> Let's split the frame in two windows only, one for the *scratch* and the
 > other for *Messages*. From here you have a few lines only in the
 > *Messages* buffer. And untill now, every new lines introduced because of
 > a command output is followed. Moreover, I can see the "ghost" cursor at
 > the end of the buffer and it follows every *Messages* buffer updates.
 >
 > Then, from the *scratch* buffer, I call the describe-variable key
 > sequence ( C-h v xxx ). The current buffer being *scratch*, the
 > *Messages* buffer will be replaced with the *help* buffer.
 >
 > After finishing to read the description I want to quit it, because I
 > want the *Messages* buffer to come back. For that, I do: C-x o, thus
 > switching to the focus to the *Help* buffer, and I just type in 'q' to
 > quit it and instantly focus back to the *scratch* buffer.
 >
 > But: When I look into the *Messages* buffer, the "ghost" cursor is no
 > longer at the same place. It has moved in the middle of the buffer. From
 > there, the window position relative to the buffer remains the same.
 > Which is not good because I want to see the *Messages* updates when they
 > come in.

This is due to a bug in `display-buffer-record-window' which doesn't pay
attention to the buffer's `window-point-insertion-type' when storing the
window-point marker.  A similar bug occurs in `record-window-buffer', so
functions like `switch-to-prev-buffer' and `switch-to-next-buffer' are
affected by the same problem whenever they switch to a buffer whose
`window-point-insertion-type' is non-nil.

Since this bug constitutes a considerable annoyance and a regression wrt
23.4 I'd like to install the attached patch on the Emacs 24 branch.  It
principally only adds the necessary TYPE argument to the `copy-marker'
calls but looks more complex because I have to do this in the right
buffer and therefore rearranged the code a bit.  OK to install?

martin