[SR-7865] swift-corelibs-foundation does not compile on macOS after 'as' bridging was merged

[millenomi]

The new primitives are only available if Objective-C isn't. Guard them with #canImport(ObjectiveC).

(Additionally, see below for discussion.)

Resolves SR-7865.

[millenomi]

@swift-ci please test

[millenomi]

cc @spevans

[spevans]

Did the tests work in Xcode? I made the same basic change but some of the tests for NSJSONSerialization failed with issues converting some nil and Optional<Any> values.

…

On 4 Jun 2018, at 15:39, Lily Vulcano ***@***.***> wrote: The new primitives are only available if Objective-C isn't. Guard them with #canImport(ObjectiveC). Resolves SR-7865. You can view, comment on, or merge this pull request online at: #1579 Commit Summary Guard the use of the new stdlib primitives with #canImport(ObjectiveC). File Changes M Foundation/Bridging.swift (7) M Foundation/NSError.swift (5) Patch Links: https://github.com/apple/swift-corelibs-foundation/pull/1579.patch https://github.com/apple/swift-corelibs-foundation/pull/1579.diff — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[spevans]

Ignore that last message, I see now that your change is different to what I tried.

…

On 4 Jun 2018, at 15:39, Lily Vulcano ***@***.***> wrote: The new primitives are only available if Objective-C isn't. Guard them with #canImport(ObjectiveC). Resolves SR-7865. You can view, comment on, or merge this pull request online at: #1579 Commit Summary Guard the use of the new stdlib primitives with #canImport(ObjectiveC). File Changes M Foundation/Bridging.swift (7) M Foundation/NSError.swift (5) Patch Links: https://github.com/apple/swift-corelibs-foundation/pull/1579.patch https://github.com/apple/swift-corelibs-foundation/pull/1579.diff — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub, or mute the thread.

[pushkarnk]

@millenomi @spevans Can we have this merged please?

[millenomi]

I forgot to run tests in Xcode; once I do today.

[millenomi]

@swift-ci please test

[millenomi]

Here's what I found:

• On Darwin, x as AnyObject can return the native _SwiftValue class, or Foundation.NSNull() — where Foundation is the native framework from /System/Frameworks, not SwiftFoundation for nil values.

The solution is to prevent usage of as AnyObject in the SCF codebase — it is not safe to assume this expression returns a SwiftFoundation.NSObject all the time (as we instead guarantee on Linux). SCF code must use _SwiftFoundation.store() instead to produce a SwiftFoundation.NSObject; this patch uses SwiftFoundation.NSNull() and intercepts and substitutes native _SwiftValue boxes with SwiftFoundation._SwiftValue boxes.

This patch isn't done; we need to make sure that objects passed in by an author work, and to this extent I need to make SwiftFoundation._SwiftValue.fetch() able to unbox native _SwiftFoundation boxes.

[millenomi]

After some experimentation: we can just return the native box from .fetch(), since it will be casted immediately afterward to the desired type.

[millenomi]

@swift-ci please test

[millenomi]

This addresses the latter point.

[millenomi]

cc @spevans

[spevans]

Confirmed working on 2018-06-04 toolchain (macOS 10.13.5, Xcode 9.4) and Ubuntu 16.04.

[millenomi]

@swift-ci please test

[millenomi]

It’s breaking the SwiftPM build.

[millenomi]

Investigating.

[millenomi]

I cannot reproduce locally. Triggering a new build here in case it was transient.

[millenomi]

@swift-ci please test

[millenomi]

Found the issue:

We are implementing NSNull roundtrip in _SwiftValue.fetch(), but it is spurious. Darwin does not unpack NSNulls unless they are explicitly casted to an optional type:

let x: [Int?] = [nil]
print((x as [AnyObject]) as [Any]) // -> [<null>]
print(((x as [AnyObject]) as [Any]).map { type(of: $0) }) // -> [NSNull]

Since bridging already works on both platforms to unpack it in that case, we may want to stop explicitly unpacking it in our case. However, we need to figure out how to treat SwiftFoundation.NSNull in that case, since Darwin bridging will not unpack it. (A solution could be to make it _ObjectiveCBridgeable on Darwin only.)

[millenomi]

I'm going to check if us not unpacking is enough to let things be unblocked while we work on a longer term solution (which hinges on whether we desire SCF to be used on Darwin as a general purpose library or not).

[millenomi]

@swift-ci please test