Re: [RFA] boehm-gc for AIX

[Tom Tromey <tromey at redhat dot com>]

>>>>> "Bryce" == Bryce McKinlay <bryce@waitaki.otago.ac.nz> writes:

Bryce> We could certainly add something to the spec to say that any
Bryce> global variables holding Java objects must be declared in
Bryce> Java. This would probibly make things easier for a theoretical
Bryce> non-conservative GC in future. I don't think I've seen any CNI
Bryce> code which does otherwise anyway.

There's some code in libgcj that does this.
At least prims.cc (see arg_vec, main_thread).
There might be more -- though if there is, there isn't much.

I think we're pretty unlikely to have a non-conservative GC.  My
understanding is that a fully precise GC would require a substantial
amount of gcc hacking.

However, I do think we don't need to make changing libgcj to use a
precise GC any harder than it already is.  Tightening the rules on
globals would be fine.  For instance we could add a CNI call to mark a
location as a GC root.

Tom