Recent Java VM often needs to interpret scrypts written in dynamically typed languages. The common langauges used include JavaScript, whose reference implementation is available with JSR 223. Performantly executing this kind of interpretation is not so much simple. Naive interpretation is known to be 10-100 times slow.
Compiling or translating programs in foreign languages into the language directly supported by the platform, which is Java bytecode for Java VM in this case, is typical approaches to good performance. Instead of interpreting method calls in JavaScript, we can translate/compile them to native Java method calls and execute them directly on Java VM.
The compiling/translating approach involves, however, plenty of code generation in some cases. Too much generated code consumes too much memory and thus can result in poor performance.
Gilad Bracha, the fourth man of Java, has been pointing out this problem in his blog, and finally proposed a new Java VM instruction, specialized to address this issue on Feb 28, 2006.
It is in the JSR review ballot stage, and to be approved in Mar 13, 2006.
I would like to leave my memos related to my "kung-fu" for some technology insights. Kung-fu doesn't only mean a battle style, but means hacking one's own life.
Subscribe to:
Post Comments (Atom)
-
Array's pointers are reset when the original array is logically copied and thus has 2+ reference counts internally. This doesn't h...
-
The Rational Unified Process unifies the entire software development team and optimizes the productivity of every team member by putting the...
-
Great demonstrations of powerful combination of recent deep neural networks applications - stacked hourglass networks for human pose estima...
2 comments:
It is acutally approved unanimously.
http://blogs.sun.com/roller/page/gbracha?entry=jsr292_and_hotswapping
Post a Comment