Re: partial-completion-mode and deleting symlink problem

[Dan Nicolaescu]

Richard Stallman <address@hidden> writes:

  >     Does any of the following functions need to change?
  >     (assume that /tmp/DELETEME is a symlink that points to a non-existent 
file)
  > 
  >     (read-file-name-internal "/tmp/DELETEME" "/tmp" 'lambda)
  >     currently returns nil. Should it return t? 
  > 
  > I think it depends on the purpose.  In most cases, probably nil is
  > right.  For some specific purposes, for operating on symlinks, t would
  > be better.

Here's what is going on here. 

The call chain looks like this:
delete-file 
   read-file-name with the PREDICATE argument Qnil, because of that
      this code is executed:
            specbind (intern ("read-file-name-predicate"),
                  (NILP (predicate) ? Qfile_exists_p : predicate));
   PC-do-completion
        test-completions
             read-file-name-internal <- this returns Qnil because of
                this code:
                     if (!NILP (Vread_file_name_predicate))
                         return call1 (Vread_file_name_predicate, string);
                      read-file-name-predicate was bound to
                      file-exists-p, and (file-exists-p "/tmp/DELETEME")
                      return nil.


Given that some of the above functions are used in quite a few places,
I don't know what is the best solution for solving this. 
Suggestions are welcome. 
On the other hand this issue has been around forever, so it's not
something that is necessary to be solved before the release...

Thanks
                --dan