In actuality, most of the crash reports I have like this do not have any threads with active sqlite work. That's why I didn't just assume that these were 0xdead10cc terminations (along with the fact that MetricKit does include the 0xdead10cc code in other crash diagnostics). But for now I'll proceed under that assumption and see how far I get.

FB24609180 I was beginning to suspect this was the issue. I've also discovered that my app builds UI too early in the startup process, which triggers some sqlite work with transactions. Also, for at least one recent app version, I was seeing a few devices each day that would get into a loop of being background-launched, getting terminated like the above, and getting relaunched every 3-5 seconds for several hours. I have no solid theory on what would keep relaunching it.

This is a good overview of the Organizer and related data but doesn't answer any of my questions.

FB23072845

For collecting a backtrace in the field, is the best way to use NSThread.callStackSymbols? I have two sets of diagnostics saved from customers hitting this. In one, I get two viewDidLoad calls a little under a second apart. In the other, I get three - the second one is 22 seconds after the first, and the third is 5 seconds after that. In the first, I get viewWillDisappear but not viewDidDisappear. In the second, I get no disappear methods. Would a DTS ticket be useful for more detail?

I can sort of reproduce it locally, but I want to know what's most common for my users in the field. And my app interacts with various SQLite files through tons of different code paths.

My design is for using the hard-blur style to fill something like a title-bar at the top of a vertically scrolling area, like in the screenshot in the linked forum post. If the scroll edge effect isn't the best tool, what is a better choice?

I should add that I've never reproduced the issue in debug - I've only gotten diagnostics from users, and it's quite a rare issue.

The view controller in question is contained in a UINavigationController that's a child view controller of something else, but the issue occurs when the view controller is created and pushed onto the navigation stack when that UINavigationController is already visible. So I don't think it's an issue of mismatched beginAppearanceTransition / endAppearanceTransition, but I'll audit my usage of those. I'm not resetting the view to nil anywhere.

The original report is closed, so I can't add more information to it.

Thanks. It doesn't look like this landed in iOS 26.1 beta 4, but I'll check again on the next beta.

Same. FB20069803, with a sample project.

Thanks for the confirmation on the length limit. Can you provide an example of (or clarify further) how to split content into smaller chunks for the LanguageModelSession API? I want the model to be able to do something with the whole input at once. It seems like a notable limitation if I can't have the model act on any content longer than 4096 tokens in total...

The issue still exists on macOS Sequoia 15.3 Beta 1 (24D5034f).

I have reproduced the issue on 15.1 (24B83). I selected Seed 6 in the "what build" dropdown in Feedback Assistant because that was the latest version available in the dropdown, though apparently I neglected to point that out in the issue description (though I have said it in other issues - see FB15730854, FB15714657). Regardless - you can't reproduce this issue on 15.1 using the sample app I provided? Could there be other factors at play here? Is there more data I can provide to clarify?