Currently we fail to generate a stack trace if any code LibunwindStepIteration generates an error, even if the error is recoverable, or only applies to a single frame. AF-628 is a good example of why this is a bad idea. A less informative, or even damaged, stacktrace is far better than no stacktrace at all, particularly due to the lack of corefiles in Atlas. This is especially true for weird memory bugs, where code can fail in all sorts of unexpected ways.

We should refactor our backtrace code to always print whatever state it has available, even if that state is incomplete or damaged.