JRuby: Pushing the Java Platform Further
7 likes1,532 views
The document discusses how JRuby pushes the Java platform further by implementing custom core classes like Array, Hash, String, and IO to match Ruby's behavior exactly. It also describes how JRuby uses libraries like ByteList, Joni, Java Native Runtime, and FFI to provide Ruby-like regular expressions, native I/O, and OS-level features on the JVM. These custom implementations and libraries allow JRuby to overcome challenges like dynamic typing and provide a full-fledged Ruby environment atop the Java Virtual Machine.
![Why Custom?
● Array, Hash
○ Behavior must match Ruby exactly
○ Many operations must redispatch into Ruby
● String
○ Java's String is immutable
○ Java's String is UTF-16 char[] based
● Regexp
○ byte[] based String necessitates it
○ Ruby Regexp is rich in features
● IO, File
○ Blocking IO simulated atop non-blocking
○ No good abstractions over NIO](https://image.slidesharecdn.com/pushingthejavaplatformfarther-110725134935-phpapp01/85/JRuby-Pushing-the-Java-Platform-Further-7-320.jpg)
![ByteList
● A list of bytes
○ Abstraction around byte[] + encoding
○ Similar operations to StringBuffer
○ CoW-capable](https://image.slidesharecdn.com/pushingthejavaplatformfarther-110725134935-phpapp01/85/JRuby-Pushing-the-Java-Platform-Further-11-320.jpg)
![Regexp
● Operates against byte[] in String
○ All other engines operate against char[]
○ Transcoding cost == death
● Supports arbitrary encoding
○ Not all encodings transcode losslessly
○ Mismatched but compatible encodings?
● Bytecode engine
○ Avoids StackOverflow bugs in java.util.regex
○ Optimization potential
● Advanced regex features
○ Encoding-agnostic
○ Named groups](https://image.slidesharecdn.com/pushingthejavaplatformfarther-110725134935-phpapp01/85/JRuby-Pushing-the-Java-Platform-Further-14-320.jpg)
![Joni
● Java port of Oniguruma
● Ruby-friendly
○ byte[] based
○ Arbitrary encodings
○ Bytecoded engine
● Fast
○ Avoids transcoding costs
○ Sometimes faster than j.u.regex
Thanks to Marcin Mielzynski (github.com/lopex)!](https://image.slidesharecdn.com/pushingthejavaplatformfarther-110725134935-phpapp01/85/JRuby-Pushing-the-Java-Platform-Further-16-320.jpg)
![Links and Q/A
(All on Github)
http://github/
● jruby - JRuby organization
○ /jruby - JRuby itself
○ /bytelist - byte[] container
○ /jcodings - byte[] encoding support
○ /joni - byte[] regex
● /jnr - Java Runtime
○ /jnr-ffi - foreign function interface
○ /jnr-posix - POSIX features
○ /jnr-enxio - native IO
○ /jnr-unixsocket - UNIX sockets
○ /jnr-x86asm - x86 assembly
● /ffi/ffi - Ruby FFI](https://image.slidesharecdn.com/pushingthejavaplatformfarther-110725134935-phpapp01/85/JRuby-Pushing-the-Java-Platform-Further-41-320.jpg)
JRuby: Pushing the Java Platform Further
- 1. Pushing the Java Platform Further
- 2. Me Charles Oliver Nutter ● headius@headius.com ● @headius ● JRuby co-lead, fulltime since 2006 ● JVM enthusiast ● http://blog.headius.com
- 3. JRuby ● Ruby atop the JVM ○ "Just another Ruby implementation" ○ "Just another JVM language" ● A challenge for the JVM ○ Dynamic typing/dispatch ○ Own set of core classes ○ Heavy use of OS-level features
- 4. JRuby ● Ruby atop the JVM ○ "Just another Ruby implementation" ○ "Just another JVM language" ● A challenge for the JVM ○ Dynamic typing/dispatch ○ Own set of core classes ○ Heavy use of OS-level features
- 6. What Classes? ● Array: a mutable, growable list of objects ● Hash: an ordered associative collection ● String: a collection of bytes (w/ an encoding in 1.9) ● Symbol: a named identifier ● IO: a stream of bytes (w/ an encoding in 1.9) ● File: a stream of bytes from filesystem ● Range: a range of values
- 7. Why Custom? ● Array, Hash ○ Behavior must match Ruby exactly ○ Many operations must redispatch into Ruby ● String ○ Java's String is immutable ○ Java's String is UTF-16 char[] based ● Regexp ○ byte[] based String necessitates it ○ Ruby Regexp is rich in features ● IO, File ○ Blocking IO simulated atop non-blocking ○ No good abstractions over NIO
- 8. String
- 9. String ● Used for both binary and character data ○ "It's just bytes" ○ IO operations receive, return String ● Arbitrary encoding ○ Implicit encoding in 1.8 (set globally) ○ Explicit encoding in 1.9 (String aggregates Encoding) ● Mutable ○ More efficient to mutate in-place ○ Copy-on-write to reduce object churn
- 11. ByteList ● A list of bytes ○ Abstraction around byte[] + encoding ○ Similar operations to StringBuffer ○ CoW-capable
- 12. ByteList Demo
- 13. Regexp
- 14. Regexp ● Operates against byte[] in String ○ All other engines operate against char[] ○ Transcoding cost == death ● Supports arbitrary encoding ○ Not all encodings transcode losslessly ○ Mismatched but compatible encodings? ● Bytecode engine ○ Avoids StackOverflow bugs in java.util.regex ○ Optimization potential ● Advanced regex features ○ Encoding-agnostic ○ Named groups
- 16. Joni ● Java port of Oniguruma ● Ruby-friendly ○ byte[] based ○ Arbitrary encodings ○ Bytecoded engine ● Fast ○ Avoids transcoding costs ○ Sometimes faster than j.u.regex Thanks to Marcin Mielzynski (github.com/lopex)!
- 17. Joni Demo
- 19. OS-level Features ● Process management ○ getpid, waitpid, kill, signal, ... ● Filesystem operations ○ symlink, ownership, permissions, ... ● User-friendly IO ○ Interruptible, inherit TTY, "select" function, ... ● Binding native libs ○ Avoid hand-written native shim (JNI) ○ Route around JVM APIs ● UNIX sockets
- 20. Java Native Runtime http://github.com/jnr
- 21. Java Native Runtime A collection of libraries to make interfacing with OS-level features easier. ● Easy binding of native libs ● NIO-like IO atop file descriptors ● UNIX sockets ● POSIX filesystem operations Big thanks to Wayne Meissner (@wmeissner)!
- 22. JNR-FFI (foreign function interface) http://github.com/jnr/jnr-ffi
- 23. JNR-FFI ● Base of the JNR libraries ● Easy binding of native functions ● Struct mapping ● Memory/pointer management ● As fast as JNI? ○ Within 10-20% (comparing invocation overhead) ○ "a viable alternative, even for reasonably performance-sensitive libraries" - @wmeissner ○ "A hell of a lot nicer than using JNI!" - @headius
- 24. JNR-FFI Demo
- 26. JNR-POSIX ● Commonly used POSIX functions not in JDK ○ Symlinks ○ Filesystem metadata ○ Ownership, permissions ○ True "exec" ○ Process management
- 27. JNR-POSIX Demo
- 28. JNR-ENXIO (Extended Native X-platform IO) http://github.com/jnr/jnr-enxio
- 29. JNR-ENXIO ● Mimics NIO ○ Same or similar API ○ Channels, Selectors, etc ● Implemented atop JNR-FFI ○ 100% Java (other than JNR-FFI) ● Multiple backends ○ poll, kqueue currently ○ epoll, win32 needed...volunteers? ● Full set of operations for *any* IO type ○ NIO can't select on stdio, files
- 30. JNR-ENXIO Demo
- 32. JNR-UNIXSOCKET ● Built atop jnr-enxio ● UNIX sockets ○ Need I say more?
- 33. JNR-UNIXSOCKET
- 34. JNR-X86ASM (yes, it's what you think) http://github.com/jnr/jnr-x86asm
- 35. JNR-X86ASM ● API for x86/x86_64 ASM generation ● Huh? ○ Down to the metal, baybee ○ Faster bindings to native functions ○ Other weird stuff...
- 36. JNR-X86ASM Demo
- 38. Ruby FFI ● Ruby DSL for binding native libraries ● Cross-impl ○ Supported on JRuby and standard Ruby ● Fun!
- 39. Ruby FFI Demo
- 40. What Have We Learned? ● The JVM can be a great native platform ● No problem is impossible to solve ;-) ● JRuby gives you all this (and more) for free
- 41. Links and Q/A (All on Github) http://github/ ● jruby - JRuby organization ○ /jruby - JRuby itself ○ /bytelist - byte[] container ○ /jcodings - byte[] encoding support ○ /joni - byte[] regex ● /jnr - Java Runtime ○ /jnr-ffi - foreign function interface ○ /jnr-posix - POSIX features ○ /jnr-enxio - native IO ○ /jnr-unixsocket - UNIX sockets ○ /jnr-x86asm - x86 assembly ● /ffi/ffi - Ruby FFI