Re: CNI changes (Was: Binary Compatibility)

[Andrew Haley <aph at redhat dot com>]

Tom Tromey writes:
 > >>>>> "Anthony" == Anthony Green <green@redhat.com> writes:
 > 
 > Anthony> For instance, conforming to the new C++ ABI (padding-wise),
 > Anthony> and wasn't there talk at some point of requiring explicit GC
 > Anthony> markers for data?
 > 
 > Requiring root registration is on my wish-list.

Ah, yes.  This would be another CNI issue, I suppose.  

I guess we can use C++ static initializers to register static roots
automatically?  So, it would just be

static_ref<java::lang::Object> foo;

 > Ideally in 3.5 we could come out with an ABI we could support well
 > into the future.  I'm not sure whether that is completely possible,
 > but we can try.

My best guess is that 3.5 will be the draft gcj ABI, with 

 > 
 > There's a minor ABI change we need to make in JNI; see PR 4066.

"The stubs generated by `gcj -fjni' are not thread-safe.
I believe the code they generate to cache the JNI method
pointer can fail in some rare situations.
(This code relies on double-checked locking, which is
known to be bad.)"

What is wrong with the current scheme?

Andrew.