On Fri, 26 Jan 2007, Jack L wrote:
I've also read about Ferret and Lucy recently. Quite interesting
and promising.
Yes, it seems very promising indeed.
I suppose a Python wapper around Ferret's C core or Lucy will provide
better performance than a wrapper around GCJ version because GCJ seems a
few times slower than Sun JVM in the benchmarks I read.
Last I compared Java Lucene and PyLucene, PyLucene seemed twice as fast.
This was on my old Powerbook G4 running Apple's VM and gcj 3.4.
Years have gone by since... Still, at the time, PyLucene was also twice as
slow as CLucene, the native C++ port.
In general, java benchmarks are fraught with little details that can affect
their results noticeably, so take my results with a grain of salt as well.
I wonder if there is any talks about PyLucene going in that direction?
I'd love to take it into that direction but Lucy is not there yet and I don't
know when it will be. I'm following its developments. In the meantime, gcj, is
getting better. gcj 4.3, currently under development, just got a new java
front end giving it Java 1.5 support and hopefully less bugs.
I could have wrapped CLucene (written in C++) but it comes with its own set of
bugs and is behind the Java Lucene development curve because it is an actual
port. Instead I chose gcj bugs and using the same codebase as Java Lucene svn
trunk.
I'm afraid that going the Lucy route is the same tradeoff: a port, diverging
from Java Lucene, with its own set of bugs, always behind Lucene core
development.
Lucy, however, is written in C and could gain significant momentum by serving
as a base for all other such wrapper ports (Python, Ruby, Perl, Lisp, etc...)
Time should tell...
Andi..
_______________________________________________
pylucene-dev mailing list
[email protected]
http://lists.osafoundation.org/mailman/listinfo/pylucene-dev
