Filesystem Paths Are Underspecified

What is a filesystem path on Linux? (This question is even more painful on Windows.) It's tempting to believe that it's a list of strings (the path components) encoded as a single string (by using '/' as a delimiter). This is wrong.

First, a path is not a UTF-8 encoded string. Many bytestrings that are valid as UTF-8 encoded strings are not valid as filesystem paths, and many bytestrings that are valid as filesystem paths are not valid as UTF-8 encoded strings.

Second, a path is not (just) an ISO-8859-1 encoded string. Many bytestrings that are valid as ISO-8859-1 encoded strings are not valid as filesystem paths.

You could argue that a path is an ISO-8859-1 encoded string-that-doesn't-contain-the-null-character, but that's not a standard interpretation.

Third, even if you interpret a path as an ISO-8859-1 encoded string-that-doesn't-contain-the-null-character, a path component still isn't (just) a string. At best, it's a string-that-doesn't-contain-the-null-character-or-the-slash-character.

Fourth, a path is not (just) a bytestring. At best, you can say that it is a bytestring-that-doesn't-contain-0x00.

Fifth, even if you interpret a path as a bytestring-that-doesn't-contain-0x00, a path component still isn't (just) a bytestring. At best, you can say it is a bytestring-that-doesn't-contain-0x00-or-0x2F.

So where does that leave us? I think we're forced to define a filesystem path as a list of bytestrings-that-don't-contain-0x00-or-0x2F, encoded as a single bytestring-that-doesn't-contain-0x00 (by using 0x2F as a delimiter and appending 0x00 as a sentinel).