From mboxrd@z Thu Jan 1 00:00:00 1970 From: Tom Tromey To: "Boehm, Hans" Cc: "'Jeff Sturm'" , java@gcc.gnu.org, gcc@gcc.gnu.org Subject: Re: dynamic library cost (was RE: libtool, java woes) Date: Thu, 12 Apr 2001 22:17:00 -0000 Message-id: <87r8yxv06u.fsf@creche.redhat.com> References: <140D21516EC2D3119EE7009027876644049B5D5C@hplex1.hpl.hp.com> X-SW-Source: 2001-04/msg00588.html >>>>> "Hans" == Boehm, Hans writes: Hans> (As far as I know, the only way to currently get a write barrier Hans> is with virtual memory techniques. That means generational or Hans> incremental collection requires virtual memory, and you only get Hans> update information at coarse granularity. And you're presumably Hans> limited to a conservative stack scan.) That is definitely the current state of affairs. Adding write barrier support doesn't appear to be hugely difficult. Jon Olson implemented it (I haven't seen the code). Is it worth doing this with your collector? Adding non-conservative stack scanning is much harder. It has been added to gcc before (in gcc 1.something -- ancient history). It is definitely possible, but it touches a lot of the compiler. There's a paper on this by Rick Hudson. When I say "much harder", my guess (and this is really a wild guess) is that it is on the order of a man year. Tom