Andrew Haley wrote: > I don't think that's a reasonable excuse for a bad interface. It's not a bad interface, it's a bad implementation, that's a big difference. > If we had a better one, people could actually _use_ some of these > "reference" methods without having to rewrite them. That's complete rubbish. The only possible other "implementation" would be to mark the method native, that's hardly more useful. > I totally reject the notion that it doesn't matter how inefficient > the reference design is -- if it's an example, it should be a good one. > > Besides, it's not worth arguing. We need it for sun.misc.Unsafe > anyway. It is worth arguing, because it shows that you totally miss the point of the VM interface. The VM interface allows a VM to pick the most efficient and/or convenient place to hook in, in this case if you care about performance, you replace VMStackWalker. This whole notion that the Java/native interface is somehow always the best library/VM interface is just silly. The VM classes are intended to be replaced. Regards, Jeroen
