Hi Wilhelm Fitzpatrick-


> I also can't agree with every criticism in the referenced blog
> post. Generating bytecode to call reflected methods actually
> sounds pretty cool to me, since now suddenly that bytecode is
> subject to Hotspot optomizing, inlining, etc.

Generating bytecodes on the fly for reflectively invoked methods is
utter coolness. But that may not be what's happening. Here's James
Robertson's suspicions:

chewing gum and bailing wire
http://www.cincomsmalltalk.com/blog/blogView?entry=3282207050

It's weird. There used to be all these benchmarks and microbenchmarks
for all these JVMs. The performance of these JVMs just kept
ratcheting up. But no more. At least that's my impression. I love
the that there's a dozen or so JVM implementations. Competition is
good for us. Could be good for Sun, too, if they embraced it. And
gauging from the continuing supply of research papers, there's still
plenty left to do.


Cheers, Jason Aaron Osgood / Seattle WA