Mini-Java-Compiler

A compiler for MiniJava, a subset of Java, that produces LLVM IR. The generated IR is compiled to a native executable with clang. Built for the Compilers course at the Department of Informatics and Telecommunications, University of Athens.

How it works

1. Parsing – a JavaCC parser, with the syntax tree and visitor interfaces generated by JTB, parses the source file and reports syntax errors.
2. Symbol table – a first visitor (MyVisitor) collects classes, fields, methods and variables, computes field and vtable offsets, and rejects duplicate declarations.
3. Type checking – a second visitor (CheckingVisitor) checks every statement and expression: type compatibility with inheritance, method overriding rules, argument types, return types and more.
4. Code generation – a third visitor (LLVMVisitor) emits LLVM IR: objects with vtables for dynamic dispatch, heap-allocated arrays with runtime bounds checks, and short-circuit &&.

The compiler exits with a non-zero status and a descriptive message on the first error.

Example

class Factorial {
    public static void main(String[] a) {
        System.out.println(new Fac().compute(5));
    }
}

class Fac {
    public int compute(int n) {
        int result;
        if (n < 1)
            result = 1;
        else
            result = n * (this.compute(n - 1));
        return result;
    }
}

The recursive call becomes a virtual call through the object's vtable:

define i32 @Fac.compute(i8* %this, i32 %.n) {
	...
if1:
	%_11 = load i32, i32* %n
	%_12 = bitcast i8* %this to i8***
	%_13 = load i8**, i8*** %_12
	%_14 = getelementptr i8*, i8** %_13, i32 0
	%_15 = load i8*, i8** %_14
	%_16 = bitcast i8* %_15 to i32(i8*, i32)*
	%_18 = load i32, i32* %n
	%_19 = sub i32 %_18, 1
	%_17 = call i32 %_16(i8* %this, i32 %_19)
	%_20 = mul i32 %_11, %_17
	...
}

A complete example is in Example.java and its output in Example.ll.

Running it

The Docker image contains everything needed (JDK, JavaCC/JTB, clang):

docker build -t minijava .
docker run --rm minijava                      # run the test suite

# compile and run your own program
docker run --rm -v "$PWD:/src" minijava bash -c \
  "java -cp /app Main /src/Program.java && clang -o /tmp/prog /src/Program.ll && /tmp/prog"

Without Docker you need a JDK and clang 14 or older (the generated IR uses typed pointers), then:

make compile
java Main Program.java         # writes Program.ll
clang -o program Program.ll && ./program

Tests

run_tests.sh runs the 87 programs in test/typechecking:

• ERROR_* programs must be rejected by the compiler.
• Valid programs are compiled to LLVM IR and then to a native binary. Since MiniJava is a subset of Java, each one is also compiled with javac and run on the JVM, and the two outputs must match, including programs that end with an out-of-bounds array access.

The suite runs on every push via GitHub Actions.

Technologies

Java, JavaCC, JTB, LLVM IR, clang, Docker, GitHub Actions