This is the mail archive of the
mailing list for the Java project.
Re: Allocate Class instancs on heap; not conservatively scan DSOs
Daniel Jacobowitz writes:
> On Wed, Apr 26, 2006 at 09:43:08AM +0100, Andrew Haley wrote:
> > This is a small (!) part of a more ambitious set of changes I've been
> > developing on a branch. It moves all static data out of BC-compiled
> > DSOs onto the heap, so we no longer need to scan static data
> > conservatively. It's been tested extensively with JOnAS and Eclipse;
> > with Eclipse, the time spent garbage collecting is halved.
> > It's still not completely done & dusted, but it's good enough to be
> > merged onto trunk, to be improved there.
> You've got discretion on this as a front end maintainer; but we are in
> Stage 3... is there enough time to get this done & dusted before we
> branch for 4.2.0?
I certainly hope so.
gcj is a less mature project than C and C++ and our development
process is a very bad fit with the gcc release schedules. Because of
that (and the fact that gcj bugs are not release-critical) the rules
regarding Stage 3 are somewhat more relaxed. We discussed this at the
GCC Summit a few years back, and the SC agreed to give gcj maintainers
leeway to make our own decisions in this area.
If this patch doesn't turn out to be sufficiently stable for a real
release it can be disabled with a one-line change.