On Fri, 20 Apr 2007, Patrick Mayweg wrote:
> You can do the optimizations, as long as you keep in mind to do them
> only for the methods which return directly. In other JNI projects I did,
> I learned that the hard way. Testing on small test set always work fine,
> but when tried I on the real data, everything broke.
Given the layered design of JavaHL's implementation, I would prefer to
leave the existing explicit reference handling in place to keep things
more obvious for maintainers.
> A good idea would be to use -verbose:jni -Xcheck:jni as JVM switches
> during testing. with them java tells you a lot about everything you
> might do wrong.
This seems like a good idea regardless. I assume that the first
argument is supposed to be supported by all JVMs, while the second may
not be? If the arguments are supported universally, we'll have to put
in some autoconf macros to do this only with Sun's JVM.
Received on Thu Apr 26 22:58:48 2007
- application/pgp-signature attachment: stored