[swift-evolution] pitch: Unified libc import (again)

Thu Aug 17 22:06:10 CDT 2017

Okay I can sense this thread getting derailed so I’ll try to address your
comments one by one.

* stderr should go wherever stdin and stdout go. Since it’d be silly for a
function like `print(_:separator:terminator:)` or `
readLine(strippingNewline:)` to live anywhere but the standard library,
then it stands to reason that the stderr version should also live in the
standard library.

* Foundation is “supposed” to be the core library that does these things,
but this isn’t a good thing. Rationale:

    * Foundation is literally an Objective C framework that’s written with
Swift syntax. The classes extend *NSObject*. You use them by *subclassing*
them and overriding open methods. This is patently unswifty. Even parts of
Foundation that look swifty on the outside (like `URL`) are still backed by
NS reference types internally.

    * To fix this would require a complete rewrite from the ground up,
that’s more than a straight port from the Objective C framework. Many on
this list have also expressed desire to use this chance to redesign these
APIs. Since we’d basically have to rewrite all of it, any effort to
modernize Foundation is basically writing a new core library.

    * Even if a piece of Foundation was rewritten like a real Swift
library, because of the monolith pattern, you still wind up importing a lot
of legacy code that shouldn’t be floating around your project. Using the
file system tools shouldn’t change `String`.

* Foundation’s file system capabilities are really just a class wrapper
around Glibc/Darwin functions. So in a way, Foundation is already a sort of
unified import for C, except it brings in a whole load of Objective C cruft
with it.

    * Logically, this means in terms of the XY problem, Foundation is on
the Y side. Foundation is what happens if you rely on Swift’s C interop to
get the job done instead of building a true Swift solution, which would
involve Swift quarterbacking all the nasty system calls, instead of
Glibc/Darwin doing it.

In conclusion,

* What we need is a modern Swift solution for accessing file systems.

* What we have to do right now is leverage Swift’s C interop to use
Glibc/Darwin, while handing imports and platform inconsistencies manually.

* Eventually, the ideal would be for Swift to handle that in Swift, instead
of delegating to a platform-dependent C library, or at least standardize
things so that a swifty Swift library imports the right C library for you,
and exports a platform-independent set of symbols to the user, or a higher
level API.

* Foundation is in many ways the worst of both worlds. It handles the
Glibc/Darwin import for us, but at the cost of an aging framework that runs
against the grain of good Swift design idioms.

On Thu, Aug 17, 2017 at 8:16 PM, Xiaodi Wu <xiaodi.wu at gmail.com> wrote:

> On Thu, Aug 17, 2017 at 6:46 PM, Taylor Swift <kelvin13ma at gmail.com>
> wrote:
>
>> I don’t think the “is this library functionality or standard library
>> functionality” argument is worth having, but if stdout and stdin are
>> first-class citizens in the Swift world, so should stderr.
>>
>> As for bringing Foundation into the discussion, you can’t really talk
>> about Foundation without also talking about the mountains of problems that
>> come with the monolith pattern. But that’s a completely different
>> conversation to be had.
>>
>
> I'm not sure what you're getting at here, but I don't believe you've
> addressed my question, which is: it's been firmly decided that I/O belongs
> in Foundation, and Foundation does in fact offer such facilities--what is
> missing from those facilities, and how can we fill it out?
>
>
> On Thu, Aug 17, 2017 at 7:13 PM, Xiaodi Wu <xiaodi.wu at gmail.com> wrote:
>>
>>> IMO, you’re touching on at least three or four separate topics here.
>>> Daniel touched on several, but just some comments/questions:
>>>
>>> * Standard error output is something that’s been discussed here
>>> previously. I believe the last pitch had something like StandardError being
>>> added to the standard library as a TextOutputStream.
>>>
>>> * Foundation is supposed to be the core library that provides file
>>> system access facilities. What’s missing, and can we add it to Foundation?
>>>
>>> On Thu, Aug 17, 2017 at 17:50 Taylor Swift via swift-evolution <
>>> swift-evolution at swift.org> wrote:
>>>
>>>> Python’s os.path <https://docs.python.org/2/library/os.path.html> is a
>>>> nice abstract model for doing path manipulations. Maybe Swift could get a
>>>> struct like `Filepath` or something on which these operations could be done.
>>>>
>>>> On Thu, Aug 17, 2017 at 6:47 PM, Taylor Swift <kelvin13ma at gmail.com>
>>>> wrote:
>>>>
>>>>>
>>>>>
>>>>> On Thu, Aug 17, 2017 at 3:50 PM, Daniel Dunbar <
>>>>> daniel_dunbar at apple.com> wrote:
>>>>>
>>>>>>
>>>>>> > On Aug 17, 2017, at 9:26 AM, Taylor Swift via swift-evolution <
>>>>>> swift-evolution at swift.org> wrote:
>>>>>> >
>>>>>> > I know this has come up before without any action, but having the
>>>>>> standard C library be packaged under `Darwin` on OSX and `Glibc` under
>>>>>> Linux is something that’s really becoming an issue as the Swift package
>>>>>> ecosystem matures. Right now a lot of packages are a lot less portable than
>>>>>> they could be because somewhere along the dependency line, there is a
>>>>>> package that only imports the C library for one platform. Unifying it under
>>>>>> one import would allow people to write packages that are portable by
>>>>>> default.
>>>>>>
>>>>>> What we (SwiftPM) have done for now is use a `libc` target to start
>>>>>> by normalizing the name:
>>>>>>   https://github.com/apple/swift-package-manager/tree/master/S
>>>>>> ources/libc
>>>>>> (and in the past, when we find missing things in Glibc getting them
>>>>>> added to the compiler). Also, this name is technically a misnomer, but we
>>>>>> couldn’t think of a better one (“os” might have been a good one).
>>>>>>
>>>>>> Unfortunately, I think this change alone is really only the tip of
>>>>>> the iceberg. It would be nice to not have it the difference, but we found
>>>>>> we very quickly needed several other layers on top to get to having a
>>>>>> relatively stable cross-platform base:
>>>>>>   https://github.com/apple/swift-package-manager/tree/master/S
>>>>>> ources/POSIX
>>>>>>   https://github.com/apple/swift-package-manager/tree/master/S
>>>>>> ources/Basic
>>>>>>
>>>>>> My hope is that one minimal improvement we can get soon is
>>>>>> multi-package repo support in SwiftPM, which will at least allow us to
>>>>>> share those targets & functionality with other packages.
>>>>>>
>>>>>> > Since I think this got hung up in the past over “what constitutes”
>>>>>> a universal libc, I propose a unified package should just consist of the
>>>>>> functionality that is common between Darwin and Glibc right now, since
>>>>>> those are the only two supported platforms anyway.
>>>>>>
>>>>>> What would the concrete proposal be? It isn’t trivial to determine
>>>>>> that subset and make it well-defined what the exact API is. Is the proposal
>>>>>> to just to pick a standard name, and reexport Darwin and Glibc from it?
>>>>>>
>>>>>
>>>>> I don’t know if it’s actually this simple, but could it just be the
>>>>> symbols that are defined in both modules?
>>>>>
>>>>>
>>>>>>
>>>>>> > Alternatively, Swift could make it a priority to get basic
>>>>>> functionality like file IO and math functions shipped with the compiler,
>>>>>> which would probably cover 98% of cases where people currently import
>>>>>> Darwin/Glibc. A large portion of the standard C libraries are redundant to
>>>>>> the Swift standard library anyway.
>>>>>>
>>>>>> I’m not sure I agree with these statements about the percentages. For
>>>>>> some clients (I’m biased, the areas I work in tend to be in this boat),
>>>>>> what we largely need is good platform-agnostic access to the POSIX APIs.
>>>>>> This is a hard problem to make a good cross-platform API for (esp. if
>>>>>> Windows support is in your head), and which many projects struggle with
>>>>>> (Netscape :: NSPR, LLVM :: libSupport, many many more).
>>>>>>
>>>>>> The sticking point I see is this: if the proposal is just to unify
>>>>>> the name & that doesn’t actually buy us all that much (still need standard
>>>>>> layers on top), then have we really solved enough of a problem to be worth
>>>>>> standardizing on?
>>>>>>
>>>>>> +1 in general agreement with the meta-issue being an important one to
>>>>>> discuss.
>>>>>>
>>>>>
>>>>> There probably is an XY issue at play here; what we *really* need is
>>>>> a way to access the file system built into the standard library. (Math
>>>>> functions are a separate, beleaguered topic for a different thread.) Having
>>>>> support for outputting to `stderr` is also something I’d really like. Going
>>>>> through Glibc/Darwin is just one way to solve this.
>>>>>
>>>>
>>>> _______________________________________________
>>>> 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/20170817/1303c9e6/attachment.html>