Re: Reachable killed buffers

[Stefan Monnier]

>> Arguably, if (part of) that list is shared with some other data
>> structure, then we indeed shouldn't remove dead buffers from it.
> But the patch doesn't implement that either, right?
> If part of the list is shared, but the GC doesn't discover
> this until after that code runs, it'll remove dead
> buffers from that part.

Oops, you're right.  So, I think it's OK to remove the dead buffers
(like we do now), but indeed the proposed patch wouldn't be good because
it would sometimes remove them and sometimes not, depending on the order
in which the heap is traversed by the mark phase.


        Stefan