A modest proposal (I'm sure this is controversial so flame away...) A big part of the problems in i18n in IETF protocols have to do with extending protocol elements from ASCII to Unicode, and how to avoid difficulties when that happens. Protocol elements include domain names, URLs, email addresses, file names But where do these Unicode names come from? They're not arbitrarily generated by automated processes, they're constructed from strings that are selected, typed in, registered. So focus on encouraging people to choose strings that won't give problems. A large specification of all of the use cases to avoid is very difficult to write and hard to review. There are very many special cases (final sigma, umlauts, private name characters, non-normalization of combined forms) with expertise widely distributed. I'm not sure the solution is "more specs"; in fact, there are many obscure special cases, and the specs are very difficult to write and review. I wonder if there's any interest in building an open-source service that would, when given a proposed domain name or URL or email address, tell you what problems various subsets of users would have when trying to deploy that name (e.g., names that don't display properly on popular platforms, names that can't be reliably typed in correctly even if they can be viewed, those that are likely to get confused with other similar but different names). Perhaps get started at a Hackathon? I did reserve the domain name "caniuse.name" that I will offer to any sincere effort.
