Eli Zaretskii <eliz@gnu.org> writes:

>> > Isn't it better to check whether the file is already locked?  That
>> > way, we don't need any (error-prone) heuristics for when it's okay to
>> > ask the question and when it isn't.
>> 
>> If I understand the code correctly, lock_file() already checks this (the
>> '&& !NILP (Ffile_exists_p (fn))' part), but takes the wrong action in
>> that case.  So I'm not sure what we should do differently.
>
> AFAIU, 'fn' in lock_file is the file we want to lock, not the file we
> create to indicate the lock.  Am I missing something?

Oh, you're right, I was confused.  The patch below seems to work.