[There is a simple reproducer at end of the bug report.]

First, the documentation of ff-other-file-alist fails to mention that
the value of the variable does not have to be alist, it can be a symbol
as well.

More importantly, if the associated value is a function, then there's no
way for the function to signal that it cannot find a related file.

If the function returns "/nonexistent", then ff-find-the-other-file
(with the default settings) will try to create "/nonexistent".

If the function returns nil, then ff-find-the-other-file will call
ff-get-file-name and:

  (ff-get-file-name '("." "/usr/include" "/usr/local/include/*") nil nil)
  ==> "/home/nemethf/.emacs.d/News/drafts/drafts/679"

"emacs -Q -l bug.el" reproduces the problem by setting
uniquify-buffer-name-style.  However, my uniquify-buffer-name-style is
'forward and not nil.  Maybe Gnus changes uniquify-buffer-name-style
under the hood, because drafts/679 corresponds to a buffer named
"*sent wide reply to Somebody*<2>".

I tend to think this bug in line 577 of find-file.el:
http://git.savannah.gnu.org/cgit/emacs.git/tree/lisp/find-file.el?id=958924a8126cf532d44c4b446d13ed744438cc9b#n577
But I don't understand the purpose of that string-match-p.

Thanks.