A Swift Runtime library for viewing type info, and the dynamic getting and setting of properties.

The nominal type descriptor metadata is going away completely in 4.2. So there's no longer a way to get field names and types with the method this package uses. There is a new runtime method swift_getFieldAt to retrieve this data, however it uses C++ std::function and is not callable from Swift. You'd need to create a C++ wrapper that exposes a C API to use this. I've opened a PR on Swift to fix this (and make the method easily callable from Swift), but I'm not sure if they will do anything about it: apple/swift#15565. Thought I would share this info, thanks!

(copy of https://github.com/Zewo/Reflection/issues/14)

Looks like something is wrong/different w/ the RC here.

Test Case '-[RuntimeMacTests.FactoryTests testClass]' started.
(lldb) bt
* thread #1, queue = 'com.apple.main-thread', stop reason = EXC_BAD_ACCESS (code=EXC_I386_GPFLT)
  * frame #0: 0x0000000108bfea2d libswiftCore.dylib`_swift_release_dealloc + 13
    frame #1: 0x0000000103d362d4 RuntimeMacTests`FactoryTests.testClass(self=0x000000010390e4a0) at FactoryTests.swift:58
    frame #2: 0x0000000103d368d5 RuntimeMacTests`@objc FactoryTests.testClass() at FactoryTests.swift:0
    frame #3: 0x00007fffb972f0cc CoreFoundation`__invoking___ + 140
    frame #4: 0x00007fffb972ef51 CoreFoundation`-[NSInvocation invoke] + 289
    frame #5: 0x000000010034eb36 XCTest`__24-[XCTestCase invokeTest]_block_invoke.272 + 50

Getting some build errors in Swift 5, in Metadata.swift and TypeInfo.swift:

    var typeInfoConvertible: TypeInfoConvertible
    
    switch kind {
    case .struct:
        typeInfoConvertible = StructMetadata(type: type) // error: Missing argument for parameter 'metadata' in call

(Those errors pop up for every case.)

That init is defined here: https://github.com/wickwirew/Runtime/blob/master/Sources/Runtime/Metadata/MetadataType.swift#L61

extension MetadataType {
    // ...
    init(type: Any.Type) {
        let base = metadataPointer(type: type)
        let metadata = base.advanced(by: valueWitnessTableOffset).raw.assumingMemoryBound(to: Layout.self)
        self.init(type: type, metadata: metadata, base: base)
    }
}

so I'm not sure why Swift 5 is refusing to see it…

Note this happens when building the project from scratch, or when building as an SPM dependency.

Troubleshooting notes:

• I can resolve the "missing argument" errors if I change the inits to typeInfoConvertible = StructMetadata.init(type: type).
• However, StructMetadata.swift fails with a error: reference to invalid associated type 'Layout' of type 'StructMetadata', on the init declaration itself.
• Adding manual typealias Layout = FooMetadata declarations fixes those errors, and allowed me to remove the manual Foo.init(type: type)…
• but then I get our old friend

Undefined symbols for architecture x86_64:
  "_swift_getFieldAt", referenced from:
      closure #1 (Swift.Int) -> Runtime.PropertyInfo in Runtime.getProperties(of: Any.Type, offsets: [Swift.Int]) -> [Runtime.PropertyInfo] in GetFieldAt.o
ld: symbol(s) not found for architecture x86_64

• Huh… swift build from the terminal completes just fine; is this only a problem in Xcode? Is there a problem with my deployment target? I've rebuilt the .xcodeproj with swift package generate-xcodeproj…

Anywho, here's the branch if you wanna take a look: https://github.com/noahemmet/Runtime/tree/swift5

This moves cruntime to a separate module, which allows this framework to be built on Linux. Discussion: https://github.com/wickwirew/Runtime/issues/18#issuecomment-434533969 and https://github.com/noahemmet/Runtime/commit/8d685eb4ff7be23385bac54c320aaa71117b38e3

• It looked like the last few commits included *.xcodeproj and *.xcworkspace files, but if you'd like, I'm happy to add them to the .gitignore.
• Tests pass locally on my Mac but I'm not set it up test on Linux; is that something Travis will do automatically?

Thanks for everyone's help on this!

I hope those do not confuse the Cocoapod.

This stuff makes it build as a Cocoa framework. Building things works fine, but the tests actually crash on macOS (same w/ spm test btw).

I installed Runtime 2.1.0 using Cocoapods. When I opened Xcode and ran a clean build, I got the following build error in Runtime's Factory.swift: Use of unresolved identifier 'swift_allocObject.

Here are the details on my environment:

• Xcode 10.2 (10E125)
• Command Line Tools: Xcode 10.2
• Swift version: Apple Swift version 5.0 (swiftlang-1001.0.69.5 clang-1001.0.46.3)
• Cocoapods: 1.5.3 (also tried 1.6.1 with the same results)
• Runtime version: 2.1.0 (also tried 2.0.0 with the same results)

Here are the troubleshooting steps I've tried so far:

• Removed the post_install from my Podfile, and ran pod install again after deleting the Pods/ directory and Podfile.lock
• Tried different versions of Runtime, and 2 versions of Cocoapods (described above)
• Deleted DerivedData

Screenshot for more context:

As described in https://github.com/wickwirew/Runtime/issues/92, swift 5.4 change the ClassMetadataLayout for platforms that do not support Objective-C interoperability. This resulted in Runtime crashing when running on those platforms with the newest swift releases.

This PR fixes this issue by adjusting the ClassMetadataLayout on those platforms. I'm using the #if !swift(>=5.4) || canImport(Darwin). Not entirely sure if canImport(Darwin) is the proper check for availability of Objective-C interoperability (works though). Happy for any feedback.

I have another PR ready which adds some CI integration with GitHub to test the framework on all the available linux distortions. This PR passes all tests on all platforms: Actions Run. I chose to do the CI PR separately so this fix can get merged faster.

Hi,

I've been trying to get the GraphQL related https://github.com/GraphQLSwift/Graphiti project upgraded to Swift5 using the latest version of this library and came across a crash running the tests. The crash occurs in NominalMetadataType.genericArguments() when trying to get the TypeInfo for a simple Struct. I noticed that the Struct was declared nested within a closure. It appears that NominalMetadataType.genericArguments() is coming back with a non-zero size results when it ought to be zero, and sometimes crashes.

To try and reproduce the problem within this project directly I've added a new test to MetadataTests.swift called testNestedStruct.

Please see this branch on my fork https://github.com/SportlabsTechnology/Runtime/tree/investigateCrash

Weirdly the addition of this new test causes the existing testStruct test to detect a non-zero amount of genericArguments and fail.

Any ideas ands help appreciated. Thanks, Mike.

class User: NSObject {
    let id: Int = 0
    let username: String = ""
    let email: String = ""
}

let info = try typeInfo(of: User.self) // crash occurs 
let property = try info.property(named: "username")
print("name: \(property.name)")
print("type: \(property.type)")

how fix this problem? and is there any plan to support objective-c language?

Swift 4.1 apparently includes ABI changes that cause this library to fail completely. ~All tests crash with EXC_BAD_ACCESS.

Instructions for grabbing the latest Swift 4.1 snapshot here:

https://gist.github.com/tanner0101/cdb77c7f58d53af2ba2da5d39415389a

This PR aims to fix issue #64, which reports a memory leak when setting a field which stores a reference type.

Unfortunately, I'm not very familiar with raw memory manipulation, so I'm not 100% sure whether this is the correct thing to do in all cases, but it appears to address the issue when I apply this fix in our project.

I wasn't able to run the tests locally, but I'm happy to look into any tests this PR breaks.

enum MyEnum: String {
    case A = "test"
    case B
    case C
}

struct Name {
    var a: MyEnum
    var b: MyEnum
}

Name(a: .A, b: .B)

let info = try! typeInfo(of: MyEnum.self)
print(info)

TypeInfo(kind: Runtime.Kind.enum, name: "MyEnum", type: SwiftDump.MyEnum, mangledName: "MyEnum", properties: [], inheritance: [], size: 1, alignment: 1, stride: 1, cases: [Runtime.Case(name: "A", payloadType: nil), Runtime.Case(name: "B", payloadType: nil), Runtime.Case(name: "C", payloadType: nil)], numberOfEnumCases: 3, numberOfPayloadEnumCases: 0, genericTypes: [])

This is probably just a misunderstanding of mine, but I thought the payload for the A case would be "test"?

I couldn't find such information anywhere and I'm a bit stuck on this one issue. I've been trying to write a new closure over an existing closure but I haven't been successful.

This is how you can reproduce it.

func closureReplacementTest() {
    
    do {
        
        // Initialize sample data
        var freshFoo = Foo()
        let freshBar = Bar()
        
        // Work on getting the property
        let objectToTest = try typeInfo(of: Foo.self)
        let property = try objectToTest.property(named: "closure")
        
        // Replace the closure to give 100 scores instead
        try property.set(value: { (bar: Bar) in
            bar.score = 100 // Actually crashing
        }, on: &freshFoo)
        
        // Execute what's supposed to be the new closure
        freshFoo.closure(freshBar)
        
        print("Score is: \(freshBar.score)") // Expects it to be 100

    } catch {}
}

I spend a few days reading posts and documentation regarding memory management but unfortunately that's not my forte. =)

I did debug as best as I could and I noticed that during the crash the memory address of the bar variable is different that it would be when I simply run the block directly. The only thing I think it's not working quite well is the address of the variables of the closure because that's when the crash happens as the code itself of the closure seems to run just fine.

Strongly appreciate any input on this, I'll still going to spend more days trying to figure out. Thanks!

Hi, I tried to use the Runtime and faced an issue with an object that inherits from the UIView. Property offsets were wrong. When I dug deeper I found that offsets don't count the instance size of UIView itself. I was able to workaround by modifying ClassMetadata's toTypeInfo() method like this:

mutating func toTypeInfo() -> TypeInfo {
    var info = TypeInfo(metadata: self)
    info.mangledName = mangledName()
    info.properties = properties()
    info.genericTypes = Array(genericArguments())
    
    var _sc: ClassMetadata? = self
    while var sc = _sc?.superClassMetadata()?.asClassMetadata() {
        info.inheritance.append(sc.type)
        let superInfo = sc.toTypeInfo()
        info.properties.append(contentsOf: superInfo.properties)
        _sc = sc
    }
    
    // Fix offset because of the obj-c inheritance if needed.
    // Fix own properties only.
    if let sc = _sc, let superClass = pointer.pointee.superClass as? AnyObject.Type {
        let size = class_getInstanceSize(superClass) - class_getInstanceSize(NSObject.self)
        for i in 0..<numberOfFields() {
            info.properties[i].offset += numericCast(size) // <--- I also made the offset writable to simplify things
        }
    }
    
    return info
}

I doubt it's the proper way of doing it and I was just curious If the issue can be fixed. Anyway, is it possible to add support for Swift objects inherited from objective-c classes?

Hi there! Thanks for an awesome library! Lately I faced the problem with @propertyWrapper. I have tried to create a @propertyWrapper:

@propertyWrapper public struct Blob {
    public init(length: Int) {
        self.length = length
        self.wrappedValue = []
    }
    
    let length: Int
    
    public var wrappedValue: [UInt8]
}

Then I used it in my struct just like this one:

struct MyStruct {
  @Blob(length: 10) var blob: [UInt8]
  ...
}

And then I tried to createInstance , i got length = 0, I could not create instance with default length (=10). How to reflect the propertyWrapper and get the default length (= 10)? Thanks

This PR adds the functionality to initialize enums and tuples using the createInstance function that is currently missing in the runtime framework as documented in #90 .

Testing

Adds unit tests to create enums and tuples