问题
I have created a compiler in C (using lex & bison) for a dynamic typed programming language that supports loops, functions declarations inside functions, recursive calls etc. I also created a virtual machine that runs the intermediate code created by the compiler.
I was now thinking instead of compiling to my own intermediate code, compile it to java byte code.
I saw that the question about creating a JVM language has already been asked but I don't find the answer very informative.
So here are my questions:
- I guess to create a language for JVM a must is to read the JVM specification book, what other books can you suggest (except Dragon Book of course)? I'm mostly concerned about books or tutorials on how to create a JVM language, not a compiler in general.
- There are many Java libraries to read, write and change .class files like jclasslib, bcel, gnu bytecode, etc. Which one would you suggest? Also, are you aware of C libraries that do the same job?
- I was thinking about having a look at maybe another language that targets the JVM like Clojure, Jython or JRuby. But all these languages are very high level and complicated (to create a compiler for them). I was looking for a simpler (I don't mind if it's unknown or unused) programming language that targets the JVM and it's compiler is open source. Any ideas?
回答1:
I would also recommend ASM, but have a look at Jasmin, I used it (or: had to use it) for an university-project, and it works quite well, I wrote a lexer/parse/analyzer/optimizer/generator combination for a programing-language using java and jasmin, so generating JVM Code. I uploaded the code here, the interesting part should be the source-code itself. In the folder "bytecode/InsanelyFastByteCodeCreator.java" you find a pice of code which transforms an AST Tree into the input format of jasmin assembler. Is quite straight forward.
The source-language (which was tranformed to the AST by Lexer+Parser+Analyzer) is a subset of Java called MiniJava. It lacks some "complicated" features like Inheritance, Constructors, static methods, private fields/methods. None of those features are difficult to implement, but there was another task to write an X86 backend (so to generate machine-assembler), and those things tend to get difficult if you got no JVM which handles some things.
In case you wonder about the strange Class name: The task of the university project was to transform the AST into a a SSA Graph (so a graph representing the input code), then optimize the graph and then turn the graph into java byte code. That was about 3/4 of the work of the project and the InsanlyFastByteCodeCreator was just a short-cut to test everything.
Have a look at the "Java Virtual Machine" book from Jon Meyer and Troy Downing. This book heavily references the Jasmin-Assembler, it's quite helpful for understanding the JVM internals.
回答2:
Last semester I have attended a "Compiler Construction" course. Our project was exactly what you want to do.
The language I've used to write my language was Scala. It runs on a JVM but supports a lot of advanced features that Java doesn't (still fully compatible with a pure java JVM).
To output java bytecode I've used the Scala CAFEBABE library. Well documented and you don't have to go deep inside java classes to understand what to do.
Beside the book, I think you can find a lot of infos by going trough the labs we've done during the course.
回答3:
ASM can be a solution for generating bytecode. To start, check the topics on generating elements from the manual.
回答4:
I was thinking to have a look at maybe another language that targets the JVM like Clojure, Jython or JRuby. But all these languages are very high level and complicated (to create a compiler for them).
Suggestion: You could have a look at Lua Programming Language, there are JVM implementations of it like LuaJ.
Lightweight, fast, Java-centric Lua interpreter written for J2ME and J2SE, with libraries for basic, string, table, package, math, io, os, debug and coroutine packages, a compiler, luajava bindings, and JSR-233 pluggable scripting engine bindings.
(Not to be confused with LuaJava that uses a native libs with JNI approach.)
回答5:
Last weekend, I was asking myself the same question to port my toy language to the JVM.
I spend only few hours searching information,so take this references with a grain of salt.
Language Implementation Patterns. I hate antlr but this book looks very good. If you dont like antlr neither, there is a very good about parsing "Parsing Techniques. A Practical Guide."
Learn to build configuration file readers, data readers, model-driven code generators, source-to-source translators, source analyzers, and interpreters. You don’t need a background in computer science—ANTLR creator Terence Parr demystifies language implementation by breaking it down into the most common design patterns. Pattern by pattern, you’ll learn the key skills you need to implement your own computer languages.
Chapter 10 cover in 30 pages (to fast IMO) this topics. But there are other chapter that probably you will be interested.
- 10 Building Bytecode Interpreters
- 10.1 Programming Bytecode Interpreters . .
- 10.2 Defining an Assembly Language Syntax
- 10.3 Bytecode Machine Architecture . . . . .
- 10.4 Where to Go from Here . . . . . . . . . .
- P.26. Bytecode Assembler . . . . . . . . . . .
- P.27. Stack-Based Bytecode Interpreter . . .
- P.28. Register-Based Bytecode Interpreter
The Implementation of Lua 5.0 This is a great paper about register- based bytecode machines. Go an read it even for the sake of it.
Lisp in Small Pieces. This book teach how to write a 2 schme compailers that compile to C. So many lessons can be learned from this book. I own a copy of this book and it is really good for anyone interesting is lisp, but maybe not your cup of tea.
This is a comprehensive account of the semantics and the implementation of the whole Lisp family of languages, namely Lisp, Scheme and related dialects. It describes 11 interpreters and 2 compilers ...
http://www.amazon.com/Lisp-Small-Pieces-Christian-Queinnec/dp/0521562473
- 10 Building Bytecode Interpreters
Check the Dalvik7 VM, a register-based VM. The DVM operates on bytecodes that are transformed from the Java Class files compiled by a Java compiler.
There is a mailing list about the topic, jvm-languages.
Are you planning to upload the code to anyplace? I would like to take a look.
回答6:
I would recommend that you first learn how JVM assembly works, if you don't already know it.
Many instructions have the form ?name, where ? is i if the instruction works on with an integer type and a if it works with a reference type.
Basically, JVM is a stack machine with no registers, so all instructions work with data directly on the stack. You can push/pop data with ?push/?pop and move data between local variables (stack locations referenced by offsets) and the top of the stack using ?store/?load. Some other important instructions are invoke??? and if_???.
For my university's compiler course we used Jasmin to assemble the programs. I don't know if this is the best way, but at least it is an easy place to start.
Here is an instruction reference for an old version of the JVM, which might contain fewer instructions than a new one.
回答7:
The best resource for the beginning might be Ola Bini's presentation. Grab the slides as well.
回答8:
First I'd back off, modify my compiler to output actual Java instead of Java byte codes (which means creating more of a translator than compiler), and compile the Java output with whatever Java environment is convenient (which would probably generate better object code than my own compiler).
You could use the same technique (eg, compile to C#) to generate CLI byte codes, or compile to Pascal to generate P-code, etc.
It's not clear why you're considering Java codes instead of using your own VM, but if it's for performance then of course you should also consider compiling to actual machine code.
回答9:
Of course once could use Java to write a new language. With the Java reflection-API You can achive a llot. If speed don't matters too much, I would give Java the preference instead of ASM. Programming is easier and less error-prone in Java (IMHO). Take a look at the RPN language 7th. It is entirely written in Java.
来源：https://stackoverflow.com/questions/3380498/create-a-jvm-programming-language