Knowledge of Swift not required. If you know your way around OS software, can reason about the security of the code you write, and are excited about writing exhaustively tested software, we’d love to talk to you.
We’re hiring for roles in kernel/systems and userspace. Like the Platforms SOTU mentioned, we’re using Swift at all layers of the software stack now. https://www.youtube.com/live/yl2jsIoMfDU
I had the pleasure of leading the effort to ship Swift in the Secure Enclave back in 2022. Now I have multiple teams working on accelerating the transition to memory safe languages. We’re showing that with good planning and a relentless focus on testing, we can improve security, performance, and functionality. And we get to have a ton of fun working with some amazing colleagues. It’s the most enjoyable and impactful work I’ve ever done in my career.
For low-level system security, I'm a fan of https://llsoftsec.github.io/llsoftsecbook/LLSoftSecBook.pdf as an overview for any systems developer, not just compiler devs. It might not be the most approachable, but it's got great info on everything memory corruption.
> Operations like filter and map allocate memory, but that allocation is only necessary if the value escapes. The Swift standard library provides .lazy.map and .lazy.filter, but they don’t work in every case. For logic that only iterates over the filter or map, it’s much more efficient to loop with continue (or use for … in … where) and transform elements into local variables as necessary.
It does feel like a compiler/optimiser failure to have to rewrite those cases.
Beware: As of a few months ago, when I tried to use the lifetime features shown off in this post, I ran into constant compiler crashes with very simple programs, until I gave up and wrote off the features as unusable. This happened on both stable and nightly compilers. I guess they work well enough for this TrueType interpreter, but I suspect they’re using a narrow subset of what the features are supposed to support. Or maybe things have been fixed very recently.
That said, I’m looking forward to using Swift lifetimes once they actually work!
They’ve been doing it for years. I don’t remember how we first knew, but I know they’ve been using Swift in kernels for at least some of the other chips like the Secure Enclave or whatever.
I’m not sure exactly which. I assume it’s some of the code and not all. But it’s not new in the abstract.
That said I don’t think I’ve heard of it in the kernel of MacOS on the main processor. That may be new.
Either way this is certainly the most concrete announcement I remember them ever giving on this stuff.
I know internally they use an IPsec implementation written by Rust (I think in the iCloud infra). Heard this from an ex-Apple engineer Ben (forgot his last name) that did a wonderful presentation of Rust from first principles. He said that it was hard to get people in on Rust when most would argue for Swift.
Some stuff was discussed at Meet with Apple security event a few months ago, and the talks on FoundationDB rewrite, or why Swift Embedded subset came to be.
However I miss them actually having had one of those 15 - 30m WWDC sessions, where they could have gone a bit deeper into the keynote examples
Curious the direction of Webkit as there was a nebulous mention of select portions being rewritten from C++ to Swift. And yet, the new ECMAScript module (ESM) loader for Safari 27 is implemented in C++ (https://webkit.org/blog/17967/news-from-wwdc26-webkit-in-saf...).
Some corporations prefer Apache 2.0 for projects where they'll be accepting contributions, because it includes patent protection and retaliation clauses. In case like this, where source code is just being published for reference and contributions aren't accepted, those risks don't exist.
Given the age of TrueType, wouldn't nearly all patents be expired already?
Apache2's license I've heard described as mutually-assured-patent-destruction - if you use the code and make a patent claim, your rights to use the code go away.
So Apache2 offers little benefit here, and MIT may get it into more hands?
Russinovitch (Azure's CTO/CISO) gave a speech at RustConf 2025 and mentions it(DirectWriteCore) took 2 engineers 6 months resulting in 154K LOC and 5-15 percent performance increase for font shaping. https://www.youtube.com/watch?v=uDtMuS7BExE&list=PL2b0df3jKK...
One of the genius things about Swift is its interop with Objective C. Made the switch over considerably easier for developers. I’m not sure what that looks like in a Rust world.
Rust is also just a more complex language. I’m not convinced the benefits would have been worth it.
Not just interoperability with Objective C but with C (full) and C++ (increasingly better but not full) as well.
Swift is also interoperable with different versions of itself courtesy of the Swift stable ABI (Application Binary Interface)[0], which they invested a significant amount of time into at the expense of adding other new features to the language, which have come along later.
Rust offers a different approach: recompile everything and static linking.
These days I mainly write Rust but I did write a semi complex iOS app and enjoyed Swift. I just didn't love how slow the type checker was and how it got lost. I recall having to break things into smaller bits to help the compiler, and there were some oddities about the language.
The gap between the two languages is quite small, it just makes me wish Apple was also all-in on Rust
In the last year they’ve added improvements to the type checker to speed it up, those would have been released now.
They have further and much more significant changes that I think might have recently landed in the development version. That should make an even bigger difference. But it’s not in a released version yet.
And yes, none of us like that one part of Swift. Especially the DRASTIC difference compared to objective-C which really only checked syntax and little else.
It’s still probably my favorite language right now though I don’t get to write in it much.
There’s no reason to invent your own head canon, the influence was openly acknowledged when Swift was new and it continues now that the language is developed out in the open (see Swift Ownership Manifesto)
I believe Swift tends to use reference counting and copy-on-write strategies. This, like GC, is less for the programmer to think about and doesn't require the semantic checks, but sometimes the performance cost is unacceptable compared to what you'd write in Rust.
You can choose to use either refcounting or unique ownership for your types. For most use cases, refcounted (+ copy-on-write) is the best choice and is the default, but the truetype interpreter made extensive use of non-refcounted types to achieve this performance.
They have either recently added or talked about a borrow style system in the language as a way to avoid more copies and speed things up/lower memory usage/help with asynchronous programming.
Swift takes the right approach in language ergonomics, many of its use cases would be much harder with explicit .clone() all over the place, lack of back references, no standard ABI for binary libraries, interop with Objective-C and C++, no standard concurrency runtime, or error handling types.
Rust 2026 roadmap has language ergonomics on it for a reason.
That said, outside Apple ecosystem you probably better with Rust, or if one has no GC issues, OCaml, Haskell, F#, Scala, C#.
Apple has a runtime system that looks a lot like smalltalk. Everything at that layer is dynamic. They needed a language that could seamlessly interact with that system.
What's funny is from 2023 (I think), macOS just draws the UI unhinted. You have a 1080p display and you don't want to see the letters in the UI blurred to death? Tough luck, 1080p is incompatible with macOS, everybody needs "retina", and nobody cares that Windows and all Linux DEs look on 1080p just fine.
It looks like this hinter will be used only in rendering PDFs, because that's where they test the performance.
While hinting is disabled for most fonts, there are some fonts that require hinting to render correctly. We have to support hinting for those fonts, and it was easier to make it secure by rewriting hinting in Swift than it would have been to comprehensively identify every font created by those foundries.
People also use usable mice instead of touchpads, and they put the "ctrl" key where Apple thinks a useless "fn" should be. All kinds of things happen outside Apple world.
To me, it's more about what I'm used to. I have a perfectly fine several years-old monitor, so why should I throw it away?
Then let me blow your mind:) One of my daily drivers is an ex-Chromebook at 1366x768. Granted, it's also physically smallish so the DPI isn't quite as low as a macbook would be with those pixels, but still. And that's a touch cramped but it's fine.
The problem is, as soon as you are not on a Mac but Linux or Windows, you are in for an awful, truly awful lot of pain. HiDPI support is a mess because even in the rare case applications are made with HiDPI in mind they are not tested on HiDPI machines.
Other way around, most Mac software is not tested how it behaves on inferior external monitors.
What kind of windows programs are these? HiDPI is more than a decade old. A desktop application, no matter what OS it is, should always be tested with different scaling factors.
macOS has been drawing unhinted text for an eternity, and for those who can tolerate it on low-DPI screens, it's a great thing: the letter shapes look the same at all sizes, and the spacing between letters is consistent at all sizes.
I'm a high DPI snob so I haven't used a low res monitor for work in forever, but isn't the entire point of font hinting to make the text more legible at smaller pixel grid sizes? Yes, of course the shapes are more consistent since the hinter isn't touching them, but isn't the end result just less legible text?
Hinting purposefully destroys letter shapes in exchange for crispness. People who like macOS style font rendering prioritize letter shapes faithful to the font designer more than crispness.
Whether good letter shapes is more legible or crisper text is more legible is basically subjective. In the 2000s before HiDPI became popular different people really thought one was more legible than the other and vice versa. HiDPI made this basically moot.
I'm surprised the code has visible LLM smells. Though, I shouldn't be surprised. I hope the important bits are still human-controlled (and the same for Apple's many operating systems that absolutely deserve to remain stable and understood).
It would be interesting to see their internal guidance on LLM use. It’s a massive amount of new power that has to be wielded carefully. That kind of guidance might mean the survival or downfall of some big corps in the next few years.
I love how it doesn't even support reasoning output or edit tool diffs. (Yes it can show diffs sometimes when using the 100% official flow but not when using ACP)
I think these are the types of things Apple should've focused on instead of half-heartedly barging ahead with SwiftUI and breaking the language in the process
There's mention at the end. The models (and Swift itself!) have evolved a lot since this project started, so the early code is largely hand-rolled and the later changes were mostly authored by centaurs (to steal a term from chess).
But I personally reviewed every line that shipped and was absolutely insufferable about testing.
If you want to help improve the security of OS software through the magic of memory safe languages, the team that did this work is hiring: https://jobs.apple.com/en-us/search?search=Spear&sort=releva...
Knowledge of Swift not required. If you know your way around OS software, can reason about the security of the code you write, and are excited about writing exhaustively tested software, we’d love to talk to you.
We’re hiring for roles in kernel/systems and userspace. Like the Platforms SOTU mentioned, we’re using Swift at all layers of the software stack now. https://www.youtube.com/live/yl2jsIoMfDU
I had the pleasure of leading the effort to ship Swift in the Secure Enclave back in 2022. Now I have multiple teams working on accelerating the transition to memory safe languages. We’re showing that with good planning and a relentless focus on testing, we can improve security, performance, and functionality. And we get to have a ton of fun working with some amazing colleagues. It’s the most enjoyable and impactful work I’ve ever done in my career.
Any resources you'd recommend for learning about reasoning about the security of code?
For low-level system security, I'm a fan of https://llsoftsec.github.io/llsoftsecbook/LLSoftSecBook.pdf as an overview for any systems developer, not just compiler devs. It might not be the most approachable, but it's got great info on everything memory corruption.
> Operations like filter and map allocate memory, but that allocation is only necessary if the value escapes. The Swift standard library provides .lazy.map and .lazy.filter, but they don’t work in every case. For logic that only iterates over the filter or map, it’s much more efficient to loop with continue (or use for … in … where) and transform elements into local variables as necessary.
It does feel like a compiler/optimiser failure to have to rewrite those cases.
I've always wondered if JS engines could rewrite those array functions at compile time, like this: https://github.com/SomeRanDev/Haxe-MagicArrayTools
Though, it probably wouldn't work if user code modified the Array prototype.
Beware: As of a few months ago, when I tried to use the lifetime features shown off in this post, I ran into constant compiler crashes with very simple programs, until I gave up and wrote off the features as unusable. This happened on both stable and nightly compilers. I guess they work well enough for this TrueType interpreter, but I suspect they’re using a narrow subset of what the features are supposed to support. Or maybe things have been fixed very recently.
That said, I’m looking forward to using Swift lifetimes once they actually work!
The work discussed in this post shipped in the OS last year (fall 2025), so nothing here is dependent on very recent changes.
Average swift experience
During the State of Platform keynote, on the subject of Swift adoption across macOS, several examples were given, not only TrueType engine.
RIS is happening across all OS levels, if the keynote is to be believed.
They’ve been doing it for years. I don’t remember how we first knew, but I know they’ve been using Swift in kernels for at least some of the other chips like the Secure Enclave or whatever.
I’m not sure exactly which. I assume it’s some of the code and not all. But it’s not new in the abstract.
That said I don’t think I’ve heard of it in the kernel of MacOS on the main processor. That may be new.
Either way this is certainly the most concrete announcement I remember them ever giving on this stuff.
I know internally they use an IPsec implementation written by Rust (I think in the iCloud infra). Heard this from an ex-Apple engineer Ben (forgot his last name) that did a wonderful presentation of Rust from first principles. He said that it was hard to get people in on Rust when most would argue for Swift.
Edit: This is the guy: https://rustcurious.com/course/
Some stuff was discussed at Meet with Apple security event a few months ago, and the talks on FoundationDB rewrite, or why Swift Embedded subset came to be.
However I miss them actually having had one of those 15 - 30m WWDC sessions, where they could have gone a bit deeper into the keynote examples
Curious the direction of Webkit as there was a nebulous mention of select portions being rewritten from C++ to Swift. And yet, the new ECMAScript module (ESM) loader for Safari 27 is implemented in C++ (https://webkit.org/blog/17967/news-from-wwdc26-webkit-in-saf...).
No idea, maybe the private parts of the code, Safari isn't open source, or is coming later.
In any case I would have liked to have more info during the deep dive sessions.
As it is, Meet with Apple on security (a 5h long event) had much more information.
What does RIS stand for?
Rewrite in Swift
So RIS is Apple’s version of RiiR, but with better fonts.
s/better/blurrier/
Life without retina displays is possible, but pointless.
Careful, you’ll bring out all the “but it’s true to print media” people.
"we used a fuzzer to minimize a corpus of 10 million PDF files down to 4,200 without any loss of code coverage"
Did they need a fuzzer for that? They could've render them all and see what's exercised?
Its a set cover problem, and is NP-hard. 4100 of something probably runs nicely in your laptop in a reasonable amount of time.
The author discussed this a bit on Mastodon as well:
https://xoxo.zone/@numist/116716469017975106
I'm also here :)
Excellent write-up!
Thanks!
Interesting that this is published under the MIT, rather than Apple’s more favorite Apache 2, license
Why is it interesting?
Presumably because MIT is even more permissive and it’s a change in Apple’s behavior.
Some corporations prefer Apache 2.0 for projects where they'll be accepting contributions, because it includes patent protection and retaliation clauses. In case like this, where source code is just being published for reference and contributions aren't accepted, those risks don't exist.
Given the age of TrueType, wouldn't nearly all patents be expired already?
Apache2's license I've heard described as mutually-assured-patent-destruction - if you use the code and make a patent claim, your rights to use the code go away.
So Apache2 offers little benefit here, and MIT may get it into more hands?
Back in 2023 there was talks about Microsoft rewriting the font stuff in Rust for similar reasons Apple is now doing the Swift move.
I'm not sure what became of it and if it ever shipped. If anyone knows I'd be curious.
Russinovitch (Azure's CTO/CISO) gave a speech at RustConf 2025 and mentions it(DirectWriteCore) took 2 engineers 6 months resulting in 154K LOC and 5-15 percent performance increase for font shaping. https://www.youtube.com/watch?v=uDtMuS7BExE&list=PL2b0df3jKK...
So, hinting only takes place at low resolutions, I believe. How often is it used, eg viewing “typical” PDFs on “typical” screens?
As much as I enjoyed Swift, one can only wonder what the world would look like if they had gone with Rust as their default language instead.
Rust doesn't have an ABI [1]. Swift needed one to be a useable application language:
https://faultlore.com/blah/swift-abi/ (written by a core Rust developer)
[1] apart from the basic/universal C one, which prevents exposing any useful Rust semantics over the interface
One of the genius things about Swift is its interop with Objective C. Made the switch over considerably easier for developers. I’m not sure what that looks like in a Rust world.
Rust is also just a more complex language. I’m not convinced the benefits would have been worth it.
Not just interoperability with Objective C but with C (full) and C++ (increasingly better but not full) as well.
Swift is also interoperable with different versions of itself courtesy of the Swift stable ABI (Application Binary Interface)[0], which they invested a significant amount of time into at the expense of adding other new features to the language, which have come along later.
Rust offers a different approach: recompile everything and static linking.
[0] https://faultlore.com/blah/swift-abi/
C compatibility comes via Objective-C, because contrary to C++, Objective-C extends C, instead of being based on a C subset.
You missed Java as well.
Modern Swift borrows a lot from Rust! And it also has its own benefits, both ergonomic and also supporting eg generic in dynamic libraries
These days I mainly write Rust but I did write a semi complex iOS app and enjoyed Swift. I just didn't love how slow the type checker was and how it got lost. I recall having to break things into smaller bits to help the compiler, and there were some oddities about the language.
The gap between the two languages is quite small, it just makes me wish Apple was also all-in on Rust
In the last year they’ve added improvements to the type checker to speed it up, those would have been released now.
They have further and much more significant changes that I think might have recently landed in the development version. That should make an even bigger difference. But it’s not in a released version yet.
And yes, none of us like that one part of Swift. Especially the DRASTIC difference compared to objective-C which really only checked syntax and little else.
It’s still probably my favorite language right now though I don’t get to write in it much.
maybe so on the surface, but it remains quite massive underneath; these languages are fundamentally different and target entirely different use cases
I'm not sure Rust has one specific use case as its main goal, despite being immediately suitable for systems programming.
I use it for making user-facing desktop applications, to name one example.
I see Swift as a more approachable version of Rust.
If somebody is mulling over Rust but finds it too difficult to grasp, they could start off with Swift first and then move over to Rust.
One of the main advantages of Rust is a more developed and thriving ecosystem.
Swift and Rust were developed at similar times. I think of them more as having similar influences than borrowing from each other.
There’s no reason to invent your own head canon, the influence was openly acknowledged when Swift was new and it continues now that the language is developed out in the open (see Swift Ownership Manifesto)
Obviously Rust was first but over time both languages have been taking inspiration from each other. For example let-else was motivated directly because of its success in Swift: https://rust-lang.github.io/rfcs/3137-let-else.html#prior-ar...
Additionally both have influences from CLU, C++, Object Pascal, Modula-2, Mesa/Cedar, Standard ML, Cyclone.
Many features that get discussed as being Swift/Rust, trace back to one of those languages.
Similar times and the Rust originator went on to work on Swift after it.
Graydon Hoare's impact on the language is marginal than that of Chris Lattner, the originator (also, Hoare joined the team much later)
Does it borrow borrow checker?
I believe Swift tends to use reference counting and copy-on-write strategies. This, like GC, is less for the programmer to think about and doesn't require the semantic checks, but sometimes the performance cost is unacceptable compared to what you'd write in Rust.
You can choose to use either refcounting or unique ownership for your types. For most use cases, refcounted (+ copy-on-write) is the best choice and is the default, but the truetype interpreter made extensive use of non-refcounted types to achieve this performance.
You can pick and chose, and memory ownership is getting better in latest versions.
Being more ergonomic is relevant enough for increasing language adoption, that possible improvements are now on Rust roadmap.
They have either recently added or talked about a borrow style system in the language as a way to avoid more copies and speed things up/lower memory usage/help with asynchronous programming.
Yes, it has a borrow checker.
Swift takes the right approach in language ergonomics, many of its use cases would be much harder with explicit .clone() all over the place, lack of back references, no standard ABI for binary libraries, interop with Objective-C and C++, no standard concurrency runtime, or error handling types.
Rust 2026 roadmap has language ergonomics on it for a reason.
That said, outside Apple ecosystem you probably better with Rust, or if one has no GC issues, OCaml, Haskell, F#, Scala, C#.
Apple has a runtime system that looks a lot like smalltalk. Everything at that layer is dynamic. They needed a language that could seamlessly interact with that system.
Welcome to the club of doing high performance text in a memory safe language!
There are dozens of us! Dozens!
Vello has been a big inspiration and source of knowledge for my own webgpu text renderer, thank you for that!
> high performance text
Just strings or rendering strings?
If the latter, who are the other members of the club?
What's funny is from 2023 (I think), macOS just draws the UI unhinted. You have a 1080p display and you don't want to see the letters in the UI blurred to death? Tough luck, 1080p is incompatible with macOS, everybody needs "retina", and nobody cares that Windows and all Linux DEs look on 1080p just fine.
It looks like this hinter will be used only in rendering PDFs, because that's where they test the performance.
While hinting is disabled for most fonts, there are some fonts that require hinting to render correctly. We have to support hinting for those fonts, and it was easier to make it secure by rewriting hinting in Swift than it would have been to comprehensively identify every font created by those foundries.
My last 1080p monitor was around 20 years ago. I have trouble comprehending people still use them regularly.
A very quick search yielded Dell selling 1080p laptops today:
https://www.dell.com/en-us/shop/dell-laptops/dell-15-laptop/...
It is very, very common. Just not in the Mac world.
It's also the corporate standard for generic cubicle workstation monitors, though it's unusual to find a Mac in such a place anyway.
It’s very common for the people who got their last monitor 8 years ago.
It's still the most common resolution for people using desktop monitors today, according to: https://gs.statcounter.com/screen-resolution-stats/desktop/w...
Also for gaming too: https://store.steampowered.com/hwsurvey/
1920 x 1080 51.89%
2560 x 1440 21.20%
3840 x 2160 5.00%
People also use usable mice instead of touchpads, and they put the "ctrl" key where Apple thinks a useless "fn" should be. All kinds of things happen outside Apple world.
To me, it's more about what I'm used to. I have a perfectly fine several years-old monitor, so why should I throw it away?
Then let me blow your mind:) One of my daily drivers is an ex-Chromebook at 1366x768. Granted, it's also physically smallish so the DPI isn't quite as low as a macbook would be with those pixels, but still. And that's a touch cramped but it's fine.
I'm reading on that resolution right now! MacBook Air 11" running Linux that I use as a quick hacking/reading machine in bed.
The problem is, as soon as you are not on a Mac but Linux or Windows, you are in for an awful, truly awful lot of pain. HiDPI support is a mess because even in the rare case applications are made with HiDPI in mind they are not tested on HiDPI machines.
Other way around, most Mac software is not tested how it behaves on inferior external monitors.
What kind of windows programs are these? HiDPI is more than a decade old. A desktop application, no matter what OS it is, should always be tested with different scaling factors.
Ham radio software, both open-source and commercial, is a big thing, and so is many an in-house development in many businesses.
macOS has been drawing unhinted text for an eternity, and for those who can tolerate it on low-DPI screens, it's a great thing: the letter shapes look the same at all sizes, and the spacing between letters is consistent at all sizes.
I'm a high DPI snob so I haven't used a low res monitor for work in forever, but isn't the entire point of font hinting to make the text more legible at smaller pixel grid sizes? Yes, of course the shapes are more consistent since the hinter isn't touching them, but isn't the end result just less legible text?
Hinting purposefully destroys letter shapes in exchange for crispness. People who like macOS style font rendering prioritize letter shapes faithful to the font designer more than crispness.
Whether good letter shapes is more legible or crisper text is more legible is basically subjective. In the 2000s before HiDPI became popular different people really thought one was more legible than the other and vice versa. HiDPI made this basically moot.
I had this problem on the first Apple Silicon Mac Mini in 2020 so it's at least a little older than 2023.
I'm surprised the code has visible LLM smells. Though, I shouldn't be surprised. I hope the important bits are still human-controlled (and the same for Apple's many operating systems that absolutely deserve to remain stable and understood).
I assure you, every inch of the interpreter code has been stared at by humans, a lot. TBH even the assembly generated by it has.
From what I got Apple is using claude code A LOT internally
It would be interesting to see their internal guidance on LLM use. It’s a massive amount of new power that has to be wielded carefully. That kind of guidance might mean the survival or downfall of some big corps in the next few years.
Yes they are using Claude Code - not the Xcode agents.
It worries me. I hope Codex adoption picks up there.
thats a shame if true, they really should be dog-fooding that horrible agent ui in xcode to bring it up to a usable state
A lot of devs are tricked into using it because it’s official
I love how it doesn't even support reasoning output or edit tool diffs. (Yes it can show diffs sometimes when using the 100% official flow but not when using ACP)
I think these are the types of things Apple should've focused on instead of half-heartedly barging ahead with SwiftUI and breaking the language in the process
I mean they’re doing both
No mention of AI? Hand written code?
There's mention at the end. The models (and Swift itself!) have evolved a lot since this project started, so the early code is largely hand-rolled and the later changes were mostly authored by centaurs (to steal a term from chess).
But I personally reviewed every line that shipped and was absolutely insufferable about testing.