bug#24803: Redirection problem with separate minibuffer frame

[Stefan Monnier <monnier@IRO.UMontreal.CA>]

> Please revert the frame.c change so we can be sure which of the two is
> the real culprit.

Reverting the frame.c change seems to fix the problem.

BTW I also just noticed that the bogus (the one I get at the end of my
recipe) focus is "mutual":
- when the mouse points at the minibuffer window, the focus is in the
  *scratch* buffer.
- when the mouse points in the *scratch* window, the focus is in the
  minibuffer!

> Works as intended on both Windows XP and a GTK+ build with XFCE on
> Debian.  I use focus-follows-mouse plus auto-raise-frame though the
> minibuffer does _not_ get autoraised when moving the mouse there.
> Actually, that's what I would call a misbehavior here ;-)

I don't use auto-raise of any kind.  But yes, it's probably dependent on
some aspect of the window manager event sequences.

> Hmm... This seems to indicate that I do not remove the redirection when
> exiting the minibuffer.  Could you try to augment in read_minibuf_unwind
>
>   if (minibuf_level == 0)
>     resize_mini_window (XWINDOW (window), 0);
>
> to something that for each frame redirects focus to itself?

Still haven't found the time to try this, but I just want to mention
that until recently, the focus redirection was usually nil rather than
"redirected to itself".

I'm not sure if there should be a difference between these two states,
but I have the suspicion that not all the C code handles those two
states in the same way (then again, last time I looked at the
redirection code, I concluded that I just don't understand how it's
supposed to work, so maybe it's just my misunderstanding).


        Stefan