bug#20943: 25.0.50; Dired buffers are not always auto-reverterd

[Mark Karpov <markkarpov@openmailbox.org>]

OK, I've spent an hour testing it, trying to find some way to reliably
reproduce the problem. It's quite random, but I've found a way to
reproduce it in about 50% of cases.

> Does it happen with some specific type of directories, like those
> mounted in some special way?

No, I can reproduce this in my home directory, which is not mounted in
any unusual way.

> Furthermore, which kind of changes of the directory are you referring
> to? Reading `dired-buffer-stale-p', the dired buffer shall be reverted
> only if there is a new file in the directory, or a file is
> removed. Just changing (the timestamp of) a file in the directory
> shall not revert the dired buffer.

By changes I mean deletion of files or creation of files, see below.

How to reproduce:

1. emacs -Q

2. Open some directory in Dired.

3. Invoke shell or Emacs shell. Now I think this bug is most likely
   happens when directory is changed in an automated way (maybe I'm
   mistaken). Personally I use a script that transforms some files
   (script written in Python, splits music files according to CUE file +
   some magic, but it doesn't really matter). That means that some files
   are created and old files are deleted.

4. Once I press C-m or Enter, there are two possible cases (each having
   about 50 % probability): either all changes are reflected in Dired
   buffer (that's in other window, window in Emacs sense of the word),
   or all the changes go invisible and I have to switch to Dired buffer
   and press g to see results.

That's pretty much all I can say. Not very helpful, but this is how I
can more or less reliably reproduce it.

> A second thought: auto-reverting dired buffers has been changed to use
> file notifications. Does it make a difference, when you eval the form
> (setq auto-revert-use-notify nil) ?

I'll test it as soon as I can.