Crash reports downloaded by Xcode contain impossible call hierarchy

As a getting started point, it would be helpful if you fed these reports to the xcrun crashlog /Path/To/Log.crash command, and see what it tells you. That command is a shortcut to a bunch of LLDB commands that pulls apart any optimizations that are present in a single frame of those call stacks. You may find that there are several layers of calls in-between the call stack frames due to optimizations, and things may make more sense once the complier's optimization passes are unwound.

If that doesn't explain things for you, please also include the output of that command in your follow-up.

— Ed Ford,  DTS Engineer

This seems to be in fact the same issue that was pointed out in https://developer.apple.com/forums/thread/760029

The output of xcrun crashlog /path/to/crash.crash contains

[  4] 0x0000000100bf81b8 MyApp`MyApp.MainViewController.showSettings(Any) -> () + 428 at MainViewController.swift:132:30
[  5] 0x0000000100ba252f MyApp`merged @objc MyApp.ConfirmMoveViewController.openSourceInFinder(Any) -> () + 79

merged @objc, if I understand correctly, indicates some compiler optimization. So it would seem that on the way to calling the showSettings method some reused part of the openSourceInFinder method is used, but the crash log makes it appear like openSourceInFinder was genuinely called.

Could this perhaps be something to be optimized in how macOS creates crash logs, or is there a way to consistently recognize this?

Are those frames from the LLDB output coherent to your understanding of how your app works? Based on your first post, I think they might not be, but want to make sure. If they still aren't, that will guide where we go from here.

— Ed Ford,  DTS Engineer

Well, the issue is that the crash report makes it look like a certain method is called which cannot be, and if I ignore that particular frame line, then everything is coherent. So if the explanation, as I assumed, is that merged @objc means that the method was not actually part of the call stack, then everything's ok. I just think that it shouldn't be listed there at all, because it's only confusing.

Would you mind opening an enhancement request for this, with both the crash log you have from the system and the one expanded by LLDB attached? (And please post the FB number here.) I'd like to treat this as a chance for improvement in the display of your crash log so that it was more intuitive to you what has happening.

Speaking broadly and not specifically to your examples, backtraces in optimized builds sometimes are surprising because of the optimizations. That's not to say that the crash report is wrong — it is correct to the actual instructions addresses that the system called in backtrace order to the crash — but that it may have either more or fewer frames than you expect from the source code because of what the optimizations added or removed. That's why I want to be specific that the above enhancement request regards displaying the optimization info that you needed to understand what happened, and not that what the crash report recorded the wrong instruction addresses.

— Ed Ford,  DTS Engineer

Thanks, I created FB16969117.