need for speed, java vs c++ and the stack

[pepe]

The only real difference is that escape analysis knows ahead of time that an object is essentially divorced from the heap, while the Eden space can only find out at runtime.

Yes, for long run processes (raytracing, radiosity and such things ), it might be more interesting to use escape analysis, but for games GGC makes a perfect job at the moment.
Problem i have with escape analysis is its use with runtime generated code. Wondering how it will perform…

[quote]Me, I’m very happy with the GCs that exist today. And I’m sure things will get even better in the future!
[/quote]
Same for me. GGC turned out to give quite impressive results that i would not have imagined.

[princec]

Alien Flux put escape analysis to good use when it was Jet compiled. It never skips a single frame. No amount of tuning the GC quite manages the same effect, and even if it did, it still costs a quite a bit more to run than simply not having to collect ing the first place. And having said that, I just don’t care any more and new() with wild abandon in Super Elvis and pauses be damned, they’re hardly noticeable anyway. EA is really only proper useful in soft-realtime environments (like the TV graphics stuff I do).

As for that separate VM thing - Java has always been eminently capable and and perfectly designed to do it. The JVM is at its core, after all, just a virtual processor. Classloaders already do all that separation stuff. The problem is separating the GC and threading semantics effectively and sensibly. Not so easy. But not so impossible either.

Cas

[mthornton]

Classloaders don’t do all the separation that is required. Notable problems include native code (what happens if two isolated instances try to load/use the same native library), and a small number of static values in the core system (which is managed by the single system class loader and hence can’t have multiple copies).