I’ve had my first taste of JRuby this week-end, I wanted to see if I could get Raven running on it and how easy (or hard) that would be. My first impression was very positive. The distribution is definitely nicely built, including all the necessary script to run a jirb (the JRuby equivalent of irb) session or run RubyGems install in no time. Installing Rake and Raven was a 5mn deal. I’ve had a little surprise the first time I ran Rake though, if you run it from the bin directory, you’ll end up in an infinite loop (load ‘rake’ will load the rake script instead of the gem, which results in the script calling itself).
Then I got into trying to have Raven running on JRuby and that required far more time. Here are all the problems I ran into, successively:
- The Net::HTTP constructor will complain if you give a string for the port instead of an int (matz interpreter doesn’t care either way).
- The Kernel.Integer() method normally throws an error when what you’re passing isn’t an integer, JRuby returns 0.
- If you do a raise GemTooManyInstallError.new(artifacts), “Too many artifacts”, what will be effectively passed to the exception constructor by JRuby is the “Too many artifacts” string, and not the artifacts object. Weird bug.
- In matz Ruby, doing a compare like [1,6,1] <=> [188.8.131.52] will return 1. JRuby always considers the size first and will return -1 (who said that size didn’t matter). This one actually took me quite some time to spot as it was deeply hidden.
- The operator == isn’t redefined on Process::Status so you can’t do $? == 0 to test that a process executed successfully. You’ll have to do $?.exitstatus == 0.
- The method ctime doesn’t exist on File::Stat, that’s a bit annoying as I couldn’t find any replacement.
Altogether it’s only small problems and details but the real problem is more the accumulation. Raven’s codebase isn’t large at all and that’s a lot of fixes or even hacks to do. That being said, I’ve been able to run Raven after a few hours of work, which is still pretty good. Beside I didn’t expect to have Raven running right away, I can’t even imagine what a nightmare it must be to implement the whole (and large) Ruby API in every single details!
What is a bit more annoying is the startup time. And that I can;t change as it has not much to do with JRuby but much more with the Java VM. Pure execution is slightly (I’d say about 30 to 40%) slower but still reasonable. However a build system is used for many short running sessions and I’m afraid the quite long startup will impact Raven users pretty badly. Especially when you compare it to the immediate response of the matz interpreter.
I can’t wait for a fork of the now GPL’ed Java VM to happen to make it more suitable for that kind of application!