This is a good idea.

I am currently leaning towards using a threading.Lock around `Pdb.interaction`, or `pdb.set_trace` with pdb++.

This would at least mitigate the issue where `pdb.set_trace` is used in some code that gets executed again, and two pdb instances would use the same input, e.g. when debugging web applications.

See also https://bugs.python.org/issue23163 for/about improving docs in this regard.