RJudy is a Ruby extension module that provides an interface to the Judy arrays library. Judy's home page is at www.sourcejudy.com.
Judy was originally implemented at
Hewlett-Packard, and was subsequently open-sourced in June 2002 (with an
LGPL license). Doug Baskins, the author of the Judy algorithm, has written a nice summary
of what makes Judy arrays so fast, and
why you might want to use them.
Build and Installation Instructions
Before you can build, install and use this extension module you'll need to
first obtain and install the Judy header
files and library. The source tarball for the latest version of Judy can be found at its SourceForge
project page (sourceforge.net/projects/judy).
If you get lucky, you may also be able to find precompiled binaries for
Once you've built and installed Judy
itself, you should be ready to build and install this extension module.
From the top-level directory, first configure the build by typing:
ruby install.rb config
Next, start the build process by typing:
ruby install.rb setup
Finally, install it by typing:
ruby install.rb install
Note that on Unix systems, you'll probably need to be logged in as
"root" for that last step.
Once you've built and installed RJudy, you can use any of its bundled
classes (Judy1, JudyL or JudySL) by simply require-ing the
"judy" feature, i.e.
judy1Array = Judy::Judy1.new
judyLArray = Judy::JudyL.new
judySLArray = Judy::JudySLArray.new
For a listing of the instance methods for each of these classes, refer to
their respective documentation pages. For more background information about
Judy arrays, refer to the Judy home page.
Test cases for each of the Judy classes
are provided in the "tests" subdirectory. These tests use the
Test::Unit framework, available for download from testunit.talbott.ws.
To run one of the individual tests, type something like:
To run all of the tests, type:
You shouldn't get any failures; if you do, please contact email@example.com and
let him know about it!
RJudy is Copyright (c) 2002 by Lyle Johnson, and is provided under the GNU
Lesser General Public License (www.gnu.org/licenses/lgpl.html).
See the LICENSE file in the top-level directory of the RJudy source
distribution for more details.
Changes Since Version 0.1
- The previous implementations of Judy1, JudyL, JudySL and JudyHash used the
calloc() function for memory allocation instead of Ruby's ALLOC(). As a
result, the normal Ruby garbage collection heuristics would not come into
play and no garbage collection would take place during insertions into a
JudyHash instance. This has been corrected; thanks to Mauricio Fernandez
for pointing out this flaw.
- The previous implementation of JudyHash did not make a duplicate of new
string keys inserted into the hash, it just stored a reference to the input
string key. This was inconsistent with Ruby's built-in Hash which does make
a duplicate of new string keys. This has been corrected; thanks to Mauricio
Fernandez for pointing out this flaw.
- The previous version of the words.rb example wasn't quite fair because it
failed to run the garbage collector between tests. As a result, the tests
run first will have an advantage over tests run later, when the garbage
collector may get called more frequently. I've rewritten this program to
instead use the benchmark library's bmbm method, which should run
the garbage collector in between tests and hopefully even up the playing
field. Thanks to Mauricio Fernandez for pointing out this flaw in the
- Per Doug Baskins' advice, replaced all calls to Judy library functions (such as JudyLGet()
and JudyLNext()) with calls to their macro equivalents (e.g. JLG() and
JLN()). This should improve overall performance, especially for hash