On 02/25/10 07:40, Wanja Gayk wrote: > Am 24.02.2010, 00:22 Uhr, schrieb Lawrence D'Oliveiro > <l...@geek-central.gen.new_zealand>: > >>> Java - The JVM code been hacked to death by Sun engineers (optimised) >>> Python - The PVM code has seen speed-ups in Unladen or via Pyrex.. >>> ad-infinitum but nowhere as near to JVM >> >> Python is still faster, though. > > In synthetic microbenchmarks without any practical importance - possibly.. > >> I think a key reason is that its VM supports >> reference-counting, which the Java folks never quite got the grasp of. > > Reference counting is about the worst technique for garbage collection.
I disagree IMO, assuming there is no circular reference, reference counting is the simple, lightweight, and best garbage collection technique. Reference counting can free memory (or return it to cache) right after the refcount turns zero; other traditional garbage management techniques runs periodically and there's a delay between getting to zero reference to actual memory release. The other weakness of traditional garbage managers is the symptomatic "stopping the time" whenever the GC is running, which makes it unsuitable for time-sensitive program. The only weakness of reference counting is its inability to detect circular reference, which is why CPython have an additional garbage collector. > Modern Java VM won't count references. They will just trace the active > references from the rootand mark all objects it finds as active and save > these. The remaining ones are garbage. The reason why this is faster is > that there are usually less live objects than dead ones and there are > less refereces to look at. Reference counting is even faster since there is only one object to be looked at: the object that is being dereferenced. There is no need to look at other live objects, there is no need to look other dead objects, just the current object. -- http://mail.python.org/mailman/listinfo/python-list
