Oniguruma Java port speeds up JRuby
This is a glorious day! Joni (Marcin's incredible Java port of the Oniguruma regexp engine) has been merged to JRuby trunk. It seems to work really well right now.JRuby team member Marcin Mielczynski took the job of porting the Oniguruma Regex engine to Java code - Oniguruma is the Regex engine included in Ruby 1.9.x.
This might just be the last installment in the (seemingly) never ending story about JRuby and Regular Expression (Regex) engines. Early JRuby versions used Java's built-in Regex library (included since Java 1.4) to implement Ruby's Regexes. While this was the simplest solution, not requiring any 3rd party libraries or ports, it also brought problems that made it unsuitable for JRuby. Since JRuby aims to be a compatible implementation of Ruby 1.8.x (or future versions), it's necessary to support the same Regexes. Java's implementation turned out to be incompatible, partially because of algorithm details that caused it to fail for some of the expressions. Ola explains the steps that followed:
To fix that, we integrated JRegex instead. That's the engine 1.0 was released with and is still the engine in use. It works fairly well, and is fast for a Java engine. But not fast enough. In particular, there is no support for searching for exact strings and failing fast, and the engine requires us to transform our byte-strings to char or String. Not exactly optimal. Another problem is that compatibility with MRI suffers, especially in the multi byte support.All these problems seem to be - or will be - solved with Joni. Regex performance has been a big problem in the past (e.g. see Lessons from building Oracle Mix on JRuby on Rails), but Joni seems to help with that too. Charles Nutter looked at REXML performance with the new code:
After running through a series of basic optimizations, most of the key expressions we worried about were performing as well as or much better than JRegex, so Ola went through with the conversion over the past couple days. Marcin is continuing to work on various optimizations, but both Ola and I have been playing with the new code. And it's looking great.The linked article continues with the benchmark results comparing the code before and after the merge, which shows significant speed ups with the Joni code.
These issues also show a problem shared by many alternative Ruby implementations. Rubinius, a Ruby implementation written in (mostly) Ruby, uses the simple solution of including Oniguruma. Ruby implementations based on VMs such as the JVM or .NET, however, have the problem that including a native library makes deployment more difficult (they'd need to ship platform specific versions). Not just that, as Marcin explains in a comment on Ola's blog, there are other integration issues:
We've been thinking about [including Oniguruma] already. There are few reasons:
Threading: Oniguruma uses global locks when initializing code range tables or managing shared AST nodes (like Character Class hashtable). Oniguruma bytecode interpreter also uses thread locks (it can be turned off but we get it for free in java land, and it'd be a hack to mix foreign threading with java one).
Exceptions: it would be hard to recover from segfaults. Converting Oniguruma errors to Ruby exceptions would also be an ugly hack.
JNI: it requires data separation, so all strings/bytes would have to be copied.
Additional binary distribution: good luck compiling it one Mainframe :D