Sunday, May 15, 2011

Rubinius on JRuby … ?

Of course, compiled Rubinius binary, *.rbc, file doesn’t work on JRuby. This post is about JRuby’s rubinius branch. I’m not sure how may people are aware that, but JRuby does have a rubinius branch. As far as I looked at that branch, it is not to be merged into master, at least, in near future. Maybe it is headius’ pet project at this moment, and is implemented as a JRuby extension. Yes, it is a JRuby extension. The branch attempts “extending JRuby,” like I wrote about in my blog post (http://yokolet.blogspot.com/2011/05/extending-jruby.html). Since this extension is still small, it would be a good practice to figure out how extension works. So, I’m going to write how you can decipher existing JRuby extension. Hopefully, this will help you to write your own extension.


To try the rubinius branch, you have two preparations to be done. The first one is to build JRuby of the rubinius branch. It is easy. Just clone out JRuby, checkout rubinius branch, and run ant command as in below:

git clone git://github.com/jruby/jruby.git
cd jruby
git checkout rubinius

ant clean-all
ant

Next, you need Rubinius source. You don’t need to build rubinius just to see how it works. But, the rubinius extension uses Rubinius’ kernal sources, which is included only in the source archive. That’s why you need the source. Rubinius source archive is available to download from http://rubini.us/. Get the archive and unzip it.


Up until now, you had rubinius branch JRuby and Rubinius source. Set the environment variables. Suppose the JRuby’s home directory is /Users/yoko/Projects/jruby, then, set JRUBY_HOME and PATH like in below. Adjust them to fit in to your system:

export JRUBY_HOME=/Users/yoko/Projects/jruby
PATH=$JRUBY_HOME/bin:$JRUBY

Suppose, Rubinius source are in /Users/yoko/Projects/rubinius-1.2.2, then set RBX_KERNEL as in below:

export RBX_KERNEL=/Users/yoko/Projects/rubinius-1.2.2/kernel



Everything should be ready. Let’s try this out.

bash-3.2 rubinius$ jruby -S irb
irb(main):001:0> require 'rubinius'
=> true

Yay! Rubinus was successfully loaded on JRuby. What’s next? Look at the RubiniusLibrary.java (https://github.com/jruby/jruby/blob/rubinius/src/org/jruby/ext/rubinius/RubiniusLibrary.java). At the line 51, “Rubinius” module is defined.

RubyModule rubinius = runtime.getOrCreateModule("Rubinius");

So, there should be a constant, Rubinius.

irb(main):002:0> Rubinius
=> Rubinius
irb(main):003:0> Rubinius.class
=> Module

So far, so good. Then, at the line 56, you can see

RubyTuple.createTupleClass(runtime);

This means you should go to RubyTuple.java (https://github.com/jruby/jruby/blob/rubinius/src/org/jruby/ext/rubinius/RubyTuple.java). On the line 55-62 of RubyTuple.java, createTupleClass method is defined. In this method, "Tuple" class is defined under the "Rubinius" module. Then, annotated methods, which have Java annotation @JRubyMethod, are defined. Looking at the rest of the code in RubyTuple.java, you can see three annotated methods (new, [], and []=), and one override method (dup) are there. Let’s try these.

irb(main):013:0> Rubinius::Tuple
=> Rubinius::Tuple
irb(main):014:0> tuple = Rubinius::Tuple.new 3
=> #<Rubinius::Tuple:0x3df89785>

This constructor needs one argument because rbNew method is defined as:

public static IRubyObject rbNew(ThreadContext context, IRubyObject tupleCls, IRubyObject cnt) {

Here's the rule. First two arguments are given internally, and the rest of the arguments are given from users. So, I typed “3” as an argument. Let’s keep going on.

irb(main):018:0> tuple[0]
=> nil
irb(main):019:0> tuple[0]=123
=> 123
irb(main):020:0> tuple[0]
=> 123
irb(main):021:0> tuple_dup = tuple.dup
=> #
irb(main):022:0> tuple_dup[0]
=> 123

All right, methods worked.


Next, get back to RubiniusLibrary.java and let’s look at the lines 84-88.

runtime.getObject().deleteConstant("Hash");
runtime.getLoadService().lockAndRequire(rbxHome + "/common/hash.rb");
RubyClass hash = (RubyClass)runtime.getClass("Hash");
hash.defineAnnotatedMethods(RubiniusHash.class);
runtime.setHash(hash);

Soooo interesting! Could you figure out what’s going on here? Hash is redefined using Rubinius code! JRuby’s Hash is entirely written in Java (https://github.com/jruby/jruby/blob/rubinius/src/org/jruby/RubyHash.java). But, once “require rubinius” is done, the Hash is totally replaced by Rubinius’ Hash, which IS written in Ruby. See? JRuby can be extended also in Ruby, not just Java. To do double check this, let’s add one line in initialize method of kernel/common/hash.rb:

def initialize(key, key_hash, value)
@key = key
@key_hash = key_hash
@value = value
@next = nil
puts "Rubinius Hash!!" # this line is added
end

Then, restart irb and re-request rubinius.

bash-3.2 rubinius$ jruby -S irb
irb(main):001:0> require 'rubinius'
=> true
irb(main):002:0> h = Hash.new
Rubinius Hash!!
=> {}

Yay, Hash is really Rubinius’ Hash. What an idea!


As I wrote, you have a lot of options for “extending JRuby.” You can extend using mature Java APIs and, also, cutting edge Ruby code. Why don’t you try this fantastic extension? It should be fun.

1 comment:

headius said...

Yes, the Rubinius branch is still a pet project, but it shows that JRuby is quite capable of being implemented in Ruby as well as Java. I often encourage users who want to extend JRuby to first explore doing so in Ruby, and this frequently gives them what they need. It's not true that JRuby can only be implemented or extended in Java...it just has mostly been so up to now.

The Rubinius branch serves two purposes: to see what would be necessary to implement more of JRuby in Ruby; and to push JRuby to run Ruby code better and faster. Some months ago I benchmarked the Rubinius Hash impl running in JRuby, and it performed very well...only about 2x slower than running in Rubinius, a VM optimized specifically for Ruby (it's important to note here that JRuby's Java-based Hash is still much faster). Since then, we've started to incorporate Java 7's invokedynamic, which so far has shown great promise for increasing JRuby's Ruby performance nearly to Java speeds...so implementing more of JRuby in Ruby starts to seem feasible.

I'd love for others to play with the Rubinius branch and try to incorporate other Rubinius core classes into JRuby. It's fun, it helps stretch JRuby, and it shows how easy it is to extend JRuby using not just Java code but Ruby code as well.

Enjoy!