[swift-evolution] [PITCH] ADD AN @RESTRICTED DECLARATION ATTRIBUTE

[Stuart Breckenridge]

Named platforms always take precedence over the *, in essence.  

Sent from my iPad

> On 27 May 2016, at 15:44, Xiaodi Wu <xiaodi.wu at gmail.com> wrote:
> 
> I dislike `@restricted` for several reasons. First of all, but for a desire for brevity, @available and @restricted would do the same thing. I don't think that's justifiable or intuitive. Second, I don't think it's at all readable: are the named platforms restricted from using the function, or is the function restricted to those platforms? The name could equally mean either one.
> 
> Brent's suggestion is perfectly cogent:
> 
> ```
> @available(OSX, introduced: 10.8)
> @available(iOS, introduced: 9.4)
> @available(*, unavailable)
> ```
> 
> I would even go so far as to say that neither `only` nor `@restricted` are needed. By that I mean, if you type this:
> 
> ```
> @available(OSX 10.8, iOS 9.4)
> ```
> 
> That's not valid. You'll be prompted to write this:
> 
> ```
> @available(OSX 10.8, iOS 9.4, *)
> ```
> 
> I'd propose to give the user the additional option of writing this:
> 
> ```
> @available(OSX 10.8, iOS 9.4)
> @available(*, unavailable)
> ```
> 
> In other words, the rule for using the @available shorthand would be that you must mention * somewhere; it can be at the end of the shorthand, or it can be on its own line. This would maintain the desired verbosity so that the latter option is not as accessible as the preferred one, but it avoids the recitation of all unavailable platforms currently required.
> 
> 
>> On Thu, May 26, 2016 at 11:59 PM, Stuart Breckenridge <stuart.breckenridge at icloud.com> wrote:
>> Are we back in a position where a different attribute (back to @restricted or similar keyword) would clear up the readability concerns?
>> 
>> The current equivalent in @available terms:
>> 
>> Short form:
>> 	@available(OSX 10.8, iOS 8.0, *)
>> 	@available(tvOS, unavailable)
>> 	@available(watchOS, unavailable)
>> 
>> Long form:
>> 	@available(OSX, introduced: 10.8, deprecated: 10.10)
>> 	@available(iOS, introduced: 8.0, deprecated: 8.4, obsoleted: 9.0)
>> 	@available(tvOS, unavailable)
>> 	@available(watchOS, unavailable)
>> 
>> Short form replacement:
>> 	@restricted(OSX 10.8, iOS 8.0) // Restricted to OSX, iOS from 10.8 and 8.0, respectively.
>> 
>> Long form replacement:
>> 	@restricted(OSX, introduced: 10.8, deprecated: 10.10)
>>         @restricted(iOS, introduced: 8.0, deprecated: 8.4, obsoleted: 9.0)
>> 
>> 
>> I would amend the draft proposal: * would not be permitted with @restricted. The rationale being that if you are restricting to everything, or marking everything as unavailable, @available(*,unavailable) is a better candidate.
>> 
>> Stuart
>> 
>>>> On 27 May 2016, at 11:06, Brent Royal-Gordon <brent at architechies.com> wrote:
>>>> 
>>>> What I'm saying here is that even if you accept that the asterisk means all platforms, it does not follow that another argument in the same place should refer to the platform names *listed in the attribute*. That's not at all precedented in the meaning of the asterisk.
>>> 
>>> The problem is, the most natural way to express this is absolutely this:
>>> 
>>> 	@available(OSX 10.8, iOS 8, *)	// includes other platforms
>>> 	@available(OSX 10.8, iOS 8)		// excludes other platforms
>>> 
>>> But we don't want to do that because we want people to use `*` by default. You could maybe do something like:
>>> 
>>> 	@available(OSX 10.8, iOS 8, * unavailable)
>>> 
>>> But I think that reads very strangely, and it doesn't seem consistent with the other parameters. (Well, unless you can write `OSX unavailable`, which would actually be kind of convenient, but would still read oddly in a declaration named `@available`.) Hence my suggestion:
>>> 
>>> 	@available(OSX 10.8, iOS 8, only)
>>> 
>>> Which is meant to be read as "Available in OS X (starting in 10.8) and iOS (starting in 8) only".
>>> 
>>> (It would be nice if the syntax actually said "OSX 10.8+", and maybe even permitted a range for things that have been retired, like `OSX 10.8..<10.10`. The main problem I see with supporting a range is that the most natural interpretation of the right end is an unavailability version, but the most useful thing to have there would be a deprecation version.)
>>> 
>>>> Furthermore, the asterisk is put after a list of things *only in the shorthand syntax*, whereas you are proposing that "only" should be usable in the full syntax as well, in a position where the asterisk is not allowed. It is something else entirely to say that a feature is available in OSX version so-and-so only and then to say it's available in iOS version so-and-so only, when in fact you mean that it's not available *on other platforms*.
>>> 
>>> Yes, that was not what I intended when I suggested `only`.
>>> 
>>>> What makes more sense to me would be allowing (if it isn't already allowed) `@available(*, unavailable)` to follow a previous @available declaration in order to mean that the feature is unavailable on all platforms not otherwise specified.
>>> 
>>> Yes, something like this makes sense to me too. Basically, I think the shorthand form should look like:
>>> 
>>> 	@available(OSX 10.8, iOS 9.4, only)
>>> 
>>> And the longhand form like:
>>> 
>>> 	@available(OSX, introduced: 10.8)
>>> 	@available(iOS, introduced: 9.4)
>>> 	@available(*, unavailable)
>>> 
>>> (The switch from `introduced=` to `introduced:` is an already-approved Swift 3 change.)
>>> 
>>> Perhaps you should even be required to say either `@available(*)` or `@available(*, unavailable)` if there are longhand `@available` attributes on the type. 
>>> 
>>> -- 
>>> Brent Royal-Gordon
>>> Architechies
>>> 
>> 
> 
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20160527/9ec9aeab/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: smime.p7s
Type: application/pkcs7-signature
Size: 4005 bytes
Desc: not available
URL: <https://lists.swift.org/pipermail/swift-evolution/attachments/20160527/9ec9aeab/attachment.p7s>