[swift-evolution] [swift-evolution-announce] [Review] SE-0169: Improve Interaction Between private Declarations and Extensions

Xiaodi Wu xiaodi.wu at gmail.com
Sat Apr 8 14:58:07 CDT 2017

On Sat, Apr 8, 2017 at 12:01 AM, BJ Homer via swift-evolution <
swift-evolution at swift.org> wrote:

> If private is required to be tied to types forever and always, then yes,
> this proposal should be accepted.

Let's be clear: private is not and will never be tied to types. It is tied
to *scopes*. This proposal has received a lot of responses suggesting that
the proposed rules should be extended to allow extensions in the same
module to access private members, but that fundamentally misunderstands
access modifiers in Swift.

Two things here. First, Swift allows types to be extended after the fact:
outside the same declaration, outside the same file, even outside the same
module. Second, Swift has access modifiers where each level of access is a
superset of the previous. That is, if you change `private let a` to
`fileprivate let a`, there is no place where `a` becomes no longer
accessible; the same goes for each pair of access modifiers. These two
design decisions are, as far as I'm aware, "now and forever" aspects of

Given that these two things are true, *no access modifier is or can be tied
to types*. This proposal is about extending the meaning of *scope* to
include same-file extensions. If you try to think of private as a
type-private modifier, you will invariably conclude that it is broken,
because you can't see private members in extensions outside the same scope.
You will also see the proposed private as broken, only less so, because you
can't see private members in extensions outside the same file. However,
unless I'm greatly mistaken, it is not the intention of this or any other
proposal to create an access level tied to types.

To do otherwise is to suggest that the use of "private" by beginners and
> the use of same-file extensions by beginners are incompatible. The former
> is silly; "private" is what beginners will naturally reach for. The latter
> is possible, but historically the use of same-file extensions has been
> quite idiomatic. So, I reluctantly agree we should accept this proposal.
> -BJ
> > On Apr 7, 2017, at 10:34 PM, John McCall via swift-evolution <
> swift-evolution at swift.org> wrote:
> >
> >
> >>> On Apr 7, 2017, at 8:12 PM, Jakub Suder via swift-evolution <
> swift-evolution at swift.org> wrote:
> >>>
> >>> What is your evaluation of the proposal?
> >>
> >> If this is the last option we have to change the status quo, any
> renaming is off the table, no further changes after Swift 4, and it's
> either this or being stuck with 'fileprivate' until the end of time, then
> +1 from me. It will increase the convenience of access control for people
> like me who see types and their extensions as parts of the same entity,
> just spread visually across neighboring blocks. In almost any other
> language these two would indeed be one entity, since most languages don't
> have any way of dividing types into pieces this way.
> >>
> >> However, IMHO any of these would be a better solution:
> >
> > I'd like to respond briefly to this to clarify the Core Team's decisions
> about what solutions are under consideration, both now and in the future.
> By doing this, I don't mean to pressure you towards any particular stance.
> The Core Team asked for this to be proposed because we wanted to know how
> the community felt about it; we are not specifically trying to get it
> approved, at least as a group.
> >
> >> 1) Rename 'private' to something else ('scoped'?) and rename
> 'fileprivate' back to 'private'
> >
> > The Core Team has rejected making such a major change in the
> interpretation of 'private'.  'private' will be tied to scopes, now and
> forever.  The only question is whether extensions of the same type within a
> file should be considered part of the same scope for the purposes of
> 'private'.  Swift 4 is the deadline for making that change; if it, too, is
> rejected, 'private' will be purely lexical forever.
> >
> >> 2) Rename 'fileprivate' to something more friendly (I liked the 'local'
> suggestion that Vladimir made today)
> >
> > The Core Team is willing to consider adding a new keyword to replace
> 'fileprivate', but not in Swift 4.
> >
> > Speaking just for myself, I don't think we'd accept such a change purely
> for aesthetics; it would have to be because 'fileprivate' seemed
> inappropriate for some new generalization, e.g. if we added sub-file
> submodules and wanted 'fileprivate' to allow access only within the
> submodule.  That is assuming a lot about how a future submodule feature
> would work, and we aren't going to design that feature specifically with a
> goal of replacing this keyword, and frankly we don't know when we're going
> to take that on at all.  I would caution people against assuming that
> 'fileprivate' will be renamed.
> >
> >> 3) Postpone this until we add submodules, but with the assumption that
> it will be possible to make some source-breaking changes at that point
> >
> > The Core Team is not willing to change the basic design of 'private' in
> future releases of Swift.  If some feature — perhaps submodules — demands
> that we define its interaction with 'private', the design of that
> interaction will have to feel natural and consistent with the at-that-point
> extant design of 'private'.  For example, it would not be acceptable if,
> say, adding a submodule declaration to a file suddenly changed the
> interpretation of 'private'.
> >
> > An option (4) that you didn't list but which I should cover for
> completeness would be to add new keywords in the future with new
> interpretations.  This is something that the Core Team is willing to
> consider.  However, speaking just for myself again, I find it unlikely that
> we would add new access control features just to express
> increasingly-precise refinements; it would have to be something that felt
> necessary because of some real inadequacy in the existing access-control
> levels as applied to some other new feature (e.g. submodules).
> >
> > John.
> >
> >> The thing I don't like about this proposal (or status quo) - apart from
> the fact that it will make people who like the current strict private
> unhappy - is that 'private' even right now means kind of two different
> things:
> >>
> >> - for a property or a method defined in a class/struct, it means
> "available only inside this type"
> >> - for a private global variable/constant/function, or a private type or
> extension, it means "available in this file" i.e. the same as 'fileprivate'
> >>
> >> So if we're worried that this proposal makes the meaning of 'private'
> unclear - it already is unclear. Wouldn't it be much more clear if private
> global variables, functions and classes were required to use the access
> level that means "available in this file", since that's how they actually
> work? (as long as this level is not named 'fileprivate' :)
> >>
> >> And the other access level, the "available only inside this type"
> (private / scoped), could only be used for things that are actually
> contained inside a type, and wouldn't have two different meanings depending
> on whether it's at the root of the file or nested inside something else.
> >>
> >> I really believe that even though this is kind of painful for everyone,
> it's worth spending time to figure out a solution that satisfies most of us
> and makes the language clearer and friendlier in the long term, even if it
> means breaking compatibility again. Do we want to be stuck with an
> imperfect solution 10 years from now, because we didn't want to do this
> last breaking change now?
> >>
> >>
> >>> Is the problem being addressed significant enough to warrant a change
> to Swift?
> >>
> >> I think almost everyone here agrees it is significant.
> >>
> >>
> >>> Does this proposal fit well with the feel and direction of Swift?
> >>
> >> This is a difficult question. IMHO it would definitely fit much better
> with the direction of Swift if we bit the bullet and did whatever we agree
> will make the language simpler and better long term, regardless how many
> changed lines in git this will cause when Swift 4 is released.
> >>
> >>
> >>> If you have used other languages or libraries with a similar feature,
> how do you feel that this proposal compares to those?
> >>
> >> I've used many languages, but I think in none of them it was a common
> thing (or even a possibility) to split types into several blocks like we do
> in Swift with extensions, so the main issue we're talking about didn't
> exist there. In Ruby you can kind of do the same thing with modules - if
> you have a private method in a module, you can access it from methods in
> the main type, even if they're defined in a different file. (But Ruby isn't
> very strict about access control in general, e.g. it allows you to call
> private methods on any object via #send).
> >>
> >>
> >>> How much effort did you put into your review? A glance, a quick
> reading, or an in-depth study?
> >>
> >> I've read or skimmed through most of the messages in the recent threads
> about this and the last proposal.
> >>
> >> Disclaimer: I have very little experience with Swift 3 (but plenty with
> Swift 2.x).
> >>
> >> Kuba
> >>
> >> _______________________________________________
> >> 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/20170408/82f1ca05/attachment.html>

More information about the swift-evolution mailing list