Syntax-Directed Translation for Javac: Modified Javac to compile a new language (C-Minus-Minus).

• In this project I modified Javac (openjdk-6-src-b22-28_feb_2011) for building a compiler-compiler using Syntax-Directed Translation (SDT).
• I added a new construct to the Javac parser called grammar, representing a full-fledged Syntax-Directed Definition which maps a CFG to its production rules
  • You can find the new grammar construct here in the javac source code
• I then used that grammar to build a compiler for a simple language called C-Minus-Minus (short: CMM) [CMM specs pdf] [CMM samples] [my CMM compiler source code] and convert it to a proprietary assembly language called BASS (BASS documentation + examples).
• I even patched a bug in Javac (enum parsing was broken) that was in that build (openjdk-6-src-b22-28_feb_2011) [official download link]
• Within my CMM compiler, the language is defined in the CMMGrammar.
  • NOTE: You see this right - This .java file does not contain a class, interface etc... but instead it defines a public grammar CMMGrammar - That is why I had to make modifications to Javac! 😊

The CMMCompiler uses the CMMGrammar to compile everything in a few simple steps.

1. The CMMParser builds the AST, based on the Grammar rules
   • The parser internally calls the Scanner which generates a token stream
2. When the AST is ready, we start running through it and execute the grammar-defined rules.
   • The grammar can define an arbitrary amount of grammarpasses, and the CMMGrammar defines two, each with their own production rules (code to be executed when matching the lefthand-side patterns in the AST): prep0 to prepare everything and gen to generate code (using the Gen class)
3. The CMMGrammar is only 724 lines long (including comments!). Our modification to the Javac parser then takes that and converts it into a regular Java class before letting the Javac code generator worry about the rest.
   • NOTE: You find that the equivalent CMMGrammar class emitted by our modified Javac has over 5000 lines, that's almost 7 times as many lines, and does contain any comments!
4. I then used this compiler to successfully compile these 7 sample CMM programs to BASS assembly code

You can find the initial proposal and 2011 final report in this folder. NOTE: The report was not very well written (I loved coding those thousands of lines, but report writing I did not enjoy back then...)