all the problems you call out are implementation bugs, not specification bugs. If people can't implement it right, it's not the spec's fault - though we should

all the problems you call out are implementation bugs, not specification bugs. If people can't implement it right, it's not the spec's fault - though we should definitely strive for "easy to implement". "Freezing TLS" doesn't really make sense either - you'll probably be able to use TLS 1.1/1.2 for years to come, and it takes a LONG time for a new version of the protocol to roll out and gain support. If a TLS implementation wanted to focus on bugs and quality instead of implementing new features, they totally could.