Skip to content

Call Swift functions from Rust with ease!

License

Apache-2.0, MIT licenses found

Licenses found

Apache-2.0
LICENSE-APACHE
MIT
LICENSE-MIT
Notifications You must be signed in to change notification settings

Brendonovich/swift-rs

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

swift-rs

Crates.io docs.rs

Call Swift functions from Rust with ease!

Setup

Add swift-rs to your project's dependencies and build-dependencies:

[dependencies]
swift-rs = "1.0.5"

[build-dependencies]
swift-rs = { version = "1.0.5", features = ["build"] }

Next, some setup work must be done:

  1. Ensure your swift code is organized into a Swift Package. This can be done in XCode by selecting File -> New -> Project -> Multiplatform -> Swift Package and importing your existing code.
  2. Add SwiftRs as a dependency to your Swift package and make the build type .static.
let package = Package(
    dependencies: [
        .package(url: "https://github.com/Brendonovich/swift-rs", from: "1.0.5")
    ],
    products: [
        .library(
            type: .static,
        ),
    ],
    targets: [
        .target(
            // Must specify swift-rs as a dependency of your target
            dependencies: [
                .product(
                    name: "SwiftRs",
                    package: "swift-rs"
                )
            ],
        )
    ]
)
  1. Create a build.rs file in your project's root folder, if you don't have one already.
  2. Use SwiftLinker in your build.rs file to link both the Swift runtime and your Swift package. The package name should be the same as is specified in your Package.swift file, and the path should point to your Swift project's root folder relative to your crate's root folder.
use swift_rs::SwiftLinker;

fn build() {
    // swift-rs has a minimum of macOS 10.13
    // Ensure the same minimum supported macOS version is specified as in your `Package.swift` file.
    SwiftLinker::new("10.13")
        // Only if you are also targetting iOS
        // Ensure the same minimum supported iOS version is specified as in your `Package.swift` file
        .with_ios("11")
        .with_package(PACKAGE_NAME, PACKAGE_PATH)
        .link();

    // Other build steps
}

With those steps completed, you should be ready to start using Swift code from Rust!

If you experience the error dyld[16008]: Library not loaded: @rpath/libswiftCore.dylib when using swift-rs with Tauri ensure you have set your Tauri minimum system version to 10.15 or higher in your tauri.config.json.

Calling basic functions

To allow calling a Swift function from Rust, it must follow some rules:

  1. It must be global
  2. It must be annotated with @_cdecl, so that it is callable from C
  3. It must only use types that can be represented in Objective-C, so only classes that derive NSObject, as well as scalars such as Int and Bool. This excludes strings, arrays, generics (though all of these can be sent with workarounds) and structs (which are strictly forbidden).

For this example we will use a function that simply squares a number:

public func squareNumber(number: Int) -> Int {
	return number * number
}

So far, this function meets requirements 1 and 3: it is global and public, and only uses the Int type, which is Objective-C compatible. However, it is not annotated with @_cdecl. To fix this, we must call @_cdecl before the function's declaration and specify the name that the function is exposed to Rust with as its only argument. To keep with Rust's naming conventions, we will export this function in snake case as square_number.

@_cdecl("square_number")
public func squareNumber(number: Int) -> Int {
    return number * number
}

Now that squareNumber is properly exposed to Rust, we can start interfacing with it. This can be done using the swift! macro, with the Int type helping to provide a similar function signature:

use swift_rs::swift;

swift!(fn square_number(number: Int) -> Int);

Lastly, you can call the function from regular Rust functions. Note that all calls to a Swift function are unsafe, and require wrapping in an unsafe {} block or unsafe fn.

fn main() {
    let input: Int = 4;
    let output = unsafe { square_number(input) };

    println!("Input: {}, Squared: {}", input, output);
    // Prints "Input: 4, Squared: 16"
}

Check the documentation for all available helper types.

Returning objects from Swift

Let's say that we want our squareNumber function to return not only the result, but also the original input. A standard way to do this in Swift would be with a struct:

struct SquareNumberResult {
    var input: Int
    var output: Int
}

We are not allowed to do this, though, since structs cannot be represented in Objective-C. Instead, we must use a class that extends NSObject:

class SquareNumberResult: NSObject {
    var input: Int
    var output: Int

    init(_ input: Int, _ output: Int) {
        self.input = input;
        self.output = output
    }
}

Yes, this class could contain the squaring logic too, but that is irrelevant for this example

An instance of this class can then be returned from squareNumber:

@_cdecl("square_number")
public func squareNumber(input: Int) -> SquareNumberResult {
    let output = input * input
    return SquareNumberResult(input, output)
}

As you can see, returning an NSObject from Swift isn't too difficult. The same can't be said for the Rust implementation, though. squareNumber doesn't actually return a struct containing input and output, but instead a pointer to a SquareNumberResult stored somewhere in memory. Additionally, this value contains more data than just input and output: Since it is an NSObject, it contains extra data that must be accounted for when using it in Rust.

This may sound daunting, but it's not actually a problem thanks to SRObject<T>. This type manages the pointer internally, and takes a generic argument for a struct that we can access the data through. Let's see how we'd implement SquareNumberResult in Rust:

use swift_rs::{swift, Int, SRObject};

// Any struct that is used in a C function must be annotated
// with this, and since our Swift function is exposed as a
// C function with @_cdecl, this is necessary here
#[repr(C)]
// Struct matches the class declaration in Swift
struct SquareNumberResult {
    input: Int,
    output: Int 
}

// SRObject abstracts away the underlying pointer and will automatically deref to
// &SquareNumberResult through the Deref trait
swift!(fn square_number(input: Int) -> SRObject<SquareNumberResult>);

Then, using the new return value is just like using SquareNumberResult directly:

fn main() {
    let input = 4;
    let result = unsafe { square_number(input) };

    let result_input = result.input; // 4
    let result_output = result.output; // 16
}

Creating objects in Rust and then passing them to Swift is not supported.

Optionals

swift-rs also supports Swift's nil type, but only for functions that return optional NSObjects. Functions returning optional primitives cannot be represented in Objective C, and thus are not supported.

Let's say we have a function returning an optional SRString:

@_cdecl("optional_string")
func optionalString(returnNil: Bool) -> SRString? {
    if (returnNil) return nil
    else return SRString("lorem ipsum")
}

Thanks to Rust's null pointer optimisation, the optional nature of SRString? can be represented by wrapping SRString in Rust's Option<T> type!

use swift_rs::{swift, Bool, SRString};

swift!(optional_string(return_nil: Bool) -> Option<SRString>)

Null pointers are actually the reason why a function that returns an optional primitive cannot be represented in C. If this were to be supported, how could a nil be differentiated from a number? It can't!

Complex types

So far we have only looked at using primitive types and structs/classes, but this leaves out some of the most important data structures: arrays (SRArray<T>) and strings (SRString). These types must be treated with caution, however, and are not as flexible as their native Swift & Rust counterparts.

Strings

Strings can be passed between Rust and Swift through SRString, which can be created from native strings in either language.

As an argument

import SwiftRs

@_cdecl("swift_print")
public func swiftPrint(value: SRString) {
    // .to_string() converts the SRString to a Swift String
    print(value.to_string())
}
use swift_rs::{swift, SRString, SwiftRef};

swift!(fn swift_print(value: &SRString));

fn main() {
    // SRString can be created by simply calling .into() on any string reference.
    // This will allocate memory in Swift and copy the string
    let value: SRString = "lorem ipsum".into();

    unsafe { swift_print(&value) }; // Will print "lorem ipsum" to the console
}

As a return value

import SwiftRs

@_cdecl("get_string")
public func getString() -> SRString {
    let value = "lorem ipsum"

    // SRString can be created from a regular String
    return SRString(value)
}
use swift_rs::{swift, SRString};

swift!(fn get_string() -> SRString);

fn main() {
    let value_srstring = unsafe { get_string() };

    // SRString can be converted to an &str using as_str()...
    let value_str: &str = value_srstring.as_str();
    // or though the Deref trait
    let value_str: &str = &*value_srstring;

    // SRString also implements Display
    println!("{}", value_srstring); // Will print "lorem ipsum" to the console
}

Arrays

Primitive Arrays

Representing arrays properly is tricky, since we cannot use generics as Swift arguments or return values according to rule 3. Instead, swift-rs provides a generic SRArray<T> that can be embedded inside another class that extends NSObject that is not generic, but is restricted to a single element type.

import SwiftRs

// Argument/Return values can contain generic types, but cannot be generic themselves.
// This includes extending generic types.
class IntArray: NSObject {
    var data: SRArray<Int>

    init(_ data: [Int]) {
        self.data = SRArray(data)
    }
}

@_cdecl("get_numbers")
public func getNumbers() -> IntArray {
    let numbers = [1, 2, 3, 4]

    return IntArray(numbers)
}
use swift_rs::{Int, SRArray, SRObject};

#[repr(C)]
struct IntArray {
    data: SRArray<Int>
}

// Since IntArray extends NSObject in its Swift implementation,
// it must be wrapped in SRObject on the Rust side
swift!(fn get_numbers() -> SRObject<IntArray>);

fn main() {
    let numbers = unsafe { get_numbers() };

    // SRArray can be accessed as a slice via as_slice
    let numbers_slice: &[Int] = numbers.data.as_slice();

    assert_eq!(numbers_slice, &[1, 2, 3, 4]);
}

To simplify things on the rust side, we can actually do away with the IntArray struct. Since IntArray only has one field, its memory layout is identical to that of SRArray<usize>, so our Rust implementation can be simplified at the cost of equivalence with our Swift code:

// We still need to wrap the array in SRObject since
// the wrapper class in Swift is an NSObject
swift!(fn get_numbers() -> SRObject<SRArray<Int>>);

NSObject Arrays

What if we want to return an NSObject array? There are two options on the Swift side:

  1. Continue using SRArray and a custom wrapper type, or
  2. Use SRObjectArray, a wrapper type provided by swift-rs that accepts any NSObject as its elements. This can be easier than continuing to create wrapper types, but sacrifices some type safety.

There is also SRObjectArray<T> for Rust, which is compatible with any single-element Swift wrapper type (and of course SRObjectArray in Swift), and automatically wraps its elements in SRObject<T>, so there's very little reason to not use it unless you really like custom wrapper types.

Using SRObjectArray in both Swift and Rust with a basic custom class/struct can be done like this:

import SwiftRs

class IntTuple: NSObject {
    var item1: Int
    var item2: Int

    init(_ item1: Int, _ item2: Int) {
       self.item1 = item1
       self.item2 = item2
    }
}

@_cdecl("get_tuples")
public func getTuples() -> SRObjectArray {
    let tuple1 = IntTuple(0,1),
        tuple2 = IntTuple(2,3),
        tuple3 = IntTuple(4,5)

    let tupleArray: [IntTuple] = [
        tuple1,
        tuple2,
        tuple3
    ]

    // Type safety is only lost when the Swift array is converted to an SRObjectArray
    return SRObjectArray(tupleArray)
}
use swift_rs::{swift, Int, SRObjectArray};

#[repr(C)]
struct IntTuple {
    item1: Int,
    item2: Int
}

// No need to wrap IntTuple in SRObject<T> since
// SRObjectArray<T> does it automatically
swift!(fn get_tuples() -> SRObjectArray<IntTuple>);

fn main() {
    let tuples = unsafe { get_tuples() };

    for tuple in tuples.as_slice() {
        // Will print each tuple's contents to the console
        println!("Item 1: {}, Item 2: {}", tuple.item1, tuple.item2);
    }
}

Complex types can contain whatever combination of primitives and SRObject<T> you like, just remember to follow the 3 rules!

Bonuses

SRData

A wrapper type for SRArray<T> designed for storing u8s - essentially just a byte buffer.

Tighter Memory Control with autoreleasepool!

If you've come to Swift from an Objective-C background, you likely know the utility of @autoreleasepool blocks. swift-rs has your back on this too, just wrap your block of code with a autoreleasepool!, and that block of code now executes with its own autorelease pool!

use swift_rs::autoreleasepool;

for _ in 0..10000 {
    autoreleasepool!({
        // do some memory intensive thing here
    });
}

Limitations

Currently, the only types that can be created from Rust are number types, boolean, SRString, and SRData. This is because those types are easy to allocate memory for, either on the stack or on the heap via calling out to swift, whereas other types are not. This may be implemented in the future, though.

Mutating values across Swift and Rust is not currently an aim for this library, it is purely for providing arguments and returning values. Besides, this would go against Rust's programming model, potentially allowing for multiple shared references to a value instead of interior mutability via something like a Mutex.

License

Licensed under either of

at your option.

Contribution

Unless you explicitly state otherwise, any contribution intentionally submitted for inclusion in the work by you, as defined in the Apache-2.0 license, shall be dual licensed as above, without any additional terms or conditions.