bug#12170: save-excursion fails boundary case with recenter

[martin rudalics <rudalics@gmx.at>]

 > As mentioned by Martin, this is a misunderstanding about what
 > save-excursion does and what `point' is.

As a matter of fact, I thought so initially but did not mention it.

 > Every buffer can have many different `point's (it basically has one per
 > window, accessible via `window-point' and changeable via
 > `set-window-point', plus one for itself, called `point').
 > `save-excursion' preserves only `point'.
 > `recenter' changes `window-point'.

The scenario shows up with one single window showing that buffer.

 > But `point' and `window-point' are linked (point is set to window-point
 > and vice-versa in various occasions), so they're often confused.
 >
 > It seems your real problem is not that `point' changes but that the
 > cursor ends up in a different position than the one you wanted (the
 > cursor position, is represented by `window-point' rather than by
 > `point'), right?
 >
 > If so, you want to preserve window-point.  And there's nothing quite
 > like save-excursion to preserve window-point.  You can try
 > save-window-excursion, tho it will do a lot more than you asked for.
 > Or otherwise manually read window-point at the beginning and
 > set-window-point at the end.

Neither of these cut it.  Maybe the purpose of `recenter' is to
overrule, if necessary, any further point movement until the next
redisplay.  But I still don't understand how putting in a call of
`set-window-start' with NOFORCE t can calm it.

martin