RE: PR 10920 GC reports "too many roots"

["Boehm, Hans" <hans_boehm at hp dot com>]

I haven't thought about it enough.  I don't think it's trivial, since a lot of
the root sets would need to be added and perhaps allocated during a GC.  GC_scratch_alloc might
still work, but there's some danger that you would make it much less likely to
be able to recover from an out-of-memory condition, since it would then tend to
show up as a failure to allocate the root sets during a GC.  Perhaps we could
overallocate during startup to make it unlikely that the table would need to
expand later on.

I think there really shouldn't be a problem unless/until we end up with a separate
root set per class.  I'll think about better ways to handle that.

Hans

> -----Original Message-----
> From: Jeff Sturm [mailto:jsturm@one-point.com]
> Sent: Monday, November 24, 2003 12:27 PM
> To: Boehm, Hans
> Cc: 'Ãyvind Harboe'; java@gcc.gnu.org
> Subject: RE: PR 10920 GC reports "too many roots"
> 
> 
> On Mon, 24 Nov 2003, Boehm, Hans wrote:
> > The table size is MAX_ROOT_SETS.  GC_dump() should print 
> what's currently
> > in the table.  It would be good to understand why it's 
> overflowing.  But simply
> > increasing MAX_ROOT_SETS is likely to improve matters.
> 
> Hans, is there any good reason this table is not yet sized 
> dynamically?
> 
> Jeff
>