published : 2009-09-22
updated : 2010-01-19
A compiler for fun and education. Written (mostly) in Ruby and based on the tutorial "Let's Build a Compiler" by Jack Crenshaw1.
The semantics are simple and familiar to most programmers. Eager evaluation, assignment with the equals sign, arithmetic using + - * and /, loops, if/else statement, etc. Integers are the only data type.
While the parser still closely resembles Crenshaw's recursive descent parser, back-end generates x85 machine code using a homegrown assembler in ~1000 lines of Ruby (just 650 lines of real code).
NOTE: OS X is the only platform that compiles working binaries right now. ELF support for Linux coming ... eventually.
You need Ruby and gcc. Ruby is standard on Macs but you'll need to install Xcode for gcc. You can also compile it yourself or use MacPorts, or homebrew.
You need Ruby and ld - which lives in the binutils package.
% sudo aptitude install ruby binutils
The build script should detect your platform. If not append 'elf' or 'macho' to the command.
% ./build.rb filename.code [elf | macho]
The resulting native executable is called 'filename' and you should be able it run it directly.
The parser starts by parsing a block of code. A block consists of one or more statements. Whitespace is largely ignored beyond delimiting tokens, so statements can be grouped on one line or spread out over multiple lines. With no explicit terminator this can look strange so we will see how it works out when the syntax evolves into something more complicated.
There are variables and integers. That's honestly about it. There are no functions or function calls, no closures, arrays, hashes, or anything else.
Supported statements are:
e.g. foo = 4096
e.g. if x < 0 a=0 else a=1 end
e.g. while x > 0 x=x-1 end
e.g. until x == 0 x=x-1 end
e.g. repeat x=x-1 if x == 0 break end end
e.g. for i=1 to 5 x=x+1 end
e.g. do 5 x=x+1 end
e.g. a=1 print
Print is strange, it prints the last value calculated in hex and that is all.
Supported operations are the following, in increasing order of precedence:
- add + and subtract -
- multiply * and divide /
- relations: == != < > <= >=
- boolean not !
- [unimplemented] or ||
- [unimplemented] and &&
- bit or | and bit xor ^
- bit and &
- unary plus + and minus -
Parentheses are used to force a specific order of evaluation.
As far as booleans go, 0 is false and everything else is true. Right now there are only integers so this makes sense.
It wasn't much fun generating assembly text, so I wrote an x86 assembler library in Ruby. It implements just the instructions needed for this compiler and is by no means complete. It only does 32-bit and no prefixes are supported. It's basically just a handful of instructions and mod-rm encoding. I use the system's linker and have no intention of writing my own, don't worry!
ELF support is still in C and not published in the repository. The class to output Mach-O binaries is found in asm/machofile.rb.
The asm/ directory holds the assembler but also the Mach-O code, for now. This is my first assembler and first time working with the x86 ISA, so it probably isn't great. It outputs horribly inefficient code and there are no optimizations.
I did not write this compiler with the intention of anyone else reading it but there are a reasonable amount of comments.
Whatever interests me really, I don't know yet.. Right now I need to clean up some of the code, now that object files of any size can be generated and tests pass again.