iOS: Fixing a Crash on Launch Issue on Apple Silicon Macs
My iOS app Mango Baby is made available for Apple Silicon Macs. Ever since macOS Sonoma, it would crash on launch.
A similar crash on launch issue happened for the release of macOS Ventura. It was because the ActivityKit is not available on macOS. I fixed it by protecting calls to the ActivityKit APIs against isiOSAppOnMac at runtime. But apparently a new issue has appeared for macOS Sonoma.
Attempt 1
First I asked on Mastodon, and Matt Massicotte very kindly provided tips and walked me through the debugging process.
TIP: For crashes on launch, Apple-captured crash logs are much more helpful than Crashlytics. They can be accessed in Xcode following these instructions.
Through the crash logs, I discovered the following stacktrace:
Termination Reason: Namespace DYLD, Code 1 Library missing
Library not loaded: /System/Library/Frameworks/ActivityKit.framework/ActivityKit
Referenced from: <9B5A3687-FF24-3E38-9007-1DE6A69EFEEB> /Volumes/VOLUME/*/Baby.app/PlugIns/BabyWidget.appex/BabyWidget
Reason: tried: '/System/105Support/System/Library/Frameworks/ActivityKit.framework/ActivityKit'
(terminated at launch; ignore backtrace)
Thread Crashed:
0 dyld 0x0000000189391b48 __abort_with_payload + 8
1 dyld 0x000000018939e108 abort_with_payload_wrapper_internal + 104 (terminate_with_reason.c:102)
2 dyld 0x000000018939e13c abort_with_payload + 16 (terminate_with_reason.c:124)
3 dyld 0x0000000189325518 dyld4::halt(char const, dyld4::StructuredError const*) + 304 (DyldProcessConfig.cpp: 2890)
4 dyld 0x00000001893221e8 dyld4::prepare(dyld4::APIs&, dyld3::Mach0Analyzer const) + 3884 (dyldMain.cpp:0)
5 dyld 0x0000000189320f44 start + 1948 (dyldMain.cpp:1238)
Then I remembered that I had to weak link the ActivityKit framework in order to make Mango Baby continue available on Macs. This is done by adding -weak_framework ActivityKit to Other Linker Flags. But this stacktrace is from my BabyWidget app extension for widgets, and I didn't have the flag for my Widget target. I submitted an update along with other features.
Since I didn't know any way of testing iOS apps running on Macs, I didn't verify the fix. And it turns out Mango Baby is still crashing.
Attempt 2
Later I learned from Franklin on Threads that you can enable the Test iPhone and iPad Apps on Apple Silicon Macs option on App Store Connect. Then, you can at least use TestFlight to test your iOS app running on Mac. This still isn't ideal, but it's already 100 times better than nothing.
I checked new crash logs again, and they look like:
Exception Type: EXC_BAD_ACCESS (SIGSEGV)
Exception Subtype: KERN_INVALID_ADDRESS at 0x0000000000000000
Exception Codes: 0x0000000000000001, 0x0000000000000000
Thread 0 Crashed:: Dispatch queue: com.apple.main-thread
0 ??? 0x0 ???
1 Baby 0x102c75b1c closure #1 in closure #1 in closure #1 in closure #2 in AppModel.init(storage:) + 32 (AppModel.swift:488) [inlined]
2 Baby 0x102c75b1c partial apply for closure #1 in closure #1 in closure #1 in closure #2 in AppModel.init(storage:) + 60
3 SwiftUI 0x1c96f4c38 0x1c7d86000 + 26668088
4 SwiftUI 0x1c96f4f28 0x1c7d86000 + 26668840
5 libswiftCore.dylib 0x190f6025c withExtendedLifetime<A, B>(_:_:) + 28
6 SwiftUI 0x1c96f3420 0x1c7d86000 + 26661920
7 SwiftUI 0x1c96f2f74 0x1c7d86000 + 26660724
Thread 0 crashed with ARM Thread State (64-bit):
x0: 0x0000000000000000 x1: 0x0000000000000000 x2: 0x0000000000000103 x3: 0x0000600000296400
x4: 0x0000000000000000 x5: 0x0000000000002760 x6: 0x00000001d94074e8 x7: 0x00000000ffffffff
x8: 0x000000016d4658f0 x9: 0x000000016d465910 x10: 0x0000000000002760 x11: 0x00000000982478cd
x12: 0x00000000000007fb x13: 0x00000000000007fd x14: 0x00000000984480d3 x15: 0x00000000000000d3
x16: 0x0000000000000000 x17: 0x00000001d98d6c30 x18: 0x0000000000000000 x19: 0x000000016d4659d0
x20: 0x000000016d465950 x21: 0x0000600002336710 x22: 0x00006000039a3020 x23: 0x0000000000000000
x24: 0x000000016d465930 x25: 0x000000010372ebe8 x26: 0x00000001d98d3ec8 x27: 0x000000000000000f
x28: 0x0000000000000000 fp: 0x000000016d465d90 lr: 0x0000000102c4ca70
sp: 0x000000016d4658f0 pc: 0x0000000000000000 cpsr: 0x20001000
far: 0x0000000000000000 esr: 0x82000006 (Instruction Abort) Translation fault
Binary Images:
0x102998000 - 0x1035b3fff com.mangoumbrella.Baby (2023.12) <ca277200-885e-3b08-8ee2-93648cc69108> /private/var/folders/*/Baby.app/Baby
Following this helpful Investigating memory access crashes article from Apple, I ran the following command to discover the exact line it's crahsing:
$ atos -arch arm64 -o atos -arch arm64 -o /Applications/Mango\ Baby.app/Wrapper/Baby.app/Baby -l 0x102998000 0x0000000102c4ca70
closure #2 in AppModel.processRemoteRecordEvents(events:) (in Baby) (AppModel.swift:1009)
And the line is:
if !ProcessInfo.processInfo.isiOSAppOnMac {
LiveActivitiesManager.shared.syncAll() // Crash here.
}
This doesn't make sense to me since this line shouldn't execute at all. It's protected by the isiOSAppOnMac check, and all the ActivityKit API calls are isolated in my LiveActivitiesManager class.
My only guess right now is that this might have something to do with branch predictions and the runtime will load the instructions even if they aren't executed at runtime in the end. With this guess, I have one solution in my mind: introduce a Protocol and two concrete classes for all my ActivityKit related functions. Something like:
protocol LiveActivitiesManagerProtocol {
func syncAll()
}
class LiveActivitiesManager {
static let shared: LiveActivitiesManagerProtocol = {
if ProcessInfo.processInfo.isiOSAppOnMac {
return NoopLiveActivitiesManager()
} else {
return RealLiveActivitiesManager()
}
}()
}
fileprivate class NoopLiveActivitiesManager: LiveActivitiesManagerProtocol {
func syncAll() {}
}
fileprivate class RealLiveActivitiesManager: LiveActivitiesManagerProtocol {
func syncAll() {
// Use ActivityKit APIs.
}
}
This code is now released in Mango Baby v2023.14 and it no longer crashes on Sonoma.