~Robert Widmann
のメッセージ:
“The Phase Distinction” is a semantic one, not one built into the import
system itself.
I understand. To rephrase my question: why introduce this semantic
distinction to Swift?
What I meant is that even the system I’m modeling this on makes a
distinction between import directives that actually expose identifiers to
modules and import directives that modify identifiers that are already in
scope.
This is, IMO, very complex. I appreciate enormously the conceptual
simplicity of the current Swift approach which, for all of its
deficiencies, has only one import directive that does what it says on the
tin: it exposes identifiers. I'm not bothered if it gains the ability to
expose identifiers differently from one file to the next without keywords
firewalled from each other to preserve the notion of phases of import.
We are *not* changing the unqualified Swift import system. Take a
gander at the proposal again, or even the first draft. Swift has a
particularly strange syntax for qualified imports that hasn’t received
attention since it was first introduced 2 major versions ago. That thing
allows quite a variety of senseless variants that can be both completely
expressed by and subsumed by `using` and `hiding`.
My sense, which I think has been echoed by others, is that the proposed
solution is syntactically complex, and now that I understand that you're
thinking through a multi-phase concept, also conceptually multilayered. I'm
not arguing that the existing syntax for qualified imports doesn't need
changing, only that there is room for radical simplification of the
proposed solution IMO. As I re-read this proposal once more, it strikes me
that the motivating issues identified (needlessly specific, does not
compose, etc.) don't clearly argue for the specific direction proposed as
opposed to alternatives like Joe's.
Perhaps they need to reread the proposal. Syntactically complex how?
We're introducing two keywords and using tuple syntax. Our grammar changes
are laid bare and take up 7 lines. I think there might be a general air of
confusing semantics changes with syntax changes.
I can't speak for the general air, but putting on my disinterested reader
hat, I can see why the confusion might arise--
The Motivation section begins:
"The existing syntax for qualified imports..."
And the Proposed Solution begins:
"The grammar and semantics of qualified imports..."
But the Detailed Design begins:
"Qualified import syntax will be revised..."
It's neither here nor there in terms of the proposal content, but suffice
it to say that if one strings together the topic sentences in your
proposal, the overarching narrative to be gleaned here is: "The current
syntax for qualified imports is no good; therefore, we revise the semantics
of qualified imports by changing the syntax." Sure.
Ne’er the twain shall meet.
Yes, you explained this concept very clearly as it applies to Agda. But I
just don't see why we should care to have this distinction. Yet you are
very adamant about it. What am I missing?
We should care because that is precisely what the two operations *do*.
`using` and `hiding` introduce things or remove things from scope which is
a very different operation from taking something that is already in scope
and giving it a new name.
Perhaps I'm not phrasing my question very cogently. Of course, if we are
to have `using`, `hiding`, and `renaming`, we must observe the distinctions
between them.
If you don’t want to think of them as part of the same import process,
think of them instead in terms of their Swift equivalents today.
import Foundation using (Date) == import struct Foundation.Date
import Foundation hiding (Date) == import Foundation; @unavailable(*,
“…") typealias Date = Foundation.Date
import Foundation using (TimeInterval) renaming (TimeInterval, to: Time)
== import typealias Foundation.TimeInterval; typealias Time =
Foundation.TimeInterval
Notice how it takes two declarations to create a renaming? It is not
simple to drop being explicit about which names are actually in scope and
expect a renaming to just implicitly slip in a using declaration. Nor is
it simple to imagine _ as some magical namespace that you can pack away
unwanted definitions into. using and hiding are very physical things and
the rules for their behavior should be obvious and unambiguous - the
proposal contains some examples of valid and invalid declarations to help
with that.
The examples worry me, in fact. That we might need to contemplate the
behavior of a statement such as `import Foo using () hiding () hiding ()
using () hiding ()` suggests it's perhaps a little over-engineered for the
purpose. Why allow chaining of `using` and `hiding` anyway? The only
example given is of a nested type, which suggests nesting would be the way
to go:
We allow chaining specifically to avoid that nesting behavior. Let's
break up that chained import line by line to see why
import Swift using (String, Int, Double, Character)
hiding (String.UTF8View)
import Swift // Scope contains {all identifiers in Swift}
using (String, Int, Double, Character) // Scope contains {String.*, Int.*,
Double.*, Character.*}
hiding (String.UTF8View) // Scope contains
{{String.* - String.UTF8View.*}, Int.*, Double.*, Character.*}
We express the exact same example in a much more human-readable way. You
can read this out loud in plain English if you don't believe me: "import
everything in Swift that is in String, Int, Double, and Character except
String.UTF8View"
I have to disagree with your reasoning here. Unless I'm mistaken, the
readability we're most concerned with is that of the written text, not its
spoken form.
Nesting is an almost exclusively _visual_ way of organizing text and it
adds real clarity on the page. Of course, if your litmus test for
readability is literally trying to read it out loud, you would conclude
that nesting is inferior. However, you'd make that same conclusion about so
many other choices in Swift syntax (take, for instance, `:` instead of
`extends` or `->` instead of `returns`). I conclude, on the other hand,
that Swift is clearly not aiming to be AppleScript-like in this respect.
In the comments you've used the nested notation `{{String.* -
String.UTF8View.*}, Int.*, Double.*, Character.*}` to explain what it is
your proposed syntax means. I really do find that, with all the punctuation
in it, clearer than the chained notation. Of course, written `(String
hiding UTF8View), Int, Double, Character`, it would be clearer still.
Critically, it eliminates the possibility of absurd chains of `hiding ()
using () hiding () using ()`, which create many more ways to represent the
same subset of imports than can a nested syntax.
Now reread the chaining example in the proposal.
import Swift using (String hiding (UTF8View), Int, Double)
A qualified import is defining a procedure to import a subset of
identifiers. That’s it.
Right, and I think an entirely different way of thinking about this would
be much easier to learn and teach. Whether using, hiding, and renaming are
to be supported now, later, or never, my mental picture of how it fits
together is quite simple:
Analogy--suppose I am a pickle merchant. I import Foo-branded pickles
from vendor X. I must re-label them with the right nutritional information
before I can sell in this country. I can have labels printed saying that
they are Foo-branded pickles. I can have them branded as Bar-branded
pickles. Or I can have the labels deliberately misprinted, and then these
pickles will never see the light of day. Point is, each of these is an
active choice; even if I sell these as Bar-branded pickles, it's not that
these pickles reached the domestic market as Foo-branded pickles, after
which I scratched out the label with a Sharpie. These pickles had no
domestic brand until I gave it one.
Back to importing modifiers--I import type Foo from module X. In my code,
I need to make a choice to call this type Foo, or Bar, or nothing at all.
In other words, there is only one directive, importing, and I am importing
`Foo as Foo`, `Foo as Bar`, or `Foo as _`. Meanwhile, `import X using Foo`
or `import X.Foo` (whatever the color of the bikeshed) would just be a
shorthand for `import X using Foo as Foo` or `import X.Foo as Foo`. In this
conceptualization, if I choose to import Foo as Bar, it's not that I'm
importing Foo into the scope, then changing the identifier to Bar. The only
identifier it ever has in this scope is Bar.
And I’m the one with the complex semantics? :)
I'm just trying to put into words what I'm familiar with after working in
other languages such as Python.
Python may not be the right mindset for this. Their import story is much
simpler because of their module system and generally simpler programming
model.
How about this:
Using and Hiding relate to each other the way && and || do for bools. If
&& can be said to “prefer to return false, but return true given no other
alternative” and || can be said to “prefer returning true, but return false
given no other alternative”, then hiding can be said to “prefer importing
all identifiers unless told not to in specific instances” and using can be
said to “prefer importing no identifiers unless told to in specific
instances”.
import Module.Name using (A, B, C, …) === import Module.Name hiding
(ALL_NAMES - {A, B, C, ...})
import Module.Name hiding (A, B, C, …) === import Module.Name using
(ALL_NAMES - {A, B, C, ...})
That seems a particularly simple explanation to me. Let me know if
anything else is unclear.
Your mental framework is clear. It's one that's just not found in very
many other languages. Many of these have import declarations (or similar)
with simpler syntax, yet they seem to address at least some of the problems
that motivate your proposal. I guess my question in the end is, why have
you chosen Agda as the basis for qualified imports in Swift and not one of
these other languages?
I chose Agda because it's the only major language I could find that
treated identifiers like notation and used its module system for *real*
organization of code; it's just a name, it can change if you want it to.
The entire language is flexible. You can redefine functions with just an
=, you can give new syntactic transformations without having to write crazy
macros or worrying about hygiene. You get so much support from the type
system too that all of these features just work together and you can sit
back and feel satisfied that your tools pushed you to write *good code. *
I wrote it with Agda in mind because they took the time to think about
the interactions between modules, code, and scope in a way we just haven't
had the time to yet.
Many of these imports have simpler syntax but don't chain so lose
expressiveness (Java). Or may have simpler surface syntax but don't
properly interact with a moduleless system (Python, Java). Or may have a
simpler syntax and give way to ambiguities (Python) when used in Swift.
Agda is unambiguous, scalable, extensible, and simple.
Please don't confuse what's new with what's complex. But at the same time
if there are any unexplainable loopholes we should know about them. I just
haven't heard much I couldn't point to the proposal about yet so I don't
see a reason to change.
Thanks for this insight. It's clear that you're aiming at a far richer
system in the future. However, it's hard to gauge how far Swift will
eventually scale/extend what you're proposing here. I don't think there's
any denying that the solution you've chosen introduces more involved
semantics and a more verbose syntax than what's strictly necessary to
address the specific motivating problem you give in this particular
instance. Without a better sense of the outer bounds of how far this system
will eventually be pushed, it's hard to judge whether the eventual payoff
will be "worth it." And without any mention of the grander plan, I suspect
the feedback you're going to get will continue along the path of trying to
tear away whatever you're trying to put in place for the future which is
not discernibly necessary for solving the immediate problem at hand.
···
On Wed, Jul 20, 2016 at 10:33 PM, Robert Widmann <devteam.codafi@gmail.com> wrote:
2016/07/20 19:01、Xiaodi Wu via swift-evolution <swift-evolution@swift.org>
On Wed, Jul 20, 2016 at 8:10 PM, Robert Widmann <rwidmann@apple.com> > wrote:
On Jul 20, 2016, at 5:47 PM, Xiaodi Wu <xiaodi.wu@gmail.com> wrote:
On Wed, Jul 20, 2016 at 6:30 PM, Robert Widmann <rwidmann@apple.com> >> wrote:
On Jul 20, 2016, at 4:17 PM, Xiaodi Wu <xiaodi.wu@gmail.com> wrote:
On Wed, Jul 20, 2016 at 4:57 PM, Robert Widmann <rwidmann@apple.com> >>> wrote:
On Jul 20, 2016, at 2:52 PM, Robert Widmann via swift-evolution < >>>> swift-evolution@swift.org> wrote:
On Jul 20, 2016, at 2:35 PM, Xiaodi Wu <xiaodi.wu@gmail.com> wrote:
On Wed, Jul 20, 2016 at 4:24 PM, Robert Widmann <rwidmann@apple.com> >>>> wrote:
On Jul 20, 2016, at 2:19 PM, Xiaodi Wu <xiaodi.wu@gmail.com> wrote:
On Wed, Jul 20, 2016 at 4:06 PM, Robert Widmann <rwidmann@apple.com> >>>>> wrote:
On Jul 20, 2016, at 2:04 PM, Robert Widmann via swift-evolution < >>>>>> swift-evolution@swift.org> wrote:
On Jul 20, 2016, at 1:59 PM, Xiaodi Wu <xiaodi.wu@gmail.com> wrote:
Why is hiding in-scope but renaming out-of-scope?
Because hiding and renaming can be used in combination to subset out
APIs, not alter them.
I mistyped. Should be "Because hiding and using can be used in
combination to subset out APIs, not alter them."
Sure, I buy that.
Both are additive to Swift,
As part of this proposal, both are source-breaking.
I don't see how. If hiding were cut from the proposal, adding it later
with even the exact syntax you propose should break no pre-existing
code--am I wrong?
Renaming the way we originally laid it out would certainly be
additive. The way you have it laid out would overlap a bit with hiding,
sure, but it is still additive and (IMO, but I’m probably among a tiny
minority of users that has used a proof assistant’s syntax as the basis for
a proposal!) a good thing to have.
Sorry, I fear I've incorrectly communicated the point I was trying to
make. I'm not advocating here for inclusion of renaming as part of this
proposal. I simply think that--even though I buy your claim that hiding and
using both subset out APIs--hiding has more affinity with renaming and the
two facilities probably ought to be considered together, whenever that is.
Thus, I'm suggesting that it would be feasible to postpone discussion
of hiding until such future time as a fully fleshed out renaming scheme is
proposed. A revamped source-breaking import syntax without either hiding or
renaming could be put in place now, and future addition of hiding and/or
renaming would not have to be source-breaking. Is there something wrong
with this argument?
There is still a useful to distinction to be made between explicitly
renaming an API and explicitly hiding an API. Scala’s syntax to rename to
underbar is a convenient notation for that kind of thing, but it goes
against making qualified imports explicit and it means that renaming
necessarily has to import identifiers into scope as well as rename them.
What the OP (maybe it was you, sorry if it was) meant by “equivalent”
missed the point that
import Swift hiding (String)
doesn’t translate into
import Swift renaming (String, to: _)
it translates into
import Swift hiding () renaming (String, to: _)
Renaming introducing identifiers into scope seems like a phase-shift
and is not something the verb “rename” implies should happen here. It’s an
interesting little hole in Agda’s module system that you can use
open A hiding (xs) renaming (ys to zs)
to mean
open A using (A; xs; ys) renaming (ys to zs)
Actually, scratch that. Their documentation explicitly mentions that
hiding and renaming may not be mixed because of the phase distinction and
recommend the using translation above as the way to go.
This is very illuminating. I think I've rather misunderstood what it is
you're proposing. I wonder if others did also.
The syntax you proposed seemed cumbersome to me because my mental model
of importing (informed by my probably superficial understanding of vanilla
procedural programming languages) has only one phase: importing. This is
why I proposed radically simplifying the spelling. To me, all of these
operations are just sugar on a single import phase, where "stuff" from
outside the module is "brought into" the module, either with the same name
("using"), a different name ("renaming"), or no name ("hiding").
But what you're saying here--if I understand correctly--is that you're
proposing a multi-phase import system, where the possible phases, which can
be composed in varying orders, are "using", "hiding", and "renaming". This
is much, much more elaborate than I had contemplated. So beyond the
bikeshedding of syntax, I'd ask: why do we need this multi-phase model of
importing?
and as has been argued by others, the former is a special case of the
latter.
A special case that cannot cause large-scale file-relative changes to
APIs. Renaming is primarily used in other languages that treat free
functions as more canonical than we do, or allow operator definitions that
can be used as notation.
I don't know about 'primary use,' but the most common use I've
experienced in Python, for example, is the mundane task of importing module
Foo2 as Foo.
And I still want that kind of syntax. I just want to get the breaking
changes out of the way to make room for it in the future.
Right. See above about my argument as to which parts of your proposal
have to be source-breaking, and which don't.
In those cases, you often have your own notation you’d like to use.
In Swift, such changes should be rare enough that if you can’t solve them
with a disambiguating qualified import then you can just redeclare the
identifier some other way (typealias, top-level let, wrapper class,
whatever).
You've already stripped out renaming of members from the proposal. I
agree wholeheartedly. The only flavor of renaming I'm thinking of here is
equivalent to a fileprivate typealias and hiding, which cannot be done in
this version of the proposal because hiding always comes before
typealiasing and you can't typealias what isn't imported. It isn't about
altering APIs any more than a fileprivate typealias can be thought of as
altering APIs.
In the sense that you can’t use the original identifier if you rename
it, it is an alteration. John brought up a great point about exporting
these things and how it could be a potentially dangerous thing. Even used
locally, there’s the potential for people to specify 500 lines of import
renaming crap that has to be copypasta’d throughout the codebase to
maintain that particular style - not a use-case I’ve ever seen, but the
potential is there.
This is, I think, a spurious argument. I can equally have 500 lines of
private typealiased crap that has to be copypasta'd.
On Wed, Jul 20, 2016 at 15:55 Brandon Knope <bknope@me.com> wrote:
I meant is there any reason for requiring parentheses
On Jul 20, 2016, at 4:53 PM, Robert Widmann <rwidmann@apple.com> >>>>>>> wrote:
Renaming is out of scope for this proposal, that’s why.
On Jul 20, 2016, at 1:26 PM, Brandon Knope <bknope@me.com> wrote:
I prefer this 100x more
Is there any reason why this wouldn't work?
Brandon
On Jul 20, 2016, at 4:13 PM, Xiaodi Wu <xiaodi.wu@gmail.com> wrote:
Yeah, I'd be happy to lose the parentheses as well.
In the last thread, my take on simplifying the proposed syntax was:
import Swift using String, Int
// or, for hiding:
import Swift using Int as _
The key simplification here is that hiding doesn't need its own
contextual keyboard, especially if we support renaming (a huge plus in my
book), as renaming to anything unused (or explicitly to `_`) is what hiding
is all about.
On Wed, Jul 20, 2016 at 15:01 Brandon Knope <bknope@me.com> wrote:
On Jul 20, 2016, at 3:08 PM, Xiaodi Wu via swift-evolution < >>>>>>>> swift-evolution@swift.org> wrote:
As Joe and others mentioned in the previous thread, this syntax
could be greatly simplified in ways that resemble analogous facilities in
other languages. In particular I think it's alarmingly asymmetrical that,
in your proposal, `import Swift using (String)` imports *only* String while
`import Swift hiding (String)` imports *everything but* String. This
becomes evident when chained together:
import Swift using (String, Int)
// imports only String and Int
import Swift using (String, Int) hiding (String)
// imports only Int
import Swift hiding (String, Int)
// imports everything except String and Int
import Swift hiding (String, Int) using (String)
// imports *nothing*? nothing except String? everything except Int?
confusing.
By contrast, Joe's proposed syntax (with some riffs) produces
something much more terse *and* much more clear:
import Swift.*
import Swift.(Int as MyInt, *)
import Swift.(Int as _, *)
I really don't find this much clearer than the proposed one. The
proposal reads much clearer.
Joe's syntax has a lot going on in my opinion.
For the proposal, do we really need the parentheses? It makes the
syntax look heavier
Brandon
On Wed, Jul 20, 2016 at 1:52 PM, Robert Widmann via swift-evolution >>>>>>>> <swift-evolution@swift.org> wrote:
Hello all,
I’d like to thank the members of the community that have guided
the revisions of this proposal. We have decided to heed the advice of the
community and break down our original proposal on modules and qualified
imports into source-breaking (qualified imports) and additive (modules)
proposals. As qualified imports is the change most suited to Swift 3, we
are pushing that proposal now as our final draft.
It can be had inline with this email, on Github
<https://github.com/apple/swift-evolution/pull/440>, or as a gist
<https://gist.github.com/CodaFi/42e5e5e94d857547abc381d9a9d0afd6>\.
Thanks,
~Robert Widmann
Qualified Imports Revisited
- Proposal: SE-NNNN
<https://gist.github.com/CodaFi/NNNN-first-class-qualified-imports.md>
- Authors: Robert Widmann <https://github.com/codafi>, TJ
Usiyan <https://github.com/griotspeak>
- Status: Awaiting review
- Review manager: TBD
<qualified-imports.md · GitHub;
Introduction
We propose a complete overhaul of the qualified imports syntax and
semantics.
<qualified-imports.md · GitHub;
Motivation
The existing syntax for qualified imports from modules is
needlessly explicit, does not compose, and has a default semantics that
dilutes the intended meaning of the very operation itself. Today, a
qualified import looks something like this
import class Foundation.Date
This means that clients of Foundation that wish to see only Date must
know the exact kind of declaration that identifier is. In addition, though
this import specifies exactly one class be imported from Foundation, the
actual semantics mean Swift will recursively open all of Foundation's
submodules so you can see, and use, every other identifier anyway - and
they are not filtered from code completion. Qualified imports deserve to be
first-class in Swift, and that is what we intend to make them with this
proposal.
<qualified-imports.md · GitHub
solution
The grammar and semantics of qualified imports will change
completely with the addition of *import qualifiers* and *import
directives*. We also introduce two new contextual keywords: using
and hiding, to facilitate fine-grained usage of module contents.
<qualified-imports.md · GitHub
design
Qualified import syntax will be revised to the following
import-decl -> import <import-path> <(opt) import-directive-list>
import-path -> <identifier>
-> <identifier>.<identifier>
import-directive-list -> <import-directive>
-> <import-directive> <import-directive-list>
import-directive -> using (<identifier>, ...)
-> hiding (<identifier>, ...)
This introduces the concept of an import *directive*. An import
directive is a file-local modification of an imported identifier. A
directive can be one of 2 operations:
1) *using*: The *using* directive is followed by a list of
identifiers for non-member nominal declarations within the imported module
that should be exposed to this file.
// The only visible parts of Foundation in this file are // Foundation.Date, Foundation.DateFormatter, and Foundation.DateComponents//// Previously, this was// import class Foundation.Date// import class Foundation.DateFormatter// import class Foundation.DateComponentsimport Foundation using (Date, DateFormatter, DateComponents)
2) *hiding*: The hiding directive is followed by a list of
identifiers for non-member nominal declarations within the imported module
that should be hidden from this file.
// Imports all of Foundation except `Date`import Foundation hiding (Date)
As today, all hidden identifiers do not hide the type, they merely
hide that type’s members and its declaration. For example, this means
values of hidden types are still allowed. Unlike the existing
implementation, using their members is forbidden.
// Imports `DateFormatter` but the declaration of `Date` is hidden.import Foundation using (DateFormatter)
var d = DateFormatter().date(from: "...") // Validvar dt : Date = DateFormatter().date(from: "...") // Invalid: Cannot use name of hidden type.
d.addTimeInterval(5.0) // Invalid: Cannot use members of hidden type.
Import directives chain to one another and can be used to create a
fine-grained module import:
// This imports Swift.Int, Swift.Double, and Swift.String but hides Swift.String.UTF8Viewimport Swift using (String, Int, Double)
hiding (String.UTF8View)
Directive chaining occurs left-to-right:
// This says to 1) Use Int 2) Hide String 3) rename Double to Triple. It is invalid// because 1) Int is available 2) String is not, error.import Swift using (Int) hiding (String)// Valid. This will be merged as `using (Int)`import Swift using () using (Int)// Valid. This will be merged as `hiding (String, Double)`import Swift hiding (String) hiding (Double) hiding ()// Valid (if redundant). This will be merged as `using ()`import Swift using (String) hiding (String)
Because import directives are file-local, they will never be
exported along with the module that declares them.
<qualified-imports.md · GitHub
on existing code
Existing code that is using qualified module import syntax (import
{func|class|typealias|class|struct|enum|protocol} <qualified-name>)
will be deprecated and should be removed or migrated.
<qualified-imports.md · GitHub
considered
A previous iteration of this proposal introduced an operation to
allow the renaming of identifiers, especially members. The original intent
was to allow file-local modifications of APIs consumers felt needed to
conform to their specific coding style. On review, we felt the feature was
not as significant as to warrant inclusion and was ripe for abuse in large
projects.
_______________________________________________
swift-evolution mailing list
swift-evolution@swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution
_______________________________________________
swift-evolution mailing list
swift-evolution@swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution
_______________________________________________
swift-evolution mailing list
swift-evolution@swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution
_______________________________________________
swift-evolution mailing list
swift-evolution@swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution
_______________________________________________
swift-evolution mailing list
swift-evolution@swift.org
https://lists.swift.org/mailman/listinfo/swift-evolution