Re: marking classes in walkObject

Date view Thread view Subject view Author view

From: Godmar Back (gback@cs.utah.edu)
Date: Fri Dec 18 1998 - 10:26:19 EST


 No problem, I'll fix it. I'll also replace MARK_IFNONZERO with MARK_OBJECT
with the understanding that MARK_OBJECT will not attempt to mark
null pointers.

Also, expect some more changes in mem/gc- --- we're still finishing up
Jason's stuff. Since it doesn't use mmap anymore, you can come up with
a new name, too.

Since the new scheme might have a need to realloc the block
table, we'll need to move the locking to gc-incremental.c because
we need to be careful the allocator doesn't hold any pointers in
the block table while it may be relocated. More later.

        - Godmar

>
> On Dec 18, 1998, Godmar Back <gback@cs.utah.edu> wrote:
>
> > Alexandre, given that objects can outlive the loaders of the class to
> > which they belong, how can you avoid marking the class when you walk the
> > object? An object may hold the last reference to a class that keeps it
> > alive.
>
> Oops, I had missed that. The (broken) reasoning was that the object
> would keep the class alive, and the class would keep the loader alive,
> and then the loader will keep the class alive, so the object doesn't
> have to mark it. Will you please fix that, or should I do that?
>
> Sorry about that :-(
>
> --
> Alexandre Oliva http://www.dcc.unicamp.br/~oliva aoliva@{acm.org}
> oliva@{dcc.unicamp.br,gnu.org,egcs.cygnus.com,samba.org}
> Universidade Estadual de Campinas, SP, Brasil
>


Date view Thread view Subject view Author view

This archive was generated by hypermail 2b29 : Sat Sep 23 2000 - 19:57:22 EDT