set-window-start moves point without scrolling

set-window-start moves point without scrolling

[Michael Heerdegen]

Hello,

this seems strange to me: in emacs -Q, eval this:

(progn
  (setq-default scroll-margin 2)
  (pop-to-buffer (get-buffer-create "test"))
  (insert "1\n2\n3")
  (goto-char 1)
  (set-window-start nil (window-start)))

Then, point in buffer "test" is at the beginning of line 3.  This
doesn't happen in Emacs 23, only in 24.  And it doesn't seem right - the
doc of `set-window-start' tells that point is only changed if would be
out of sight otherwise.  Is this a bug or a feature?


Regards,

Michael.

Re: set-window-start moves point without scrolling

[Eli Zaretskii]

> From: Michael Heerdegen <michael_heerdegen@web.de>
> Date: Mon, 18 Mar 2013 23:44:01 +0100
> 
> (progn
>   (setq-default scroll-margin 2)
>   (pop-to-buffer (get-buffer-create "test"))
>   (insert "1\n2\n3")
>   (goto-char 1)
>   (set-window-start nil (window-start)))
> 
> Then, point in buffer "test" is at the beginning of line 3.  This
> doesn't happen in Emacs 23, only in 24.  And it doesn't seem right - the
> doc of `set-window-start' tells that point is only changed if would be
> out of sight otherwise.  Is this a bug or a feature?

Feature.  You set the scroll-margin to 2, so Emacs avoids putting
point inside the margin.

Does this give you trouble in any real-life situation?

Re: set-window-start moves point without scrolling

[Michael Heerdegen]

Eli Zaretskii <eliz@gnu.org> writes:

> > (progn
> >   (setq-default scroll-margin 2)
> >   (pop-to-buffer (get-buffer-create "test"))
> >   (insert "1\n2\n3")
> >   (goto-char 1)
> >   (set-window-start nil (window-start)))
> > 
> > Then, point in buffer "test" is at the beginning of line 3.

> Feature.  You set the scroll-margin to 2, so Emacs avoids putting
> point inside the margin.

Ok, I see, thanks.

> Does this give you trouble in any real-life situation?

Dunno.  The background: yesterday, I tried "multiple-cursors" by Magnar
Sveen (whom I CC'd):

  https://github.com/magnars/multiple-cursors.el

The idea is, yes, to simulate multiple cursors.

To avoid performing of automatic window scrolling when the current
editing command is performed at the positions of the "other" cursors,
these actions are wrapped in this macro:

(defmacro mc/save-window-scroll (&rest forms)
  "Saves and restores the window scroll position"
  `(let ((p (set-marker (make-marker) (point)))
         (start (set-marker (make-marker) (window-start)))
         (hscroll (window-hscroll)))
     ,@forms
     (goto-char p)
     (set-window-start nil start)
     (set-window-hscroll nil hscroll)
     (set-marker p nil)
     (set-marker start nil)))

(the local vars should become uninterned symbols, but that's not related)

This code triggers the effect in my example above: point (i.e., the
"real" cursor) is moved as a side effect if point was in the first
visible window line and scroll-margin > 0.

How would a better implementation of this macro look like?  Would it be
sufficient to use a non-nil third argument for `set-window-start', or
would this not reverse automatic scrolling reliably?

BTW, I already tried to wrap the `set-window-hscroll' inside
`save-excursion' or (let ((scroll-margin 0)) ...), but this doesn't help
(I guess because scrolling is performed afterwards while redisplaying.)


Thanks so far,

Michael. 

Re: set-window-start moves point without scrolling

[Eli Zaretskii]

> From: Michael Heerdegen <michael_heerdegen@web.de>
> Date: Tue, 19 Mar 2013 17:26:42 +0100
> Cc: Magnar Sveen <magnars@gmail.com>
> 
> To avoid performing of automatic window scrolling when the current
> editing command is performed at the positions of the "other" cursors,
> these actions are wrapped in this macro:
> 
> (defmacro mc/save-window-scroll (&rest forms)
>   "Saves and restores the window scroll position"
>   `(let ((p (set-marker (make-marker) (point)))
>          (start (set-marker (make-marker) (window-start)))
>          (hscroll (window-hscroll)))
>      ,@forms
>      (goto-char p)
>      (set-window-start nil start)
>      (set-window-hscroll nil hscroll)
>      (set-marker p nil)
>      (set-marker start nil)))
> 
> (the local vars should become uninterned symbols, but that's not related)
> 
> This code triggers the effect in my example above: point (i.e., the
> "real" cursor) is moved as a side effect if point was in the first
> visible window line and scroll-margin > 0.
> 
> How would a better implementation of this macro look like?  Would it be
> sufficient to use a non-nil third argument for `set-window-start', or
> would this not reverse automatic scrolling reliably?

I think using a non-nil 3rd argument to set-window-start should do
what you want.  At least it does that in your recipe that started this
thread.  My reading of the code which deals with the effect of the
optional arg to set-window-start is that passing a non-nil value there
inhibits the code which forces point outside the scroll margin in this
case.

> BTW, I already tried to wrap the `set-window-hscroll' inside
> `save-excursion' or (let ((scroll-margin 0)) ...), but this doesn't help
> (I guess because scrolling is performed afterwards while redisplaying.)

Yes.

Re: set-window-start moves point without scrolling

[Michael Heerdegen]

Eli Zaretskii <eliz@gnu.org> writes:

> I think using a non-nil 3rd argument to set-window-start should do
> what you want.  At least it does that in your recipe that started this
> thread.  My reading of the code which deals with the effect of the
> optional arg to set-window-start is that passing a non-nil value there
> inhibits the code which forces point outside the scroll margin in this
> case.

That sounds good - thanks for checking! 


Regards,

Michael.