Indeed. We don't know where the bug happens when tracing, so each
trace-point just writes its hash value into it. And nothing restores it,
since nothing knows that the resume is "done" (a lot of problems happen
due to X interactions much later than the 'core' resume thing).