<div dir="ltr">Let me first say that this new draft does a great job of explaining the current situation and the goals of the proposal. The revised “Motivation” section in particular is very clearly written and introduces the concepts effectively.<div><br></div><div>I found the previous versions to be highly confusing, and I did not understand from them what was being proposed and why. After reading this one, I feel that I finally comprehend the situation, both in terms of how metatypes behave today and how the authors would like them to work.</div><div><br></div><div>This is still an area where I do not have much expertise, so I look forward to hearing what better-versed people think of the overall approach as well as the specific details.</div><div><br></div><div>Nevin</div><div><br></div><div><br></div></div><div class="gmail_extra"><br><div class="gmail_quote">On Wed, Sep 28, 2016 at 6:18 AM, Adrian Zubarev via swift-evolution <span dir="ltr"><<a href="mailto:swift-evolution@swift.org" target="_blank">swift-evolution@swift.org</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex"><div style="word-wrap:break-word"><div><p>Formatted version: <a href="https://github.com/DevAndArtist/swift-evolution/blob/refactor_metatypes/proposals/0126-refactor-metatypes.md" target="_blank">https://github.com/<wbr>DevAndArtist/swift-evolution/<wbr>blob/refactor_metatypes/<wbr>proposals/0126-refactor-<wbr>metatypes.md</a></p>
<hr>
<h1>Refactor Metatypes</h1>
<ul>
<li>Proposal: <a href="http://0126-refactor-metatypes-repurpose-t-dot-self-and-mirror.md" target="_blank">SE–0126</a></li>
<li>Authors: <a href="https://github.com/DevAndArtist" target="_blank">Adrian Zubarev</a>, <a href="https://github.com/Anton3" target="_blank">Anton Zhilin</a>, <a href="https://github.com/brentdax" target="_blank">Brent Royal-Gordon</a></li>
<li>Status: <strong>Revision</strong></li>
<li>Review manager: <a href="http://github.com/lattner" target="_blank">Chris Lattner</a></li>
<li>Revision: 2</li>
<li>Previous Revisions: <a href="https://github.com/apple/swift-evolution/blob/83707b0879c83dcde778f8163f5768212736fdc2/proposals/0126-refactor-metatypes-repurpose-t-dot-self-and-mirror.md" target="_blank">1</a></li>
</ul>
<h2>Introduction</h2>
<p>This proposal removes <code>.Type</code> and <code>.Protocol</code> in favor of two generic-style syntaxes and aligns global <code>type(of:)</code> function (SE–0096) to match the changes.</p>
<p>Swift-evolution thread (post Swift 3): </p>
<ul>
<li><a>[Pitch] Refactor Metatypes</a></li>
</ul>
<p>Older swift-evolution threads: <a href="https://lists.swift.org/pipermail/swift-evolution/Week-of-Mon-20160718/025115.html" target="_blank">[1]</a>, <a href="https://lists.swift.org/pipermail/swift-evolution/Week-of-Mon-20160718/024772.html" target="_blank">[2]</a>, <a href="https://lists.swift.org/pipermail/swift-evolution/Week-of-Mon-20160704/023818.html" target="_blank">[3]</a></p>
<h2>Motivation</h2>
<p>Every type <code>T</code> has an instance, accessible through <code>T.self</code>, which represents the type itself. Like all instances in Swift, this “type instance” itself has a type, which is referred to as its “metatype”. The metatype of <code>T</code> is written <code>T.Type</code>. The instance members of the metatype are the same as the static or class members of the type.</p>
<p>Metatypes have subtype relationships which reflect the types they represent. For instance, given these types:</p>
<pre><code>protocol Proto {}
class Base {}
class Derived: Base, Proto {}
</code></pre>
<p><code>Derived.Type</code> is a subtype of both <code>Base.Type</code> and <code>Proto.Type</code> (and <code>Any.Type</code>). That means that <code>Derived.self</code> can be used anywhere a <code>Derived.Type</code>, <code>Base.Type</code>, <code>Proto.Type</code>, or <code>Any.Type</code> is called for.</p>
<p>Unfortunately, this simple picture is complicated by protocols. <code>Proto.self</code> is actually of type <code>Proto.Protocol</code>, not type <code>Proto.Type</code>. This is necessary because the protocol does not, and cannot, conform to itself; it requires conforming types to provide static members, but it doesn’t actually provide those members itself. <code>Proto.Type</code> still exists, but it is the supertype of all types conforming to the protocol.</p>
<p>Making this worse, a generic type always uses <code>T.Type</code> to refer to the type of <code>T.self</code>. So when <code>Proto</code> is bound to a generic parameter <code>P</code>, <code>P.Type</code> is the same as <code>Proto.Protocol</code>.</p>
<p>This shifting of types is complicated and confusing; we seek to clean up this area.</p>
<p>We also believe that, in the long term, the dot syntax will prevent us from implementing certain future enhancements that might be valuable:</p>
<ul>
<li>Moving the implementation of metatypes at least partly into the standard library.</li>
<li>Adding members available on all type instances for features like read-write reflection or memory layout information.</li>
<li>Conforming metatypes to protocols like <code>Hashable</code> or <code>CustomStringConvertible</code>.</li>
<li>Offering straightforward syntaxes for dynamic features like looking up types by name.</li>
</ul>
<h2>Proposed solution</h2>
<p>We abolish <code>.Type</code> and <code>.Protocol</code> in favor of two generic-style syntaxes:</p>
<ul>
<li><p><code>Type<T></code> is the concrete type of <code>T.self</code>. A <code>Type<T></code> only ever has one instance, <code>T.self</code>; even if <code>T</code> has a subtype <code>U</code>, <code>Type<U></code> is not a subtype of <code>Type<T></code>.</p></li>
<li><p><code>Subtype<T></code> is the supertype of all <code>Type</code>s whose instances are subtypes of <code>T</code>, including <code>T</code> itself:</p></li>
<li><p>If <code>T</code> is a struct or enum, then <code>Type<T></code> is the only subtype of <code>Subtype<T></code>.</p></li>
<li><p>If <code>T</code> is a class, then <code>Type<T></code> and the <code>Type</code>s of all subclasses of <code>T</code> are subtypes of <code>Subtype<T></code>.</p></li>
<li><p>If <code>T</code> is a protocol, then the <code>Type</code>s of all concrete types conforming to <code>T</code> are subtypes of <code>Subtype<T></code>. <code>Type<T></code> is not itself a subtype of <code>Subtype<T></code>, or of any <code>Subtype</code> other than <code>Subtype<Any></code>.</p></li>
<li><p>Structural types follow the subtype/supertype relationships of their constituent types. For instance:</p></li>
<li><p><code>Type<(NSString, NSString)></code> is a subtype of <code>Subtype<(NSObject, NSObject)></code></p></li>
<li><p>Metatypes of functions are a little bit more special (<a href="https://en.wikipedia.org/wiki/Covariance_and_contravariance_(computer_science)" target="_blank">the subtyping relation on functions flips around for parameter types</a>):</p>
<ul>
<li><code>Type<(Any) -> Void></code> is a subtype of <code>Subtype<(Int) -> Void></code> etc.</li>
<li><code>Type<(Void) -> Int></code> is a subtype of <code>Subtype<(Void) -> Any></code></li>
</ul></li>
</ul>
<p>In this new notation, some of our existing standard library functions would have signatures like:</p>
<pre><code>func unsafeBitCast<T, U>(_: T, to type: Type<U>) -> U
func ==(t0: Subtype<Any>?, t1: Subtype<Any>?) -> Bool
func type<T>(of: T) -> Subtype<T> // SE-0096
</code></pre>
<p>That last example, <code>type(of:)</code>, is rather interesting, because it is actually a magic syntax rather than a function. We propose to align this syntax with <code>Type</code> and <code>Subtype</code> by renaming it to <code>Subtype(of:)</code>. We believe this is clearer about both the type and meaning of the operation.</p>
<pre><code>let anInstance: NSObject = NSString()
let aClass: Subtype<NSObject> = Subtype(of: anInstance)
print(aClass) // => NSString
</code></pre>
<h4>More details:</h4>
<ul>
<li><p>Every static or class member of <code>T</code> which can be called on all subtypes is an instance member of <code>Subtype<T></code>. That includes:</p></li>
<li><p>Static/class properties and methods</p></li>
<li><p>Required initializers (as methods named <code>init</code>)</p></li>
<li><p>Unbound instance methods</p></li>
<li><p>The <code>Type<T></code> of a concrete type <code>T</code> has all of the members required by <code>Subtype<T></code>, plus non-required initializers.</p></li>
<li><p>The <code>Type<T></code> of a protocol <code>T</code> includes only unbound instance methods of <code>T</code>.</p></li>
<li><p>If <code>T</code> conforms to <code>P</code>, then <code>Subtype<T></code> is a subtype of <code>Subtype<P></code>, even if <code>T</code> is a protocol.</p></li>
<li><p>The type of <code>Subtype<T>.self</code> is <code>Type<Subtype<T>></code>.</p></li>
<li><p>The type of <code>Type<T>.self</code> is <code>Type<Type<T>></code>, which is not a subtype of any type except <code>Subtype<Type<T>></code>. There is an infinite regress of <code>Type<...<Type<T>>></code>s.</p></li>
<li><p><code>Subtype</code>s are abstract types similar to class-bound protocols; they, too, support identity operations. </p></li>
<li><p><code>Type</code>s are concrete reference types which have identities just like objects do.</p></li>
</ul>
<p><code>swift
Int.self === Int.self // true
Int.self === Any.self // false
</code></p>
<p></p><u></u><u></u><strong>Visual metatype relationship example (not a valid Swift code)</strong><u></u><p></p>
<pre><code>protocol Foo {
static func foo()
func instanceMethodFoo()
}
protocol Boo : Foo {
static func foo()
static func boo()
func instanceMethodFoo()
func instanceMethodBoo()
}
class A : Foo {
static func foo() { ... }
func instanceMethodFoo() { ... }
}
class B : A, Boo {
static func boo() { ... }
func instanceMethodBoo() { ... }
}
/// Swift generates metatypes along the lines of:
///
/// Syntax: `meta protocol Subtype<T>` - only metatypes can conform to these meta protocols
/// Syntax: `final meta class Type<T>` - metatype
/// Note: `CapturedType` represents `Self` of `T` in `Subtype<T>`
// For Any:
meta protocol Subtype<Any> : meta class {
var `self`: Self { get }
}
final meta class Type<Any> : Subtype<Any> {
var `self`: Type<Any> { ... }
}
// For Foo:
meta protocol Subtype<Foo> : Subtype<Any> {
var `self`: Self { get }
func foo()
func instanceMethodFoo(_ `self`: CapturedType) -> (Void) -> Void
}
final meta class Type<Foo> : Subtype<Any> {
var `self`: Type<Foo> { ... }
func instanceMethodFoo(_ `self`: Foo) -> (Void) -> Void { ... }
}
// For Boo:
meta protocol Subtype<Boo> : Subtype<Foo> {
var `self`: Self { get }
func boo()
func instanceMethodBoo(_ `self`: CapturedType) -> (Void) -> Void
}
final meta class Type<Boo> : Subtype<Any> {
var `self`: Type<Boo> { ... }
func instanceMethodFoo(_ `self`: Boo) -> (Void) -> Void { ... }
func instanceMethodBoo(_ `self`: Boo) -> (Void) -> Void { ... }
}
// For A:
meta protocol Subtype<A> : Subtype<Foo> {
var `self`: Self { get }
func foo()
func instanceMethodFoo(_ `self`: CapturedType) -> (Void) -> Void
}
final meta class Type<A> : Subtype<A> {
var `self`: Type<A> { ... }
func foo() { ... }
func instanceMethodFoo(_ `self`: A) -> (Void) -> Void { ... }
}
// For B:
meta protocol Subtype<B> : Subtype<A>, Subtype<Boo> {
var `self`: Self
func foo()
func boo()
func instanceMethodFoo(_ `self`: CapturedType) -> (Void) -> Void
func instanceMethodBoo(_ `self`: CapturedType) -> (Void) -> Void
}
final meta class Type<B> : Subtype<B> {
var `self`: Type<B> { ... }
func foo() { ... }
func boo() { ... }
func instanceMethodFoo(_ `self`: B) -> (Void) -> Void { ... }
func instanceMethodBoo(_ `self`: B) -> (Void) -> Void { ... }
}
</code></pre>
<p></p><u></u><p></p>
<p></p><u></u><u></u><strong>Some examples</strong><u></u><p></p>
<pre><code>// Types:
protocol Foo {}
protocol Boo : Foo {}
class A : Foo {}
class B : A, Boo {}
struct S: Foo {}
// Metatypes:
let a1: Type<A> = A.self //=> Okay
let p1: Type<Foo> = Foo.self //=> Okay
let p2: Type<Boo> = C.self //=> Error -- `C` is not the same as `Foo`
let any_1: Subtype<Any> = A.self //=> Okay
let any_2: Subtype<Any> = Foo.self //=> Okay
let a_1: Subtype<A> = A.self //=> Okay
let p_1: Subtype<Foo> = A.self //=> Okay
let p_2: Subtype<Foo> = Foo.self //=> Error -- `Type<Foo>` is not a subtype of `Subtype<Foo>`
// Generic functions:
func dynamic<T>(subtype: Subtype<Any>, `is` _: Type<T>) -> Bool {
return type is Subtype<T>
}
func dynamic<T>(subtype: Subtype<Any>, `as` _: Type<T>) -> Subtype<T>? {
return type as? Subtype<T>
}
let s1: Type<S> = S.self
dynamic(subtype: s1, is: Foo.self) //=> true
dynamic(subtype: s1, as: Foo.self) //=> an `Optional<Subtype<Foo>>`
</code></pre>
<p></p><u></u><p></p>
<h2>Future Directions</h2>
<ul>
<li><p>We could allow extensions on <code>Type</code> and perhaps on <code>Subtype</code> to add members or conform them to protocols. This could allow us to remove some standard library hacks, like the non-<code>Equatable</code>-related <code>==</code> operators for types.</p></li>
<li><p>It may be possible to implement parts of <code>Type</code> as a fairly ordinary final class, moving code from the runtime into the standard library.</p></li>
<li><p>We could offer a <code>Subtype(ofType: Type<T>, named: String)</code> pseudo-initializer which would allow type-safe access to classes by name.</p></li>
<li><p>We could offer other reflection and dynamic features on <code>Type</code> and <code>Subtype</code>.</p></li>
<li><p>We could move the <code>MemoryLayout</code> members into <code>Type</code> (presumably prefixed), removing the rather artificial <code>MemoryLayout</code> enum.</p></li>
<li><p>Along with other generics enhancements, there may be a use for a <code>Subprotocol<T></code> syntax for any protocol requiring conformance to protocol <code>T</code>.</p></li>
</ul>
<h2>Impact on existing code</h2>
<p>This is a source-breaking change that can be automated by a migrator. </p>
<p>We suggest the following migration process; this can differ from the final migration process implemented by the core team if this proposal will be accepted:</p>
<ul>
<li><code>Any.Type</code> is migrated to <code>Subtype<Any></code>.</li>
<li>If <code>T.Type</code> is in function parameter, where <code>T</code> is a generic type parameter, then it’s migrated to <code>Type<T></code>.</li>
<li>Every <code>T.Protocol</code> will be replaced with <code>Type<T></code>.</li>
<li>Every <code>T.Type</code> in a dynamic cast will be replaced with <code>Subtype<T></code>.</li>
<li>If static members are called on a metatype instance, then this instance is migrated to <code>Subtype<T></code>.</li>
<li>Return types of functions are migrated to <code>Subtype<T></code>.</li>
<li>Variable declarations is migrated to <code>Subtype<T></code>.</li>
</ul>
<h2>Alternatives considered</h2>
<p>Other names for <code>Type</code> and <code>Subtype</code> were considered:</p>
<ul>
<li>Type: SpecificType, Metatype or ExactType.</li>
<li>Subtype: Supertype, Base, BaseType, ExistentialType or TypeProtocol.</li>
</ul>
<p>Alternatively the pseudo initializer <code>Subtype(of:)</code> could remain as a global function:</p>
<pre><code>public func subtype<T>(of instance: T) -> Subtype<T>
</code></pre><span class="HOEnZb"><font color="#888888">
<p></p></font></span></div><span class="HOEnZb"><font color="#888888"><div><div style="font-family:Helvetica,Arial;font-size:13px;color:rgba(0,0,0,1.0);margin:0px;line-height:auto"><br></div><br><div><div style="font-family:helvetica,arial;font-size:13px">-- <br>Adrian Zubarev<br>Sent with Airmail</div></div></div><div><p></p></div></font></span></div><br>______________________________<wbr>_________________<br>
swift-evolution mailing list<br>
<a href="mailto:swift-evolution@swift.org">swift-evolution@swift.org</a><br>
<a href="https://lists.swift.org/mailman/listinfo/swift-evolution" rel="noreferrer" target="_blank">https://lists.swift.org/<wbr>mailman/listinfo/swift-<wbr>evolution</a><br>
<br></blockquote></div><br></div>