Juri Linkov <juri@jurta.org> writes:
>> I note that `bury-buffer' removes the buffer from the frame-local
>> buffer lists of all frames, not just the selected one.  This is an
>> unwanted side-effect in this case, as we want the effects of
>> `next-buffer' to remain frame-local.  I propose to add an optional
>> parameter to `bury-buffer' to support this.
>
> I think this change should be made for all invocations of
> `bury-buffer', not only from `next-buffer'. [...] And so there is no
> need for a new optional parameter.

That's fine by me.

> Also I think recoding a buffer to the new frame parameter should be
> made in `bury-buffer' instead of `next-buffer'.    Some commands call `bury-buffer'
> directly, and it would be good to record their buffers in the
> frame-local parameter.

Ah, good idea.  I implemented this in the patch at the end of this
message.  It involves adding a new element to the frame struct, and
changing switch-to-buffer, bury-buffer and kill-buffer etc. to
maintain the list.

> Instead of introducing a new frame parameter `previous-buffer-list'
> what do you think about reusing the existing frame parameter `buffer-list'?
> It could have a special marker delimiting a list of frame's "next" visited
> buffers and "previous" visited buffers.  A nil value could serve as
> such special marker.  So for example, (frame-parameter nil 'buffer-list)
> could contain:
>
> (#<buffer *b+1*> #<buffer *b+2*> nil #<buffer *b-2*> #<buffer *b-1*>)

I thought of this as well, but it would be an incompatible change that
is not strictly necessary.  The 'buffer-list frame parameter has been
in Emacs since (I think) 1997; some external packages may rely on its
current contents.  Also, using a special marker seems a bit less clean
to me than using a separate parameter.  (It is also less efficient.)

The below patch also renames `prev-buffer' to `previous-buffer', and
extends it to support the buffer-predicate frame parameter.  It
incorporates all fixes that have been suggested in this thread, except
for the customization option for disabling frame-local behaviour.
(`other-buffer' has been unconditionally using the frame-local buffer
list for years; I don't see a point in making it configurable for
next-buffer/previous-buffer either.)

If there are no objections, I will document the changes and install
them in CVS.