- Well, the new iPhone has offically launched with today's store release. Naturally, RustyBrick was on hand to get our brand new iPhones.
- Most programs on Linux are dynamically linked. So when you are creating a unikernel with OPS out of a linux application OPS goes out and finds all the libraries it's dynamically linked to and throws it onto the disk image. This works well if you are on linux - 99% of everything on linux is dynamically linked.
- View Rusty Folsom’s profile on LinkedIn, the world's largest professional community. Rusty has 7 jobs listed on their profile. Technical Services Administrator.
Rust is a systems programming language focused on speed and safe concurrency, and which I’ve been using for personal projects heavily since the 1.0 release last year. Most of these projects have been replacements for existing scripts in my workflows or new command line tools, but I wanted to create a Mac application and determine if it would benefit from Rust’s memory efficiency, safety, and robust library ecosystem.
Our Rusty Mac Adopt a Teen Program has 481 teens this year! At just $100, you can make our teenagers feel special and loved during this busy holiday season. For more information. For a MP 5,1 the easiest way is to upgrade to the current Mac OS version which is Big Sur. Keep in mind that there is a lot more to IT security than an updated OS (the most important factor in any security chain is the user), but it means your MP 5,1 is roughly at a same security level as a newer Mac.
I’ve done iOS and Mac application development for many years and it's worth noting that the hardest part of Cocoa development has always been learning the frameworks rather than the languages. This experiment is about applying Cocoa and Rust knowledge to create something safe and yet easy to work with.
Getting started with Cocoa crates
There are already crates for working with the Objective-C runtime, such as the [CODE]objc[/CODE] and [CODE]block[/CODE] crates, which are for using the runtime directly and interfacing with Apple’s block extensions respectively. The [CODE]objc[/CODE] crate in particular provides the [CODE]msg_send![/CODE] macro, which is a basic interface to messaging Objective-C objects. Here’s an example of creating an [CODE]NSObject[/CODE]:
-- CODE language-rust --
unsafe {
let cls = Class::get('NSObject').unwrap();
let obj: *mut Object = msg_send![cls, new];
}
The [CODE]cocoa[/CODE] crate builds on this to provide an interface to using frameworks including AppKit for drawing windows and views onscreen. It also has an interesting take on implementing Objective-C classes in that translates them to traits which are implemented by a generic [CODE]NSObject[/CODE] type. This snippet creates an app and a window, and presents it on screen:
-- CODE language-rust --
unsafe {
let _pool = NSAutoreleasePool::new(nil);
let app = NSApp();
app.setActivationPolicy_(NSApplicationActivationPolicyRegular);
let window = NSWindow::alloc(nil).initWithContentRect_styleMask_backing_defer_(
NSRect::new(NSPoint::new(0., 0.), NSSize::new(200., 200.)),
NSTitledWindowMask as NSUInteger,
NSBackingStoreBuffered,
NO
).autorelease();
let title = NSString::alloc(nil).init_str('Hello World!');
window.setTitle_(title);
window.makeKeyAndOrderFront_(nil);
app.run();
}
Pretty cool, though as is, the entire interface is unsafe, missing the hopeful goal of the experiment. This approach could still be interesting when writing the application core code in Rust, and then packaging it using Cocoa bindings.
Wrapping Cocoa APIs in “safety”
Given those caveats, couldn’t we create Rust wrappers for Objective-C classes? Of course! After some trial and error, I had a base trait to use for wrapping and interacting with Objective-C objects:
-- CODE language-rust --
use objc::runtime::Object;
pub type Id = *mut Object;
pub trait ObjCClass: Sized {
/// Returns pointer to underlying objc object
fn ptr(&self) -> Id;
/// Creates an instance from an objc object pointer, failing
/// if the pointer is not an instance of the wrapped class
fn from_ptr(ptr: Id) -> Option<self>;</self>
/// The printed name of the class
fn class_name() -> &'static str;
/// Type-safe reference to an instance with a nil pointer
fn nil() -> Self;
/// Performs an `isKindOfClass` check to whether a particular
/// pointer is an instance of the wrapped class
fn ptr_is_class(ptr: Id) -> bool;
/// Change an instance of one class into another, failing if
/// the pointer is not an instance of the preferred class.
/// Useful for converting between inherited classes e.g.
/// NSDictionary to NSMutableDictionary.
fn coerce<t: objcclass='>(&self) -> Option<t> {</t></t:>
T::from_ptr(self.ptr())
}
/// Designate this instance as suitable for being released
/// once it is out of scope
fn autorelease(&self) -> Self;
/// Drop the Objective-C reference. The object is then invalid
fn release(&mut self);
}
Note that this creates a Rust object with a reference to an Objective-C object, but the overall effect is minimal as most interaction still happens in Objective-C runtime land.
Using this trait was most easily done creating a handy macro named [CODE]impl_objc_class[/CODE], and then wrapping the average class became easy! Here’s an example which wraps a few methods on [CODE]NSString[/CODE].
-- CODE language-rust --
const UTF8_ENCODING: NSUInteger = 4;
impl_objc_class!(NSString);
impl NSString {
/// Creates an `NSString` from a `str`.
pub fn from(content: &str) -> Self {
let ptr: *mut Object = unsafe {
let string: *mut Object = msg_send![class!('NSString'), alloc];
msg_send![string, initWithBytes:content.as_ptr()
length:content.len()
encoding:UTF8_ENCODING]
};
NSString { ptr: ptr }
}
/// The length of the string as measured in UTF-8 code points
pub fn len(&self) -> usize {
unsafe { msg_send![self.ptr, lengthOfBytesUsingEncoding:UTF8_ENCODING] }
}
}
The class can now be used directly, and without [CODE]unsafe[/CODE]:
-- CODE language-rust --
let greeting = NSString::from('hello');
assert_eq!(greeting.len(), 5);
Resources still need to be released (or auto-released, if applicable) when they are no longer needed, but classes became much easier to use. I explored some options such as implementing a [CODE]Drop[/CODE] trait to automatically discard Objective-C objects once the Rust reference goes out of scope, but this behavior is not always desirable, especially when working with references to applications and windows which are expected to stay for the life time of the application, or at least longer than the current scope.
Packaging Rust into an app
While we can use the snippets of the cocoa crate to run an executable, the executable is not packaged as an app bundle, which would be suitable for having an app icon, putting an app in the dock, or being registered as a default application (like being the mail client used for [CODE]mailto:[/CODE] links, for example). For that, we’d need to package the executable into an app bundle.
An easy way to create an app bundle which launches Rust code is to create a Cocoa app with a Rust and dependent app target. This requires a few steps in Xcode:
- Create a new app using the Cocoa app template
- Add a second “External build system” target to the application which creates the Rust binary
- Add the second target to the default app target as a dependency
- Add the rust executable as a bundled resource of the app target
- Replace the default AppDelegate with a script to launch the Rust binary, something like this bit of Swift:
-- CODE language-rust --
let task = Process()
task.launchPath = Bundle.main.path(forResource: 'my-rust-program', ofType: nil)
task.launch()
task.waitUntilExit()
Rusty Macomb
I’ve created an example which shows all of these parts in action, adds an app icon, and pipes output from the Rust executable to the system console.
Conclusions
Rust Mac Os
The initial results were less than ergonomic when using the existing Cocoa crate since the interface did not add additional safety, and perhaps removed some because the generic object type conformed to every Cocoa class trait. I could (and did) call the wrong methods on Cocoa class instances.
Writing my own layer of classes on top of [CODE]objc[/CODE] improved the latter, though it was more initial overhead to write wrappers before using classes, and still felt clumsy when converting between values in class clusters for example. There is potential for a “Rustier” crate for interfacing with Objective-C, or a generator which makes ergonomic method names. Despite this, I mapped a number of Objective-C classes by hand, and while my stylistic choices probably aren’t suitable for a general use library, Rust+Cocoa became very fast to use and iterate on ideas. The approach could be worth a try if you have reusable components in Rust to share with a Cocoa application, and have constructs unsuitable for use with the foreign function interface.
There’s more I could cover here about the experience, like how to declare your own Objective-C classes in Rust and implementing protocols, but that should be the topic of a later post.
I’ve made some longer examples demonstrating the snippets in this post as well as a general template usable for packaging a mac app, which is available on GitHub.
Thanks for reading!