I am not sure if this is intentional, but the new `file-name-split'
is a bit unintuitive in this edge-case:

   (file-name-split "/")        → ("" "" "")

while

   (file-name-split "/a")       → ("" "a")
   (file-name-split "a/")       → ("a" "")
   (file-name-split "a/b")      → ("a" "b")

another few peculiar cases might be

   (file-name-split "//")       → ("/" "" "")
   (file-name-split "///")      → ("" "" "")
   (file-name-split "////")     → ("" "" "")

as I'd expect '/' (in the first case) to never be part of a file name
(at least on a *nix system).

This all appears to happen only if there is no actual file name to be
found:

   (file-name-split "a/")       → ("a" "")
   (file-name-split "a//")      → ("a" "")
   (file-name-split "a///")     → ("a" "")

One simple solution might just be to remove all empty strings from the
return value of `file-name-split', as to my knowledge empty file names
are now allowed (?):