Re: scroll-restore.el

[martin rudalics]

> I think this is trying to implement too many different features, and
> we don't want them all.  I would rather implement just the feature of
> restoring point after you scroll back to the previous screen, in a
> simple way.

It's the only feature I use in practice.  But I never scroll a window
when the region is highlighted.  Scrolling with a highlighted region is
a real pain when `point' gets relocated to stay on screen.

> In general, it is best to avoid `pre-command-hook' whenever a finer
> tool can be used, and likewise `post-command-hook'.

Goes without saying.  Last summer someone on bug-gnu-emacs lamented the
absence of such a feature.  I just wanted to know whether I liked it and
use it ever since.

>     ;; Note: We can't use `point-before-scroll' for our purposes because
>     ;; that variable is buffer-local.
>
> WHY isn't `point-before-scroll' enough?

The variable must be window-local to handle the case of independently
scrolling two or more windows showing the same buffer (setting
`mouse-wheel-follow-mouse' to t makes that possible).

> Would it be enough just for the feature of restoring point?
>
>                                   We need a variable that recorded
>     ;; `window-point' before a sequence of scroll operations.  Also
>     ;; `point-before-scroll' is not handled by mwheel.el and some other
>     ;; commands that do implicit scrolling.
>
> I presume we can fix `point-before-scroll' to DTRT in those cases.  In
> general, if a primitive feature isn't suitable for the job it is meant for,
> let's make it suitable, rather than do the job in a heavy-handed way.

I'll have to look into how `point-before-scroll' is currently used by
the scroll-bar / mouse code and whether it can (or should) be safely
made window-local.

> If we want a few commands to do a certain thing, then it is a bad idea
> to use `post-command-hook' to check for them.  Instead we should
> change those commands to do it, whatever IT is.  We can add a new
> specific hook, and run it from commands and features that scroll.
>
> People often choose `post-command-hook' rather than change C code.
> But when we design changes to install in Emacs, we should not design
> them based on the premise of "avoid changing the C code."

Fully agreed.