Re: [Emacs-diffs] /srv/bzr/emacs/trunk r109890: Do not mark objects from deleted buffers, windows and frames.

[Dmitry Antipov <dmantipov@yandex.ru>]

On 09/06/2012 11:28 AM, martin rudalics wrote:

>  > Since we can't implement this just now, I'm reverting window marking;
>  > but I think that I'll spent more time on this :-).
>
> ... and please tell the people from bug 12251 about it ;-)

OK

>  > Hm. For example, killed buffers may sit in all_buffers for a while, and
>  > still
>  > have from tens to thousands reachable objects per buffer (although I didn't
>  > check whether these objects are reachable only from this dead buffer).
>
> Could you elaborate?

Elaborate what? I tried to count calls to mark_object needed to mark the buffer
and all data referenced from it, and it may be from 60-70 to a few thousands or
even more; some of these calls are no-ops since corresponding objects may be
marked already or doesn't need marking at all.

> Marker objects belong to the buffer.

Not all - make_window attaches two new markers to W->start and W->pointm; when W
is set up to display the buffer B (set_window_buffer or maybe somewhere else),
these markers are changed from 'point to nowhere' to 'point into B'.

> The start and pointm fields are
> regenerated from the saved window structure.

Their member values are regenerated assuming that W->start and W->pointm itself
are valid marker objects:

Fset_window_configuration:
...
else if (!NILP (BVAR (XBUFFER (p->buffer), name)))
   /* If saved buffer is alive, install it.  */
   {
     wset_buffer (w, p->buffer);
     w->start_at_line_beg = !NILP (p->start_at_line_beg);
     set_marker_restricted (w->start, p->start, w->buffer);  /* here */
     set_marker_restricted (w->pointm, p->pointm,            /* and here */
                            w->buffer);
     Fset_marker (BVAR (XBUFFER (w->buffer), mark),
                  p->mark, w->buffer);

This will throw an error if start/pointm aren't markers, or crash
if they're invalid or uninitialized; if we can't trust the values
stored in W's fields, we should use Fcopy_marker instead.

Dmitry