[relax-ng] Re: Bali - RELAX NG validatelet compiler

[Kohsuke Kawaguchi <kohsuke.kawaguchi@sun.com>]

> This sounds like a great addition.  Have you done any measurements at all?

This is a part of my master's degree project, and the professor wants me
to do a performance comparison, so I'm planning to do it. But not yet. I
barely managed to pass the test suite...


> - Is it faster for some kinds of grammar/instance? If so, which?

One area that I'm hoping to see some performance gain is the name
comparison. Jing does a lot of comparison of strings between tag names
and name classes. Bali encodes a name class hit test into a simple bit
test operation.

Another potential performance gain I'm looking for is a better
life-cycle control over short-lived objects. The way the runtime works
makes it possible to detect exactly when an object becomes unnecessary.
So I thought maybe I could use this to reduce the object allocation /
garbage collection overhead. But this is not tried yet.



The table generation algorithm is essentially the same as Murata-san's.
I used regexp derivation to reduce the size of the table, and threw in a
couple of optimizations. And probably the way I handle interleave is a
bit different. But nothing fundamentally new.



regards,
--
Kohsuke KAWAGUCHI
Sun Microsystems                   kohsuke.kawaguchi@sun.com