Compile snippets into bytecode in assembler

Compile snippets into bytecode in assembler

The idea is to allow users to open a method assembler, then open a prompt allowing them to type in something like System.out.println("test"). This line would then be compiled against the current source and inserted into the assembler as lines of method bytecode.

It would be nice to support arbitrary variable support too, such as: if ($0 != 0){} --> *LOAD 0

[Col-E]

Could be improved, but variable support is in!

Need to get some proper unit tests made too…

[andylizi]

How can javassist access the value on the stack?

No it can’t - at least, not in its current state. After some digging however, I believe its possible to make it work without major modifications.

Javassist has a complete lexer/parser/bytecode-generator system just built for this, like a lightweight version of JavaParser. It will first parse the statement using a pre-populated symbol table and generate an AST tree representation. Then the corresponding bytecode will be emitted at the various at*** methods in CodeGen, MemberCodeGen and JvstCodeGen. One of these methods, atVariable, goes like this:

@Override
public void atVariable(Variable v) throws CompileError {
    Declarator d = v.getDeclarator();
    exprType = d.getType();
    arrayDim = d.getArrayDim();
    className = d.getClassName();
    int var = getLocalVar(d);

    if (arrayDim > 0)
        bytecode.addAload(var);
    else
        switch (exprType) {
        case CLASS :
            bytecode.addAload(var);
            break;
        case LONG :
            bytecode.addLload(var);
            break;
        case FLOAT :
            bytecode.addFload(var);
            break;
        case DOUBLE :
            bytecode.addDload(var);
            break;
        default :   // BOOLEAN, BYTE, CHAR, SHORT, INT
            bytecode.addIload(var);
            break;
        }
}

As you can see, this kind of structuring makes it easy to implement our own bytecode emitter that works on the stack. We won’t even need to change the pesky parser part (except here), just feed it some type info about things-on-stack through the symbol table and write some custom at*** generators and it magically works. At least, it should. Never done this before but don’t see why it shouldn’t - a value is a value, no matter whether it comes from a variable or the stack. It won’t be ‘piece of cake’ level of easy, but it should work.