Natalia Panferova:

I thought it would be useful to have a single reference for the different ways we can style the toolbar in SwiftUI on macOS 26.

Phil Zakharchenko:

The toolbar id creates a stable customization domain, and each item id gives the framework something stable to persist when the user adds, removes, or reorders controls. This identifiability of toolbar items is a requirement for customization. Under the hood, when SwiftUI inevitably bridges your toolbar contents to NSToolbar, these identifiers will be directly translated to the NSToolbarItem’s itemIdentifiers.

As we continue the migration to a customizable toolbar, we are likely to encounter a few surprises related to this model.

[…]

After a closer inspection, and perhaps a few minutes of digging through the documentation, we come to the bleak realization: there is no customizable ToolbarItemGroup API.

[…]

The regular toolbar model and the customizable toolbar model overlap, but they are not interchangeable. You cannot give a ToolbarItemGroup an id, and you cannot place one directly inside toolbar(id:content:).

Majid Jabrayilov:

Let’s take a look at another example. This one is a bit tricky, because on macOS, you will get Action 1 in the top trailing position and the whole group of actions centered in the navigation bar. On iOS, you will get Action 1 pinned to trailing and all other items collapsed.

[…]

The final addition to the Toolbar API is the toolbarMinimizeBehavior view modifier allowing us to minimize toolbars while scrolling up or down. You can minimize tab, bottom, window and navigation bars with it.

With new modifiers SwiftUI gives us a much better balance between system-driven adaptation and explicit control. We can now decide which actions are more important, which items should always stay visible, which groups should intentionally live in the overflow menu, and how toolbars should react to scrolling.

Natalia Panferova:

iOS 27 includes a few smaller additions that make toolbar content easier to configure.

We can remove the default padding around an item whose content should extend to its edges with contentMarginsRemoved(_:)[…]

[…]

The status bar is now represented by ToolbarPlacement.statusBar, so its visibility can be controlled through the same toolbar API as other bars[…]

[…]

ForEach now conforms to ToolbarContent, which means we can generate toolbar items directly from a collection[…]

Natascha Fadeeva:

Together, the toolbar APIs let us express the importance of each action while SwiftUI continues to handle the toolbar’s layout across different sizes.

Previously:

• NSToolbar in Golden Gate
• SwiftUI in appleOS 27
• Golden Gate Window Corners
• Golden Gate Sidebars and Toolbars
• Where Did SwiftUI Leave You Hanging?
• Liquid Glass: Content vs. Controls

Amy Worrall (Mastodon):

McKinley is an all-in-one studio for drawing and annotating SF Symbols. It bundles vector tools with knowledge of the SF Symbols structure, so you can edit one set of objects across multiple weights, sizes and colours. Import SVG or start from scratch, live preview your work using Apple’s own renderer, and export directly to Xcode, so you never need touch the SF Symbols app again!

[…]

Gone are the days of converting your strokes and shapes to outlined paths before export. Keep everything how you drew it, and let McKinley handle the output.

[…]

Link your project’s asset catalog to your McKinley document. Then when you make a change, hit the Sync keyboard shortcut and it’s immediately populated with your latest version.

[…]

McKinley is a Mac-assed Mac app: a native AppKit UI, Quick Look previews and thumbnails for your documents, a full AppleScript dictionary for batch work, and the whole user guide offline in Help.

$60, no subscription. It’s been a while since I had to deal with custom SF Symbols, but I remember it being a major pain. I haven’t had the chance to try McKinley with a real project yet, but it looks very well done. I love seeing a new app ship with documentation and extensive AppleScript support, including sample scripts that you can install right from its script menu.

Mario Guzmán:

The icon isn’t just a symbol but a whole artistic creation on a canvas.

Look at how your tools are on panels that could be individually opened on their own auxiliary window. I can’t stand the single-window, everything buried into a single-sidebar design we get these days.

Amy Worrall:

SF Symbols rendering does a surprising amount behind your back: it interpolates nine weights from three masters, applies optical size adjustments, resolves hierarchical opacity ladders, and changes some of this between OS releases. Any preview I rendered myself might diverge from the real thing in ways nobody would notice until the symbol was in an app looking subtly wrong. (Obviously I aim to get my canvas rendering with the exact same nuance as Apple’s renderer, but programmers are fallible and the fewer ways things can diverge, the better.)

Apple provides no API for “here’s my SVG, render it as a symbol”. For a normal app using a custom SF Symbol, the symbol is compiled from an Asset Catalog in Xcode into a .car file (a compiled asset catalog) at build time. But there’s no ‘build time’ when I want to update a preview in McKinley.

My workaround involves compiling your symbol into a .car file at runtime, on every edit, by patching bytes into a template, then loading that temporary .car file into McKinley via NSBundle.

Previously:

• Octavo 1.0
• SF Viewer Removed From the App Store
• SF Symbols

Update (2026-09-08): John Gruber:

I mean just look at McKinley. It’s both beautiful and quite obviously a serious professional tool.

Update (2026-09-11): Amy Worrall:

Anyone used any great implementations of dockable palettes? (Adobe’s don’t count, I don’t think they’re great!)

I’m not going to go back to sidebars, but some kind of options/flexibility might work, if it doesn’t get in the way!

Adam Engst explains why AI software development should be approached as an evolution, starting with safe, verifiable data analysis, advancing to simple scripts, and eventually building applications. He distinguishes programming from development, arguing that non-programmers can act as product managers who direct AI, demand security checks, define requirements, and create software tailored to problems existing apps do not solve. (3) 

💾

💾

Howard Oakley:

As those new models start shipping on 22 September, the window for the first public release of Golden Gate has narrowed to 15-22 September, with greatest likelihood on the same day as the arrival of those new Macs.

[…]

At the same time as the release of Golden Gate, we should get Tahoe 26.7 and Sequoia 15.8, but it looks increasingly likely that the final version of Sonoma has been 14.8.9, released on 6 August.

Howard Oakley:

As we prepare to flee from the visual horrors of macOS Tahoe at last, this article is a reminder that it wasn’t the first major release of macOS that has had blatant bugs in its interface, although I suspect it may be unique in that those were intentional. While progress in Golden Gate has been limited, I hope you’ll find it rectifies some of the worst errors of Liquid Glass.

Many Macs will be left behind, as Golden Gate drops support for Intel.

Normally, I update around January. With Tahoe I waited until July. This time, I’m considering updating to Golden Gate soon after its release. The downside is that Xcode 26 doesn’t run on Golden Gate, so I would have to drop support for many older macOS versions (about 3% of my customers’ Macs). But it seems like Apple’s going to force that soon, anyway, as one of the next Xcode 27 updates will likely drop support for Tahoe. I want to get off Xcode 26, both to use the newer SDK and because there are various debugging/testing things that don’t work properly.

Mario Guzmán:

I am working on a new app. Right now, I am targeting back to macOS Sequoia. I think that would be a good OS to target but becoming quite old with Golden Gate just around the corner. […] I wonder if I should just target macOS Golden Gate and focus testing on that.

Previously:

• Ending the Golden Era for Backward-Compatible Apps
• Mac Studio 2026
• Mac mini 2026
• macOS 27 Public Beta
• Glow Leopard
• Avoiding Tahoe
• Ending macOS Intel Support
• Waiting to Update to Big Sur
• Waiting to Update to Catalina

Update (2026-09-04): Isaiah Carew:

i’m in the exact same boat. i’d really like to stay on macOS 15 and whatever Xcode works there. it means i can still build things in the old UI and mostly Tahoe and GG respect it.

but recent bug reports from beta testers and the inability to see these bugs on my VM has made me feel very stuck here — like apple has purposefully made it difficult to move forward and support older versions too.

Update (2026-09-11): Howard Oakley:

What if any preparations do you need to make before upgrading or updating then, or in the coming weeks?

Update (2026-09-16): Adam C. Engst:

So who should upgrade immediately, and who should wait? Let’s consider a few broad categories of users.

Our conversation with Adam Engst continues with an exploration of  how AI can help non-programmers build sophisticated software by following professional development practices. Adam explains using specifications, architecture documents, and AI-guided decisions to create reliable apps, then examines the approach his app, SetShot, takes to macOS settings, privacy, security, user feedback, Cloudflare, GitHub, and minimizing sensitive data collection. 

💾

💾

Craig Hockenberry:

Then, when you look at the sample code, it’s all SwiftUI. At least the sample compiles and runs because it uses a completely different syntax than what was shown at WWDC.

[…]

Eventually I stumbled upon TipUIPopoverViewController which inherits from UIViewController. That should work! […] But it didn’t: the close button on the popover didn’t work.

[…]

Until you realize that any changes you make are stored in a SQLite database, debugging is very confusing. It’s also easy for your own state to get out-of-sync with the Tip state: there is not a single source of truth.

[…]

Unfortunately, that sample code doesn’t scale well. If you have a view with multiple tips, you’re going to be littering your code with tip instances, observation tasks, and popover controllers. It’s a mess and a clear sign that the TipKit developers didn’t think much about UIKit.

Fatbobman:

In the previous article, we introduced the basic usage of TipKit. In this article, we will discuss some advanced topics related to TipKit, such as how to fully customize Tip views (not using TipView and popoverTip), how to use TipKit with UIKit, and how TipKit can share data across different applications, as well as how to reuse Tip declaration code. Additionally, we will attempt to address some common questions and concerns about TipKit.

Previously:

• SwiftUI in appleOS 27
• SwiftUI at WWDC 2025

John Sundell:

New in Swift 6.4 […] is a warning which is emitted when a nested closure performs a weak capture while the parent closure implicitly captures that object as a strong reference.

At first glance, that new diagnostic could be a bit puzzling, since it might seem unnecessary to have to specify such a weak capture twice (or, alternatively, move it to the parent closure’s capture list), but it turns out that it’s an incredibly useful new warning that can help us avoid a quite common kind of memory management bug.

[…]

Finally, in situations where capturing self strongly doesn’t actually lead to a retain cycle, we can silence the new warning for nested weak captures by explicitly adding self to the outer closure’s capture list, such as when creating a Task within the following example[…]

Previously:

• Swift 6.4
• Xcode 27 Announced
• How to Automate Memory Leak Detection With XCTest
• Avoiding Implicit Retain Cycles When Using Swift Function References
• Weak Self: Closure Rules of Thumb
• Swift “guard” Capture Specifier Pitch
• Implicit Capturing of Self in Swift 5.3
• Dealing With Weak in Closure-based Delegation
• Stop the weak-strong Swift Dance

Matthias Gansrigler-Hrad:

I don’t appreciate being forced by Xcode 27 to drop support for macOS 10.13-11.0. It’s my decision which OSes I want to support, not Apple’s. Yes, I can keep Xcode 26 around until submissions require Xcode 27, but what then?

Originally, Xcode 27 was documented as supporting Big Sur on Intel Macs, but now the earliest is Monterey.

I’m frustrated, too, but it’s more honest than the recent releases that claimed to support deployment on older macOS versions when Apple clearly hadn’t tested that it actually worked. I filed bugs, and Apple kept responding that they’d fixed it, when they hadn’t.

Using old versions of Xcode isn’t a great solution because they aren’t supported on Golden Gate. Swift also makes it more difficult to call newer APIs when using an older SDK. You can’t just copy/paste a method prototype or constant value and conditionally use it when running on a newer OS.

Ighor July:

I eventually found a hidden workaround that changes the errors back into warnings. That is good news, but it does not make the concern disappear. The setting is undocumented, absent from the normal Xcode interface, and currently being tested only in a beta toolchain. The larger question is what Apple intends to enforce when Xcode 27 is final and, later, when App Store submission requirements move to its SDKs.

[…]

Apple’s current Xcode system requirements list Xcode 26 with deployment targets beginning at iOS 15 and macOS 11. Yet developers have been able to enter older targets manually. In an Apple Developer Forums discussion about Xcode 26, an Apple product manager explained that the published deployment target range is aligned with simulator support and on-device debugging. The accepted answer confirms that an iOS 13 target can still build even though it falls below the range in Apple’s table.

My own tests went further. Xcode 26.5 compiled a project targeting iOS 9.0. It warned that the target was older than the supported range, but the build succeeded. On macOS, I have compiled current apps with deployment targets as old as macOS 10.9. Xcode warned me, but it still produced working applications.

[…]

For iOS and iPadOS, staying on Xcode 26 is not a lasting option. Apple has repeatedly made a current Xcode and SDK mandatory for new App Store submissions. […] If any one of those conditions fails, iOS and iPadOS developers will have no normal App Store path for publishing new updates that still support versions older than 15.

Previously:

• Xcode 27 Announced
• Xcode 15.4

Update (2026-09-07): Léo Natan:

Just want to remind you guys that if you move to Xcode 27, the only “officially” supported and sanction Xcode for GG, UIDesignRequiresCompatibility is no longer respected across the 27 OSes (apps built with older Xcodes still are). So you might suddenly find out you have a lot of work ahead of you, adding support for the dreaded Glass. And 27 Glass is not the same as 26 Glass. That can be an issue if you skipped 26 altogether but still want to support it.

Previously:

• Preparing to Update to Golden Gate

Update (2026-09-11): Nick Lockwood:

I mostly ignored the Xcode 27 betas (mainly due to the barely-usable Device Hub) so I didn’t realize the passive-aggressive prompts to drop support for iOS < 15 that you get when submitting apps is being backed up by Xcode 27 refusing to compile apps at all unless you bump the deployment target 😧

Adam Engst looks back at HyperCard as an early tool that democratized software creation, then connects its legacy to today’s AI-assisted development. He outlines how Claude helped turn an idea for tracking macOS setting changes into a working Mac app, SetShot. Adam discusses Xcode, GitHub, rapid iteration, and the value of visual feedback in his development process, and how AI lowers programming barriers without eliminating the need for thought, judgment, and effort. 

💾

💾

Dave Nanian:

Recent versions of xOS (that is, macOS, iOS, iPadOS, etc) have Data Protection Classes from A to D that get assigned to files and folders.

[…]

The thing is, though, is that these files and folders are inaccessible unless the user is logged in, and the screen is unlocked. A few seconds after the screen locks (or you log out), the keys are purged, and they cannot be read.

[…]

That means the files any application considers to be worth protection cannot be accessed when a backup usually happens.

[…]

You may have noticed that if you had an overnight backup, when you unlock your screen, a copy starts to run. It’s typically quite brief, thanks to ⚡️Turbo Smart Update, but it might surprise you anyway. […] During your overnight copy, SuperDuper kept track of the files that it couldn’t access due to their protection class. And now that it can access them, it’s making sure they get backed up.

This seems like the right thing to do. Time Machine also does it. A backup app should try to make as faithful a backup as possible. I kind of find it annoying, though. Sometimes it can’t do a Turbo Smart Update, and so the backup that should already be done may end up taking another 30 minutes or more. The best thing to do is schedule backups while you’re using your Mac to avoid SuperDuper needing to do such “follow-ups.”

I wrote some code to try to figure out which files I have that are using data protection. Interestingly, the URLFileProtection API seems to not work on macOS. I used the lower level getattrlist() instead. Nanian recommends fcntl(). I didn’t find any protected files that I actually care about, just caches and derived data from Apple’s own apps. But, in theory, Apple or a third-party app could be saving actual user data using data protection.

Previously:

• SuperDuper 4
• Launching Before UserDefaults Is Available
• Apple Addresses iOS Surveillance and Forensics Vulnerabilities

Vlad Fedorov (Hacker News):

On August 17, GitHub experienced an outage that lasted 7 hours and 47 minutes. It disrupted github.com, authentication, GitHub Actions, APIs, pull requests, issues, and Copilot, affecting developers and organizations around the world.

[…]

Neither outage was caused by a code or configuration change. Both incidents were capacity failures at their core. We failed to scale critical components before demand exceeded their capacity. Since April, monthly commits have grown from 1.4 billion to 2.9 billion. That growth explains the pressure on our systems, but it does not excuse these outages.

[…]

As part of the reliability commitments we made earlier this year, we have focused on three priorities: adding capacity, improving efficiency, and removing architectural bottlenecks. We have since added more than 3 million CPU cores, 120 petabytes of high-speed storage, and significant network capacity. We installed as much hardware as available power allowed in our existing data centers while accelerating our migration to Azure.

Alejandro Forero Cuervo:

This is the wrong way to think about this because there’s no such thing as infinite capacity. A large distributed system will be simultaneously mostly idle and (in some subcomponents) overloaded. The root cause is not “a component didn’t have enough capacity (because of auto scaling failures)”, but rather “this complex system collapses (rather than degrade gracefully) when demand exceeds capacity”.

When components reach capacity limits, the excess traffic of the lowest priority should be rejected. Rejected traffic should not be retried — in fact, not only should clients not retry these errors, these errors should cause client-side throttling. Traffic isolation should be applied — if the cause of the overload is a single client/customer system, no other system should be affected.

Nearly a decade ago I wrote about some of the techniques we applied at Google to implement these protections. Most other large internet services have since copied them, afaik.

Previously:

• Mac App Store Review Times Increasing

Marcin Wichary:

By default, on a Mac, the initial delay is 500ms, and the key repeat rate 1000ms.

[…]

My suggestion is to make them both as short as possible for testing.

[…]

First, it’s good to test whether your keystrokes behave well when repeated. Occasionally you might want your interface to suppress repeating, or do something special if the key is held longer.

Second, it’s a good test not just of repeating, but just the user pressing keys really fast.

Previously:

• Testing Tip: Always Show Scrollbars

Marco Arment:

Unforgetful isn’t a new productivity system to learn.

[…]

This is designed for a small number of short-term tasks. What do you need to do today, tomorrow, this week, or this month?

[…]

If you want tagging, folders, archives, or other ways to organize or file away tasks, this isn’t the app you’re looking for.

Unforgetful is for people who know that if they file away a task, it never gets done.

I’m an OmniFocus person. I love tagging, folders, and archives; I never want to see a notification for a to-do item. I prefer pull to push. So this is absolutely not the right tool for me. But I like the fresh thinking behind this app and how the various design choices interact to make it an opinionated tool for the right type of person.

Marco Arment:

Calendar alarms changed my life. Events without alarms don’t happen.

[…]

Most of my Reminders interaction has always been via Siri and notifications. And that fits me very well, with three major exceptions:

1. If I say “Remind me to buy milk,” but I forget to specify a time (or Siri misses it), Reminders will never notify me, and I’ll never check the app, so it’ll be lost.
2. If I inadvertently dismiss a Reminders notification without snoozing it, Reminders will never notify me about it again, so it’ll be lost.
3. When I do snooze a Reminders notification, which I do a lot, the snooze options suck.

He has good solutions to these and other problems. I’m tempted to try out Unforgetful just for the purported quick capture of dictated reminders, however there’s no watch app.

Listen to Accidental Tech Podcast to hear about unfortunate limitations of EventKit. Reminders itself is interesting to me in that the (private) data model is now so rich, but I’m put off by the UI, the lack of export, and the limited support for driving it from the keyboard.

Previously:

• TabControl Scam and the App Review Queue
• OmniFocus 4.8.10
• Dictating Literal Reminders to My Apple Watch
• CKSyncEngine
• Bulk Setting “Catch Up Automatically” in OmniFocus
• Overcast’s New Foundation
• Adventures in Siri Failures: Reminders Edition

Modular (tweet, beta, Hacker News):

Today, the Mojo language officially reaches 1.0: a milestone the language has been building toward since its first release in 2023. Mojo has grown into a general-purpose language with a vibrant developer community writing their own libraries, tools, and applications on top of it. With Mojo 1.0, developers can now build for the long-term on a stable, production-ready language foundation.

[…]

Importantly, Mojo 1.0 does not mark the end of the language’s evolution, but it is an important milestone on a longer journey. During the 1.x timeframe, changes should primarily be additive, giving developers confidence that the language will not continually shift beneath them. Breaking changes may still be made, but will be managed with care, following the standards of how mature languages (e.g. C++) evolve over time.

Modular (Hacker News):

We are happy to announce that the Mojo🔥 language is now fully open source under the Apache 2.0 license (with LLVM exceptions)! The source code for the Mojo compiler, tooling, and everything else you need to build the language are now available in our modular GitHub repository.

[…]

The Mojo standard library has been accepting contributions since 2024, and we’re grateful for everyone that has helped advance the language. One learning (particularly in today’s era of AI coding) is that we need to be deliberate about how we handle contributions. As such, we aren’t ready to take contributions to the compiler and tooling. We aim to accept contributions to the compiler and tooling by the end of this year, and we’ll share more details when we can.

Abdul Dakkak, Chris Lattner, and Jeff Niu (2023):

This talk will give an overview of Mojo 🔥, a new programming language in the Python family built on MLIR and LLVM. Mojo aims to bridge the programmability and performance gap in systems programming by combining Python’s ergonomics and ecosystem with C++ and CUDA-level performance into a single language. We will describe how Mojo leverages the MLIR and LLVM infrastructures to provide meta-programming, user-defined code transformations, hardware backends, adaptive compilation, and auto-tuning to enable developers to achieve performance without sacrificing programmability.

Gergely Orosz (2025, tweet):

Chris Lattner is one of the most influential engineers of the past two decades. He created the LLVM compiler infrastructure and the Swift programming language – and Swift opened iOS development to a broader group of engineers. With Mojo, he’s now aiming to do the same for AI, by lowering the barrier to programming AI applications.

I sat down with Chris in San Francisco, to talk language design, lessons on designing Swift and Mojo, and – of course! – compilers.

See also: LPython (Hacker News).

Previously:

• Qualcomm Acquiring Modular
• Mojo Is Open Source
• Mojo for Mac
• Mojo Programming Language

Nicolas Seriot (Hacker News, Reddit):

The core Unicode algorithms (normalization, casing, bidi, collation) are deliberately bounded, but UTS #35 transliteration rules, under their natural unbounded semantics, are not. This is a result I haven’t found published before.

These rules ship as locale data in ICU, the Unicode library shipped in operating systems, browsers, runtimes, and databases. Whether a given rule file terminates on a given input is undecidable.

[…]

To prove UTS #35 universality, we compile 2-tag systems (Post, 1943), a model proven universal (Cocke & Minsky, 1964), into transliteration rules.

[…]

Transliteration rules were designed to turn “é” into “e”. Three lines of them can compute the Collatz function.

KT_Music:

Inspired by this, I tried creating an implementation that “solves #Sudoku” using only UTS#35 transformation rules 😎

Previously:

• Outlandish SQLite Recursive Query Examples
• Solving the “Miracle Sudoku” in Prolog
• Sudoku Solver

Nicolas Seriot (Hacker News):

Engineering folklore holds that Jira (Atlassian’s project-tracking tool) is Turing-complete. Existing claims point vaguely at automation features without exhibiting a reduction. This article supplies a proof, with setup instructions and execution trace.

[…]

Minsky proved this model Turing-complete (1967). Exhibiting it in Jira’s automation language therefore establishes the reduction. Here is how the model maps onto Jira:

Minsky Machine	Jira
Register A	Count of linked issues of type Bug
Register B	Count of linked issues of type Task
Program Counter	Status of a single Epic issue
Dispatch Table	Jira Automation rules, one per instruction state
Clock	Automation-triggered transitions, or external re-triggering past chain caps

Alex Chan (Hacker News):

Many of these outages were caused by a single bug, deep in SQLite. It took months of intense forensics to track it down.

[…]

This lack of reliable trigger conditions meant we couldn’t reproduce the bug synthetically. Instead, we had to rely on deploying passive, forensic telemetry in our live environment to catch the corruption red-handed. Gathering live diagnostics for a database issue is the last thing we wanted to do, but we had no choice.

[…]

As an additional complication, the corruption didn’t occur on a regular schedule. Sometimes incidents would be hours apart, other times weeks. This made it difficult to predict progress or plan further work, because we were never sure when we’d get our next diagnostic dump. We had a six-week period between October and December when there were no corruption incidents, before they returned as an unwelcome Christmas present.

[…]

In two incidents, our transaction logs failed to replay cleanly. Upon closer inspection, we discovered that data written and committed by one transaction was inexplicably invisible to later transactions. A write had vanished into thin air without raising an error.

SQLite:

The bug is likely present in all version of SQLite from 3.7.0 (2010-07-21) through 3.51.2 (2026-01-09). It is fixed in version 3.51.3 (2026-03-13) and later. Backports of the fix are available for some earlier releases: 3.44.6 and 3.50.7.

The bug only affects databases in WAL mode when there are two or more database connections open on the same file, in separate threads or processes, and when those two connections attempt to write or checkpoint at the same instant.

Alex Chan:

This investigation is a useful reminder: running boring technology in a non-standard way is a risk. The common paths and standard configurations are incredibly well-tested and reliable. Most people use SQLite in a standard configuration and never face this sort of issue. Everything we were doing was a public, documented, supported configuration—but by taking manual control of the checkpointing process and running at our own aggressive pace, we stepped off the well-trodden operational path.

Scott Perry:

I spent the last two years of my time working on SQLite trying to chase this bug down. It was so rare that I never got an internal report; my assessment of the number of customers affected was “about a county’s worth”, some of whom were generous enough to allow their local Genius Bar to send me backups of the affected databases. I spent so many hours in a hex editor staring at the wreckage.

Top three data corruption bug of my career for sure, and the other two weren’t bugs in SQLite.

A former Tailscale employee got in touch and shared that they were checkpointing these databases every 250ms. Multiplied across a fleet of servers that is a fantastic way to surface a super rare bug—they basically built our test rack as a product, but two orders of magnitude more effective (partially due to scale, but mostly because our rack also panics devices at random in order to test the filesystem, NAND controller, etc. and rebooting takes a while)

[…]

13/10 fantastic bug. proof that 100% test coverage is a good start, and that formal methods are needed for critical applications.

Carl Sverre (Hacker News):

I whipped out my phone, and asked Claude to get to work. I had it get SQL 3.51.2 – still buggy – set up in Antithesis, and then instrument the code with a bunch of Antithesis assertions. You can see the instrumented version here.

Then I asked it to write a simple workload which exercised the WAL insert and checkpoint code. Notably, this is a completely generic workload. It just runs writes and checkpoints concurrently – things you’d expect to actually happen in production, all the time. The assertions are also generic to the bug, they’re all standard assertions you’d add to any database, things like “no lost committed writes” and “database is not corrupt” (called integrity check in sqlite).

On my first run, Antithesis caught the bug in 15 mins. Here’s the report.

Of course, it helps that he knew to look in the checkpoint code, but you could imagine using tools like this to quickly investigate hypotheses when you don’t already know the answer.

In a conversation at Macstock 2026, Adam Christianson reflects on retiring the MacCast, joining Mac Geek Gab, and starting the Debut Film Podcast with his daughter. He also explains how his views on AI have evolved, how ChatGPT and Claude support his programming work, why human oversight remains essential, and what AI may mean for […]

💾