bug#12081: 24.1; buffer-predicate often not called

[martin rudalics]

> So, would something like the following patch be ok (after adding
> appropriate documentation changes)?

Sure.  Can you write a ChangeLog entry, one additional sentence for the
doc-strings and a more useful documentation of buffer-predicate for the
Elisp manual?

> Note that with these changes,
> there might still be some potentially odd behavior.  Here's a
> contrived example.  From Emacs -Q, evaluating the following in
> *scratch*:
>
> (progn
>   (set-frame-parameter nil 'buffer-predicate (lambda (x) nil))
>   (switch-to-buffer (get-buffer-create "foo"))
>   (switch-to-prev-buffer))
>
> will leave buffer "foo" displayed in the selected window, but the
> return value of of `switch-to-prev-buffer' will be #<buffer
> *scratch*>, because `next-buffer' was assigned during the scan of the
> previous buffers even though we did not switch to it.  Is such a
> misleading return value ok?

No.  It's an obvious bug not related to `buffer-predicate'.  It went
unnoticed so far because the return value of `switch-to-prev-buffer' is
neither documented nor used anywhere.  But in the case you describe
`switch-to-prev-buffer' should return nil just as in all other cases
resulting from

      (dolist (entry (window-prev-buffers window))
        (when (and (setq new-buffer (car entry))

when new-buffer is rejected.  I hope I fixed that now.  Please have a
look.

Thanks, martin