display-buffer and temp-buffer-resize-mode

display-buffer and temp-buffer-resize-mode

[martin rudalics]

With the customization

(customize-set-variable
  'display-buffer-alist
  '(("*shell*" display-buffer-in-direction (direction . bottom))))

M-x shell will display a three lines high *shell* window on the bottom
of the frame and not - as probably expected - a window that occupies
about half of the root's height.

The reason is that 'display-buffer-in-direction' unconditionally asks
for resizing the new window via

	(setq alist
	      (append alist
		      `(,(if temp-buffer-resize-mode
		             '(window-height . resize-temp-buffer-window)
	                   '(window-height . fit-window-to-buffer))
	                ,(when temp-buffer-resize-mode
	                   '(preserve-size . (nil . t))))))

A user can fix that via

(customize-set-variable
  'display-buffer-alist
  '(("*shell*" display-buffer-in-direction (window-height . nil) (direction . bottom))))

but that's undocumented and maybe not what we should ask users to do.

Note that 'display-buffer--maybe-at-bottom' has the same problem - but
that's an internal function.

There are numerous ways to fix that, all requiring some sort of surgery:

- We could simply document the current behavior.  This means that for
   using 'display-buffer-in-direction' as helper function for say
   'display-buffer-at-bottom' we would have to append an extra
   (window-height . nil) alist entry in order to suppress the resizing
   ('display-buffer-at-bottom' currently does not suffer from that
   syndrome).

- We could inhibit the resizing.  This would require a slight amendment
   in the call in 're-builder' - all other calls in our code base provide
   an extra 'window-height' alist entry.  Nevertheless, it would be an
   incompatible change of, admittedly undocumented, behavior.  (Note that
   for 'display-buffer--maybe-at-bottom' we would then have to fix
   'hack-local-variables-confirm'.)

- We could try to refine the snippet above in some way.  For example,
   ask for 'resize-temp-buffer-window' only if the buffer to be displayed
   is "temporary".  That's a bit tricky because 'temp-buffer-resize-mode'
   is a global mode and that variable does not allow to tell whether the
   buffer to display is temporary at all.  In either case, the
   'fit-window-to-buffer' entry would have to be dropped and we still
   have an incompatible change.

Any ideas?

Thanks, martin

Re: display-buffer and temp-buffer-resize-mode

[Juri Linkov]

> There are numerous ways to fix that, all requiring some sort of surgery:
>
> - We could simply document the current behavior.  This means that for
>   using 'display-buffer-in-direction' as helper function for say
>   'display-buffer-at-bottom' we would have to append an extra
>   (window-height . nil) alist entry in order to suppress the resizing
>   ('display-buffer-at-bottom' currently does not suffer from that
>   syndrome).

Shouldn't both 'display-buffer-in-direction (direction . bottom)'
and 'display-buffer-at-bottom' be consistent with each other
in regard to resizing, and either both resize or not?

Even this comment says they should have the same implementation:

  ;; This should be rewritten as
  ;; (display-buffer-in-direction buffer (cons '(direction . bottom) alist))
  (defun display-buffer-at-bottom (buffer alist)

> - We could inhibit the resizing.  This would require a slight amendment
>   in the call in 're-builder' - all other calls in our code base provide
>   an extra 'window-height' alist entry.  Nevertheless, it would be an
>   incompatible change of, admittedly undocumented, behavior.  (Note that
>   for 'display-buffer--maybe-at-bottom' we would then have to fix
>   'hack-local-variables-confirm'.)

Also I found 'ediff-setup-windows-plain-compare' where
'display-buffer-in-direction' is without explicit 'window-height'.

But the decision whether to resize or not could depend on statistics:
if the majority of uses prefer resizing, then resize by default.

> - We could try to refine the snippet above in some way.  For example,
>   ask for 'resize-temp-buffer-window' only if the buffer to be displayed
>   is "temporary".  That's a bit tricky because 'temp-buffer-resize-mode'
>   is a global mode and that variable does not allow to tell whether the
>   buffer to display is temporary at all.  In either case, the
>   'fit-window-to-buffer' entry would have to be dropped and we still
>   have an incompatible change.

Another variant is to treat synonyms differently where e.g.
'(direction . bottom)' will resize, but '(direction . below) won't.