RE: JNI patches

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

Good point.

Another alternative on many platforms is to treat all writable regions of
memory (that aren't part of the garbage collected heap) as roots.  This
requires something like a /proc interface to find the writable mappings.
The code to do this under Linux and a few other systems is mostly there.
(This is basically how things work under win32 and Irix anyway.)

But I think we are all in agreement that this is mostly a solution in search
of a problem.  So long as most of the standard libraries use CNI, and JNI is
thus not a performance bottleneck, the more conventional the JNI
implementation, the better.

Hans

> -----Original Message-----
> From: Jeff Sturm [mailto:jsturm@one-point.com]
> Sent: Monday, April 02, 2001 4:53 PM
> To: java@gcc.gnu.org
> Subject: RE: JNI patches
> 
> 
> 
> 
> On Mon, 2 Apr 2001, Boehm, Hans wrote:
> > The collector normally does not scan objects allocated with 
> C malloc (except
> > on one or two platforms where that's hard to avoid).  The 
> easiest way to
> > change that is to redirect the C malloc to 
> GC_malloc_uncollectable, but that
> > sometimes creates startup ordering issues.  It should work 
> on Linux, but I'm
> > not sure about various other platforms.
> 
> It's not a general solution.  For one thing, with JNI 
> invocation programs 
> are free to startup (and bind to malloc, etc.) before loading the gcj
> runtime.
> 
> I don't see how local refs are avoidable if JNI is going to 
> work.  There
> are probably faster ways to handle them, though.  (I'm not 
> worried about
> global refs, which are relatively uncommon.)
> 
> Jeff
> 
>