Stefan Beller <sbeller@xxxxxxxxxx> writes: > This version adds the actual thread safety, > that is promised in the documentation, however it doesn't add any optimization, > i.e. it's a single global lock. But as we do not expect contention, that is fine. Because we have to start _somewhere_, I agree it is a good approach to first try the simplest implementation and then optimize later, but is it an agreed consensus that we do not expect contention?