Behavior on uncaught exception is different between -C panic=abort and -C panic=unwind

[tomaka]

When a program compiled with -C panic=abort panicks, the stack trace is printed and an illegal instruction is triggered (with the ud2 opcode).
When a program compiled with -C panic=unwind doesn't catch a panic in the main thread, the stack trace is printed and the program aborts.

I don't have any strong opinion in favor of triggering an illegal instruction vs aborting, but I think the behavior should be the same between panic=abort and panic=unwind.

[huonw]

and the program aborts

To be clear, I think this should read "and the program calls exit(101)", which is different to both libc::abort() and std::intrinsics::abort() (the latter is actually how the ud2s are created).

[alexcrichton]

I'm not sure I understand what this issue is about, unfortunately. The information about stack traces above seems irrelevant, and otherwise it seems obvious that the behavior of a panic changes when you change the panic runtime? Panicking with panic=unwind without a catch is undefined behavior, so it's just an implementation detail what it does.

[retep998]

I'd really prefer if we called DebugBreak whenever there was an uncaught panic on Windows (or at the very least if IsDebuggerPresent()) just to make debugging a bit easier, because otherwise with panic=unwind it is silently caught and the program exits cleanly (!) with an exit code of 101.

[briansmith]

I'd really prefer if we called DebugBreak

Call __debugbreak() instead. DebugBreak can screw up call stacks (see old fixed bugs in the Firefox bug tracker).

[retep998]

@briangreenery Of course __debugbreak() is an intrinsic so we'd have to either hope LLVM has that intrinsic, or use inline assembly.

[alexcrichton]

In any case as I mentioned before the behavior of -C panic=unwind is undefined if there's nothing to catch a panic, and with a lack of response I'm going to close this issue as well.

If there's something Windows-specific that should happen here, it would be best to open a new issue.

[retep998]

I opened a separate issue for that #34502