bug#29111: 26.0.90; Erc keep-place module broken with new default of switch-to-buffer-preserve-window-point

[martin rudalics]

> Keep place adds a hook so that, when a new message comes in on a
> non-visible buffer *and* point is at the bottom, point is moved up by
> one line. Doing this means that point stays on the last 'read'
> message. Here's the relevant code from keep-place:
>
>> (defun erc-keep-place (ignored)
>> "Move point away from the last line in a non-selected ERC buffer."
>> (when (and (not (eq (window-buffer (selected-window))
>>                      (current-buffer)))
>>             (>= (point) erc-insert-marker))
>>    (deactivate-mark)
>>    (goto-char (erc-beg-of-input-line))
>>    (forward-line -1)))

You could try to pretend that the position at which the buffer was
previously shown in a window is the position calculated here.  Add after
the

(forward-line -1)

something like

;; For all non-minibuffer windows on all frames check whether the
;; current buffer was shown in that window before.  If so, update the
;; window point positions stored for the buffer to the position we just
;; calculated.
(dolist (frame (frame-list))
  (walk-window-tree
   (lambda (window)
     (let ((prev (assq (current-buffer) (window-prev-buffers window)))
           (next (assq (current-buffer) (window-next-buffers window))))
       (when prev
         (setf (nth 2 prev) (point-marker)))
       (when next
         (setf (nth 2 next) (point-marker)))))
   frame nil 'nominibuf))

Completely _untested_ so you may have to tweak it appropriately!

> Previously, this worked fine, since moving point on non-visible buffers
> meant the movement persisted till the buffer was next 'shown'. However,
> the new default of `switch-to-buffer-preserve-window-point' ensures that
> point stays wherever it was 'last seen'. I'm not sure how it does it,
> but I think it's saving point when a buffer is hidden and restoring it
> when it's seen again
>
>> (when (and entry
>>                   (or (eq switch-to-buffer-preserve-window-point t)
>>                       displayed))
>>          ;; Try to restore start and point of buffer in the selected
>>          ;; window (Bug#4041).
>>          (set-window-start (selected-window) (nth 1 entry) t)
>>          (set-window-point nil (nth 2 entry)))

Right.  So with the hack above you would set (nth 2 entry) to the value
keep-place calculated.

>> That's a possibility, but it would be too radical to go into 26.1, so
>> I think we should explore the less drastic alternatives first.
>
> I agree with Eli that this is too big of a change at this point, but I
> think this is the best long term solution. Perhaps we could
> `make-variable-buffer-local' on `switch-to-buffer-preserve-window-point'
> when enabling keep-place (later, of course).

I can't yet assess all implications of such a solution so I'd certainly
defer it until all other measures have been exhausted.

>> Since ‘keep-place’ is some sort of a minor mode, enabling it should warn
>> the user about the ‘switch-to-buffer-preserve-window-point’
>> incompatibility.  But I'm not familiar with ‘define-erc-module’ so we'd
>> probably need someone knowledgeable to do that.
>
> I think this is probably the best idea before the emacs 26 release. The
> definition of 'keep-place' is currently
>
>> (define-erc-module keep-place nil
>>    "Leave point above un-viewed text in other channels."
>>    ((add-hook 'erc-insert-pre-hook  'erc-keep-place))
>>    ((remove-hook 'erc-insert-pre-hook  'erc-keep-place)))
>
> I think we should be able to simply do something like
>
>> (define-erc-module keep-place nil
>>    "Leave point above un-viewed text in other channels."
>>    ((add-hook 'erc-insert-pre-hook  'erc-keep-place)
>>     ;; poor name, but just an example
>>     (erc-check-if-switch-to-buffer-preserve-and-warn))
>>    ((remove-hook 'erc-insert-pre-hook  'erc-keep-place)))
>
> I would be happy to submit a patch for this, if the Erc maintainer is
> busy. Does this seem like a good solution, or does anyone see a better
> way around this?

If all else fails let's do that.  First please try the proposal above.

martin