Thanks for the report.

 > Recipe 1:
 >
 > 1. Emacs -Q
 > 2. evaluate the following code in *scratch*:
 >
 >     (display-buffer (get-buffer-create "aaa")
 >                    '(display-buffer-pop-up-window
 >                      (window-width . 20)))
 >     ;; => correct window width for buffer "aaa"
 >
 >     (display-buffer (get-buffer-create "bbb")
 >                    '(display-buffer-pop-up-window
 >                      (window-width . 20)))
 >     ;; => changed and larger window width for buffer "bbb"

Recipe 1 is difficult to reproduce here.  I made my initial frame 169
columns wide to get a side-by-side split when displaying aaa and to
_not_ get another side-by-side split when displaying bbb.  What happens
is that ‘display-buffer-pop-up-window’ refuses to make a new window for
bbb (you can reduce the value of ‘split-width-threshold’ to get a split)
so ‘display-buffer’ rather uses aaa's window to display bbb via
‘display-buffer-use-some-window’.  The latter ignores the ‘window-width’
entry but applies ‘window--even-window-sizes’ instead.

So this is not a bug but admittedly a very confusing sequence of events.

 > Recipe 2:
 >
 > 1. Emacs -Q
 > 2. evaluate the following code in *scratch*:
 >
 >     (display-buffer (get-buffer-create "aaa")
 >            '(display-buffer-in-side-window
 >              (window-width . 20)
 >              (side . right)))
 >     ;; => correct window width for buffer "aaa"
 >
 >     (display-buffer (get-buffer-create "bbb")
 >            '(display-buffer-in-side-window
 >              (window-width . 20)
 >              (side . left)))
 >     ;; => correct width for buffer "bbb", but the window width of
 >     ;; buffer "aaa" is changed and becomes larger

This is a bug and I attached a fix for it.  Please try it.

Note, however, that in general you can rely on Emacs to preserve the
size of a window like that of aaa in recipe 2 if and only if you run
‘window-preserve-size’ on that window - either explicitly or via a
`preserve-size' argument.

martin

(BTW: Does the ibuffer scenario from bug#25115 work orderly now?)