Re: Reachable killed buffers

[Dmitry Antipov]

On 09/13/2012 09:11 PM, Paul Eggert wrote:
> Right, but as I understand it in the meantime it would still exist as
> a dead buffer, and would continue to pressure the GC.  And it might
> inadvertently survive the next sweep too, no?  The goal is to reclaim
> all the dead buffers, not just some of them.

No. If finding dead buffer pointer requires the long walk through a lot
of large objects, it's not worth doing; the goal is to find dead buffer
pointers in some "special slots" where such a pointers tends to concentrate
(like prev_buffers and next_buffers of window objects) and to hope that
the most of dead buffers are reachable just once and so can be reclaimed
if we do not reference them from such a "special slots".

Dmitry