[swift-evolution] access control proposal

Wed Dec 9 11:20:51 CST 2015

Ilya,

I would be happy to write this up.  It sounds like there is a possibility to have a good discussion and possibly even a formal review even if there is some initial reluctance to accept it.

I can't guarantee how soon I will get to it though.  It may take me a week or two (maybe longer) as I have some other proposals already started and my life is pretty busy at the moment.  If you want to see something sooner than that you should probably start a draft and I'll pitch in where I can with feedback or as a co-author.

Matthew

Sent from my iPad

> On Dec 9, 2015, at 5:42 AM, Ilya <ilya at harmonicsense.com> wrote:
> 
> Hi Jordan,
> 
> Thank you, I understand your reasoning. The current model is simple to describe and implement. It’s a good solution for access control across modules. That being said, it doesn’t solve the problem of hiding implementation details at a smaller scale, which is also important. The suggestion to put each class (or part of the class) in a separate file works for large pieces of code but isn’t practical for smaller pieces. An analogy: putting each chapter of a book into a separate file makes sense but putting every paragraph in a separate file does not.
> 
> I think that Matthew had a great idea: add a scope based access control. Swift already has some support for it: local functions. I actually use it a lot and put functions into other functions to hide implementation details. It’s not possible to use it everywhere, but when it is possible, it’s great. If we could have a “local” access modifier that limited the entity only to the scope in which it’s defined, that would allow much less nesting, and it would make it possible to hide many more code.structures.
> 
> For me, hiding implementation details is a fundamental feature of the language / compiler. It is possible to get by without it with coding by convention, but it’s also possible to get by without a strong type system. Having a way to hide implementation details that is enforced by the language will help to have clearly defined interfaces and limit hacks and mistakes for misusing APIs that are meant to be private.
> 
> I appreciate what you said about an official proposal. I’ll ask Matthew again (since access by scope was his idea). If he is not up to it, I’ll submit a proposal myself. For me, this issue was one of the driving forces to join the mailing list.
> 
> —
> Ilya Belenkiy
>> On Dec 8, 2015, at 6:49 PM, Jordan Rose <jordan_rose at apple.com> wrote:
>> 
> 
>> As the primary implementer of the Swift access model when it first went in, I don't think we need something like "classified". The purpose of access control is to make sure implementation details stay hidden from other parts of the program. If you're in the same file, you're not in another part of the program. (If your files are big enough that you are in another part of the program, maybe you should split things up into more files.)
>> 
>> I am glad that you're not treating extensions as "just part of the type". A lot of people say "oh, of course my 'just for the type' modifier allows extensions to see things", forgetting that extensions be added in any file in the program, including outside the module. You can even add extensions for the sole purpose of accessing these members, and therefore there's no protection at all.
>> 
>> This doesn't mean you can't submit a proposal officially to the swift-evolution repo, or even that it won't get reviewed. (I'm not on the core team, and even core team members can be overruled by the rest of the team.) It just means I'm against it. :-) I am glad people are thinking about it, though!
>> 
>> As for "Swift 3" specifically, I'll repeat Chris's canned message: the schedule for Swift 3 is already pretty full with our documented goals, so there's really no guarantee about what will or won't get engineering time for that release.
>> 
>> Best,
>> Jordan
>> 
>>> On Dec 8, 2015, at 8:07, Matthew Johnson via swift-evolution <swift-evolution at swift.org> wrote:
>>> 
>>> I would be happy to do it, but I would like to know how broad the interest in exploring this is first, and especially whether this is something the core team is willing to consider for Swift 3 (depending on details of course).
>>> 
>>> Sent from my iPhone
>>> 
>>>> On Dec 8, 2015, at 7:37 AM, Ilya <ilya at harmonicsense.com> wrote:
>>>> 
>>>> Hi Matthew,
>>>> 
>>>> Maybe you could write the proposal? I'd be happy to do it, but it was your idea. Also, I think that you will express it better.
>>>> 
>>>> --
>>>> Ilya Belenkiy
>>>>> On Mon, Dec 7, 2015 at 2:11 PM Matthew Johnson <matthew at anandabits.com> wrote:
>>>>> Glad you like my reframing!  Is anyone planning to write a proposal for this?
>>>>> 
>>>>> Sent from my iPad
>>>>> 
>>>>>> On Dec 7, 2015, at 1:00 PM, Ilya <ilya at harmonicsense.com> wrote:
>>>>>> 
>>>>>> Yes, that's a great idea!
>>>>>> 
>>>>>> --
>>>>>> Ilya Belenkiy
>>>>>>> On Mon, Dec 7, 2015 at 1:35 PM Matthew Johnson <matthew at anandabits.com> wrote:
>>>>>>> I think what you are really trying to accomplish is to restrict access to a member to the scope in which it is declared, which just happens to be a type declaration.  
>>>>>>> 
>>>>>>> By looking at it in terms of scope the feature would be come more generally applicable.  For example, you could have a 'classified' member in an extension which is only visible inside the body of the extension.  This would be useful if the extension contains helper methods as part of its implementation.
>>>>>>> 
>>>>>>> I would absolutely support extending Swift's scope-based access control to allow for more granular control of visibility within a file by extending it to syntactic scopes.  It seems like a pretty natural enhancement to the existing feature.
>>>>>>> 
>>>>>>> Sent from my iPad
>>>>>>> 
>>>>>>>> On Dec 7, 2015, at 11:02 AM, Ilya <ilya at harmonicsense.com> wrote:
>>>>>>>> 
>>>>>>>> If it is not allowed in extensions or subclasses  (and it shouldn't), the model is still very simple.
>>>>>>>> 
>>>>>>>> I understand the current model, but I think that there is a lot of value in getting the compiler's help in ensuring that class internals cannot be broken by anything outside the class.
>>>>>>>> 
>>>>>>>> The current model doesn't solve this problem well. Putting everything in separate files seems like a very unfortunate workaround.
>>>>>>>> 
>>>>>>>> --
>>>>>>>> Ilya Belenkiy
>>>>>>>> 
>>>>>>>>> On Mon, Dec 7, 2015 at 9:41 AM Matthew Johnson <matthew at anandabits.com> wrote:
>>>>>>>>> I was not sure how I felt about Swift's access control when it was first announced but it didn't take long for me to believe it strikes a pretty good balance between complexity and ability to hide implementation details.
>>>>>>>>> 
>>>>>>>>> It follows a very straightforward scope-driven strategy.  If it is extended further I would like to see any enhancements follow the same scope-driven strategy rather than being type-driven as is seen in other languages.  Proposals along these lines might be interesting to consider and might be able to accomplish what you are hoping to accomplish while affording additional flexibility.
>>>>>>>>> 
>>>>>>>>> In a language where distributing the implementation of a type across several extensions sometimes in different files a type-driven access control model doesn't make a lot of sense to me.  I think it has a lot of potential to be confusing.  
>>>>>>>>> 
>>>>>>>>> What would it mean to limit access to a member to the type itself in Swift?  Would an extensions in the same (or different) files be able to see the member?  What about extensions in a different module?  And what about subclasses? Subclasses probably wouldn't have visibility to them, but any discussion of something like this will probably lead to discussion of protected as well.
>>>>>>>>> 
>>>>>>>>> Sent from my iPad
>>>>>>>>> 
>>>>>>>>>> On Dec 7, 2015, at 7:45 AM, Ilya via swift-evolution <swift-evolution at swift.org> wrote:
>>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>>> Also, one problem with the current approach is that it's all or nothing. The classes in one file see everything from the other classes, and it may be useful to limit that and share only some additional APIs with those related classes, but not all of the implementation details / internal state, while still hiding those same APIs from unrelated classes. Currently, there is no way of doing it, even when splitting code into separate files (the class state must be all in one place and can't be in an extension).
>>>>>>>>>> 
>>>>>>>>>> --
>>>>>>>>>> Ilya Belenkiy
>>>>>>>>>>> On Mon, Dec 7, 2015 at 8:17 AM Ilya <ilya at harmonicsense.com> wrote:
>>>>>>>>>>> My main proposal was not to change the semantics of private (although I would like that) but to introduce a way to keep class implementation details inaccessible to anything outside the class. I think that it should still be possible and useful to have the current level of sharing (I would just name it differently, given what private usually means in other languages).
>>>>>>>>>>> 
>>>>>>>>>>> Just as it is convenient to group several related classes that can refer to each other's implementation details, it is very convenient to have several classes that don't do this but are still related in other important way, for example related APIs or classes that are based on the same implementation idea (like array based data structures).
>>>>>>>>>>> 
>>>>>>>>>>> Having all code in one place rather than jumping through many small files is a very nice thing to have because it really helps to keep everything consistent. The more manual work people have to do for consistency, the more likelihood that it won't be done.
>>>>>>>>>>> 
>>>>>>>>>>> --
>>>>>>>>>>> Ilya Belenkiy
>>>>>>>>>>>> On Mon, Dec 7, 2015 at 12:21 AM John McCall <rjmccall at apple.com> wrote:
>>>>>>>>>>>> 
>>>>>>>>>>>> > On Dec 5, 2015, at 8:39 PM, Ilya via swift-evolution <swift-evolution at swift.org> wrote:
>>>>>>>>>>>> >
>>>>>>>>>>>> > I think the it would help a great deal to have an access level modifier that is really private and visible only inside the class itself. Right now, the only way to hide implementation details for a class is to hide the class code in a separate file, which is very inconvenient for several reasons:
>>>>>>>>>>>> >
>>>>>>>>>>>> > 1) the meaning of the code changes depending on which file the class is in. It's very easy to accidentally expose class internal details and then call class elements that are meant to be used only inside the class. Having a keyword for class internals will allow the compiler to ensure that only the public API for the class is used from the outside world. The user can check types on his own, but it's better that the compiler does it automatically. Similarly, the user can check that only the proper APIs are called, but it's better that the compiler does it automatically.
>>>>>>>>>>>> >
>>>>>>>>>>>> > 2) accessibility by file structure may cause some really short files.
>>>>>>>>>>>> >
>>>>>>>>>>>> > 3) It's impossible to group related classes in one file but still hide implementation details inside each class
>>>>>>>>>>>> >
>>>>>>>>>>>> > I think that it the best solution is to make private keyword do what it states -- keep the class element private to the class. But if it's really important to have a separate keyword for backward compatibility, it would be the next best thing.
>>>>>>>>>>>> 
>>>>>>>>>>>> But on the flip side, with your proposed semantics for private, it would be impossible to write a group of related types, functions, and extensions that do need to refer to each other’s internal details without exposing those details to the entire module.  That’s not really acceptable.
>>>>>>>>>>>> 
>>>>>>>>>>>> The Swift language rule encourages you to put independent definitions in different files.  That definitely means that, if you want to enforce very tight separation of concerns, you’ll end up with more and smaller files.  You haven’t explained why that’s really a *problem*, though.
>>>>>>>>>>>> 
>>>>>>>>>>>> John.
>>>>>>>>> 
>>>>>>>>> 
>>>>>>>>>> _______________________________________________
>>>>>>>>>> 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/20151209/4adf6563/attachment.html>