Why in Building Protocols, Like Code, Starting over Is Dumb
5 hours ago
- The IETF held a BoF to create a new protocol similar to TLS but using MLS for key exchange, which is considered a bad idea.
- Replacing protocol components wholesale is common in AI-generated proposals, ignoring the wisdom of preserving existing systems.
- Rewriting protocol components resembles code rewriting, but protocols have unique risks like missing unknown use cases.
- HTTP/2 lost HTTP/1.1 features like mid-request client authentication, showing the difficulty of catching all use cases.
- Changing a protocol by replacing parts forces the ecosystem to change or results in forking, which breaks interoperability.
- Forking a protocol destroys the value of universal communication, creating two incompatible systems.
- Existing protocols have technical debt but share maintenance burden; fixing via extensions is better than replacing.
- Major refreshes like TLS 1.3 show that incremental improvements can clear cruft while preserving interoperability.
- Responsible stewardship requires learning from past lessons, not ignoring them.