This works around the problem that the threadnames deduction switches
the current thread and thereby invalidates the assumption that the
thread mentioned in the *stopped response is the "current" thread.
There's already a proper fix in master including some performance
enhancements.