AP => Bluesky: support usernames that begin or end with _, -, or ~

[snarfed]

We currently convert these characters to -, which ATProto allows in the middle of handles, but not at the beginning or end.

Fediverse usernames that begin or end with these characters are rare, but definitely exist. Examples in #982.

Users need to translate handles themselves in order to search for users across the bridge, so I'm trying to keep that translation UX as simple as possible. Barring a more complicated translation process, I don't yet have a good idea for how to support these usernames.

[qazmlp]

It's a bit out there, but maybe IDNA2008 could work here (also for internationalisation)? The Bsky names are supposed to be domain names, after all, and rendering IDNs at least for supporting TLDs may be something desireable on their end too.

Unicode has a good FAQ with a bunch of links here: https://www.unicode.org/faq/idn.html
The easiest option would be to just use an IDNA2008 library, though.

The scheme isn't pretty, per se, but iinm it keeps the ascii-part of the handle readable, at least, and is unambiguous.

[snarfed]

Oof. The use case is, a Bluesky user sees someone in the fediverse, and they want to find and follow their bridged account on Bluesky. To do that, they need to manually translate the fediverse address into a bridged Bluesky handle. Punycode, IDNA2008, etc are useful, but not human-usable for that use case.

[qazmlp]

Right, but this is something they wouldn't be able to do anyway with any sort of non-trivial mapping, which you'd need to cover more than a very limited set of cases. They'd have to use https://fed.brid.gy/ to look up the account (once the request function is online), and that can do a complex conversion for them easily.

(You could still give out ambiguous mappings on a first-come-first-serve basis, of course, but personally I use such characters starting/trailing especially when the plain name is taken, so collisions seem decently likely to me.)

Bsky also indexes users by display name, so such-translated bridged users would most likely still appear quickly in search there. I can't test this since my display name is also my handle, but the bridged account is clearly indexed and others do show up in search for their display name specifically.