with a @lazy property followed by a non-lazy property crashes the compiler.

This is fixed, but didn't make it into Beta 3. Stay tuned for a later Beta,

— Chris Lattner

The code from the Developer Forums no longer causes a segmentation fault in the compiler in Xcode 6.0 Beta 4.

Source: https://devforums.apple.com/message/1000950#1000950

Other changes to standard library

  • uppercaseString and lowercaseString properties were removed from String
  • insertionSort and quickSort were removed
  • CString was removed. const char * values are now imported as ConstUnsafePointer<Int8>
  • modulusWithOverflow was replaced by remainderWithOverflow
  • Float and Double no longer conform to RandomAccessIndex, which means they can no longer be used to index a collection
  • true and false are now language literals. Bool conforms to a new BooleanLiteralConvertible protocol that allows user-defined types to support Boolean literals.
  • ArrayBuffer, ArrayBufferType, SliceBuffer and ContiguousArrayBuffer were removed (the reason being, those structures were only an implementation detail of corresponding types)
  • reverse is no longer lazy and simply returns an Array. New lazy functions can be used to lazily reverse, filter and map collections through new LazyForwardCollection, LazyRandomAccessCollection and LazySequence structures

Sources: http://airspeedvelocity.net/2014/07/21/changes-in-the-swift-standard-library-in-beta-4/ Xcode 6.0 Beta 4 release notes

Changed in Xcode 6.0 Beta 3

Array and Dictionary type declaration syntax

Before Xcode 6.0 Beta 3, the shorthand for an Array type was Type[], and Dictionary types were written Dictionary<KeyType, ValueType>. Array type shorthand was changed to [Type] and Dictionaries types now have a shorthand syntax [KeyType: ValueType] (e.g. [String: Bool])

Array value semantics

Since Xcode 6.0 Beta 3, Array has full value semantics to match Dictionary, String and other value types.

Array semantics were in flux at the time of Beta 1, and have been revised to provide full value semantics like Dictionary and String. This will be available in later betas.

— Chris Lattner

Sources: https://devforums.apple.com/thread/228695?start=75&tstart=

Modifying constant properties in designated vs. convenience initializers

What is going on here is that initializers have privledged access to 'let' properties while they run: these properties are actually mutable when accessed directly within the initializer. This is very useful when you're configurating an object during its setup, but it is absolutely required when you have an immutable property dependent on some argument to the initializer, e.g.:

class C {
  let x : Int   // immutable property
  init(input : Int) {
    x = input     // mutating an immutable property!
  }
}

This is an important part of making immutable properties (as opposed to random other immutable variables) useful and functional, but it is dangerous, and potentially allows extensions to a type to violate invariants.

Beta 3 fixes this by only allowing mutation within non-convenience initializers. Convenience inits must delegate to some other initializer anyway, so that initializer can take an argument and do the mutation.

Long story short, this is a feature, not a bug :-)

— Chris Lattner

Source: https://devforums.apple.com/message/1003240#1003240

Range operators

The half-open range operator was changed from .. to ..<.

We considered this carefully. As you can see from this thread, small syntactic issues like this are polarizing, subject to personal preferences, and have no one right answer. See also http://en.wikipedia.org/wiki/Bikeshed

For what it's worth, this approach is precendented in the groovy language. It optimizes for readability and clarity: you're unlikely to mistake one operator for the other when skimming code, and new people coming to Swift are unlikely to assume that ..< is an inclusive range operator (like most assumed when they saw "0..5")

— Chris Lattner

I'd really like it if there was only a single range operator, but that isn't possible (AFAIK):

  • You need to have a half-open range operator to be able to represent an empty range.
  • You need an inclusive range operator to represent finite enumerated sequences when you want to include the last element (e.g. enums, but also integers that you want to include the largest integer value in)..

— Chris Lattner

Sources: https://devforums.apple.com/message/1000100#1000100 https://devforums.apple.com/message/999669#999669