[swift-evolution] [Pitch] merge types and protocols back together with type<Type, Protocol, ...>

Fri May 13 10:25:44 CDT 2016

Lets sum things up before I try to write a draft proposal for this feature `type<>` aka `all<>`.

Is this feature out of scope for Swift 3?

From my point of view it’s definitely not. 

Is the name `type<>` really that bad for the compiler?

Here I’m not sure, because that is out of my experience. From a readers perspective it’s more clear than `all<>`, I would say, even though this is a perfect explanation why it should be called `all<>`:

I also see Thorsten (and in the past Brent's?) argument for calling it "all" or "All", because it's enforcing multiple constraints.
Do we miss any rules here?

`type<>` can contain only one value-type or reference-type and n protocols
the value-type or reference-type should always be the first element between angle brackets
`type<>` should always contain at least 2 types (one value-type or a reference-type and at least one protocol)
reference-types do represent a possible super/base type/class
nesting `type<>` is not allowed, however `type<>` can contain `protocol<>`
protocols conformance of `type<>` is tested with the actual type `type<>` not with the first element of `type<>` (hard to describe this one, does this makes sense to you?)
Does this proposal need more than the base `type<>`?

Maybe, but I think we should start with `type<>` before we will introduce a type operator for this.

Did I missed anything out here? 

PS: Feel free to help me with my English, because it’s not so well.

-- 
Adrian Zubarev
Sent with Airmail

Am 13. Mai 2016 bei 16:21:41, Tony Allevato (allevato at google.com) schrieb:

I think there would be a certain elegance to allowing Boolean type expressions wherever types are currently allowed, so `A & B` being a replacement for `protocol<A, B>` might look nice, and then extend that to allow concrete types as well. Then, if Swift ever decided to support union types, the `|` operator naturally fits there.

One concern though would be whether parsing would get more complicated with deeply composed expressions. If we only supported `&`, there's no real nesting going on. But if we wanted to be forward thinking and leave the door open for `|`, we might need to support things like `(String | Int) & SomeProtocol`, and I'm not enough of a parser expert to know whether that would really complicate things (e.g., could the compiler decide easily enough that those parentheses are part of a type expression and not a function type?).

`all<A, B>` would be a nice compromise in that case, and leave the door open for `any<A, B>` in the future. So I'd be supportive of either option.

On Thu, May 12, 2016 at 1:30 PM Jordan Rose via swift-evolution <swift-evolution at swift.org> wrote:
We've been over this a few times before on the list. I personally like naming this thing "Any<…>" in the same vein as "AnyObject", "AnyClass", and "AnySequence". I also see Thorsten (and in the past Brent's?) argument for calling it "all" or "All", because it's enforcing multiple constraints.

I will say that "type" is unlikely to see much traction simply because it is an incredibly common name for both properties and locals. We went through that exercise when trying to name both "static" and "dynamicType" and decided that it would be too confusing, even if we could make the parsing work.

The feature itself has definitely been shown to be useful when working with the Cocoa frameworks, if not in general. I don't see it on JIRA yet but we have it internally tracked in Radar as rdar://problem/15873071.

Jordan

On May 12, 2016, at 13:08, Adrian Zubarev via swift-evolution <swift-evolution at swift.org> wrote:

I don’t get the part how `all<>` should allow `any<>`. Could you explain that a little bit in detail (I’m not familiar with Ceylon)?

From my point of view `any<>` is something different that I pitched here. `any<>` could be proposed in its own thread, because it is way different than `type<>`. Or can we refine the rules of `type<>` to get to `any<>`?

Here is a little example where `any<>` gets strange:

func foo(value: any<String, Int>) -> any<String, Int> {

    // how would one use value here?
    // what about its properties
    // what will foo return and how to use the result
}

One benefit of `any<>` is the replacement of overloading, at least for the type part of the function.

I’d like to propose `type<>` as the base extension to the language in that direction, before we’ll move forward with more complex scenarios (just like Chris did with generic typealias).

This function is clear that it only will work if you provide a subclass of an UIView which conforms to SomeProtocol (nice addition for library design).

func foo(value: type<UIView, SomeProtocol>) -> type<UIView, SomeProtocol> {

    // use it as a UIView and SomeProtocol at the same type
    return value // return type works great
}

We can split the value just fine:

let mergedValue = foo(SomeViewThatWorks)
let view: UIView = mergedValue
let protocolValue: SomeProtocol = mergedValue

And merge it back together:

guard let newMergedValue = view as? type<UIView, SomeProtocol> else { /* do something */ }

`all<>` could be seen as an alternative name for `type<>`, but to me its not clear what `all<>` can do, whereas `type<>` is almost like `protocol<>`.

-- 
Adrian Zubarev
Sent with Airmail

Am 12. Mai 2016 bei 21:40:24, Thorsten Seitz (tseitz42 at icloud.com) schrieb:

Ceylon uses „&" for intersection types, i.e.

SomeRealClass & SomeProtocol

and the bar („|“) for union types, i.e. 

String | Int

That has proven to be very lightweight and readable in Ceylon where it is heavily used to good effect.

I agree with you that

type<SomeRealClass, SomeProtocol> 

is much nicer than protocol<> for intersection types but to keep the door open for union types, I would prefer

all<SomeRealClass, SomeProtocol>

This would allow

any<String, Int>

to be used for union types.

-Thorsten

Am 12.05.2016 um 16:09 schrieb Adrian Zubarev via swift-evolution <swift-evolution at swift.org>:

protocol<SomeRealClass, SomeProtocol> 
protocol<SomeRealStruct, SomeProtocol> 

This feels really odd to me. 

`type<SomeRealClass, SomeProtocol>` is more clear I’d say.

I think this would be a good addition to the type system and allow us to build more complex and type save code.

But still I’d love to discuss if there might be any disadvantages to this feature.

-- 
Adrian Zubarev
Sent with Airmail

Am 12. Mai 2016 bei 15:11:00, Vladimir.S (svabox at gmail.com) schrieb:

protocol<> 
_______________________________________________
swift-evolution mailing list
swift-evolution at swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution

_______________________________________________
swift-evolution mailing list
swift-evolution at swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution

_______________________________________________
swift-evolution mailing list
swift-evolution at swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20160513/0e04dc57/attachment.html>