bug#25115: 26.0.50; ibuffer fails to show confirmation buffer when there are side windows

[martin rudalics <rudalics@gmx.at>]

 > When there are side windows, ibuffer fails to show the comfirmation buffer for buffers to be killed.
 >
 > 1. emacs -Q & create a side window by
 > (pop-to-buffer "test"
 >                 '(display-buffer-in-side-window
 >                   (side . right)
 >                   (width . 0.2)))
 > 2. switch to another buffer and M-x ibuffer
 > 3. press 'D', mark two or more buffers for killing
 > 4. press 'x'
 >
 > then an error 'ibuffer-confirm-operation-on: Cannot split side window or parent of side window' is issued.
 >
 >
 > In GNU Emacs 26.0.50.2 (x86_64-pc-linux-gnu, GTK+ Version 3.18.9)
 >   of 2016-12-03 built on lgw01-50
 > Windowing system distributor 'The X.Org Foundation', version 11.0.11804000
 > System Description:    Ubuntu 16.04.1 LTS

Thanks for the report.  Since I don't use ibuffer I just tried to follow
your recipe.  In (2) I switched to *Messages* in the side window by
selecting it with the mouse from the Buffers menu.  If, after M-x
ibuffer, I now type ‘D’ in the side window, ibuffer prompts me whether I
really want to kill *Messages* so I miserably failed to "mark two or
more buffers for killing".

If instead in (3) I repeatedly type ‘m’ to mark all buffers listed in
the side window and then type ‘D’, my main window gets split, in the
window at the bottom I get a list of the selected buffers and in the
minibuffer I'm asked whether I really want to kill 3 buffers.  When,
instead of ‘D’ I type ‘x’, I am asked in the minibuffer whether I really
want to kill the buffer at the current line of the side window.

I don't understand the ibuffer code (partly because I don't understand
what an interactive Lisp closure is and how it works).  So if you can't
give me a recipe I can follow it might be best if you debug the ibuffer
code yourself to find out why and how it tries to split the side window.

Finally note that side windows are preferably populated via
‘display-buffer’, so the "switch to another buffer" and "M-x ibuffer" in
step (2) already sound like a bad idea.

Thanks, martin