cni-changes documented?

[Andrew Haley <aph at redhat dot com>]

Jakob Praher writes:
 > 
 > from lurking on the thread "Java-Gnome: jni or cni", Andrew Haley
 > introduced a redesigned cni, which would allow other vms to support it.
 > (Maybe I missed something here)
 > 
 > Are these changes documented somewhere?

We have no intention of making CNI truly multi-VM compatible.  That
isn't a design goal, and would present some severe technical
difficulties.

What we must do, however, is switch from using pointers to using smart
pointers: this is a matter of declaration rather than use.  

i.e.

   ::java::lang:foo *bar

becomes

   ref<::java::lang:foo> bar

that's all.

 > The cni.sgml is rather old (at least my cvs tells me this).
 > 
 > Andrew: In your mail titled "Binary Compatibilty ABI work" you mention
 > that you have some patches for compiling whole jars to shared objects,
 > thus enabling optimization on a jar level. What is the status on this?

It already works.

However, there are compatibility issues I'm working on.  My latest
plan is to make a CVS branch in which to commit the current state.
This is somewhat experimental, so it isn't suitable for mainline.

Andrew.