Rethinking fsinfo()

Realistically, if you want to pass raw bytes through a text-formatted thing, you should be using Base64 or something similar. This of course means that paths become unreadable to humans without decoding, but you could have a flag indicating whether a path has been escaped, and then only escape things that aren't valid UTF-8. Alternatively, you could encode the "bad bytes" with \\x00 through \\xFF, which is valid in a JSON string (the backslash is escaped, so it's "just" a backslash followed by three letters), but could be confused with a real filename (so you would need to invent further escaping for that case, as described in https://xkcd.com/1638/).

Or, if you think letting people create files with ridiculous names was a bad idea to begin with, you could simply declare non-UTF-8 paths unsupported and spit out invalid JSON if the user tries to create one. Much userspace software already does something like that anyway (see for example Python 3's surrogateescape hack). But then a lot of parsers will work just fine the vast majority of the time, and break on an obscure condition that the average engineer may not even realize is possible. So that's probably not ideal...