RubyX compiles ruby to binary (in ruby), hoping to be that X times faster
Go to file
Torsten Rüger b2260d856d And we are green again
After having over 600 failing tests at one point, this does feel good.
Even better, most of the risc/interpreter tests where i didn't change anything came gree without changing the tests. ie we have binary compatibility.
2019-08-14 11:11:26 +03:00
bin updated qemu wuth buster 2019-07-23 12:08:06 +03:00
lib And we are green again 2019-08-14 11:11:26 +03:00
stash update parser since not we run 2.4+ 2019-02-27 10:13:38 +02:00
test And we are green again 2019-08-14 11:11:26 +03:00
.codeclimate.yml disable duplication engine 2017-04-08 19:31:48 +03:00
.gitignore update parser 2019-06-01 21:01:11 +03:00
.reek fix remaining tests 2018-06-29 22:56:49 +03:00
.rubocop.yml remove test and stash from codeclimate and add vanilla rubocop 2016-12-17 13:25:20 +02:00
.ruby-version close #21 2019-02-07 18:24:35 +02:00
.travis.yml fix travis (try 1) 2019-02-07 19:05:06 +02:00
CodeStyle.md fix typos 2018-03-20 15:52:16 +05:30
Gemfile upgrade ruby and include thor in general 2019-02-07 11:05:40 +02:00
Gemfile.lock update parser 2019-06-01 21:01:11 +03:00
Guardfile moved all the normalize stuff over to the ruby layer 2018-07-19 14:47:29 +03:00
LICENSE.txt fix typos 2018-03-20 15:52:16 +05:30
Rakefile remove traces of salama 2017-08-29 18:38:51 +03:00
README.md bit of documentation 2018-08-14 11:23:19 +03:00
rubyx.gemspec update parser to use 2.6 2019-03-04 19:00:47 +02:00
ToDo.md bit of documentation 2018-08-14 11:23:19 +03:00

Build Status Code Climate Test Coverage

RubyX

RubyX is about native code generation in and of ruby. In other words, compiling ruby to binary, in ruby.

X can be read as X times faster, or a decade away, depending on mindset.

The last rewrite clarified the roles of the different layers of the system, see below. The overhaul is done and rubyx produces working binaries.

Processing goes through layers: Ruby --> Vool --> Mom --> Risc --> Arm --> binary .

Currently most basic constructs work to some (usable) degree, ie if, while, assignment, ivars, calling and dynamic dispatch all work. Simple blocks, those that ruby passes implicitly also work (lambdas not).

Work continues on memory management, which turns out to be pretty basic to do just about anything, even counting to a million.

Layers

Ruby

Ruby is input layer, we use whitequarks parser to parse ruby. The untyped ast is then transformed into a typed version. The classes and fields follow the ast output pretty much one to one. The we transform to Vool, removing much of ruby's "fluff".

Vool

Vool is a Virtual Object Oriented Language. Virtual in that is has no own syntax. But it has semantics, and those are substantially simpler than ruby.

Vool is Ruby without the fluff. No unless, no reverse if/while, no splats. Just simple oo. (Without this level the step down to the next layer was just too big)

Mom

The Minimal Object Machine layer is the first machine layer. This means it has instructions rather than statements. Instructions (in all machine layers) are a linked list.

Mom has no concept of memory yet, only objects. Data is transferred directly from object to object with one of Mom's main instructions, the SlotLoad.

Mainly Mom is an easy to understand step on the way down. A mix of oo and machine. In practise it means that the amount of instructions that need to be generated in vool is much smaller (easier to understand) and the mapping down to risc is quite straightforward.

Risc

The risc cpu architecture approach was a simplification of the cpu instruction set to a minimum. Arm, our main target, is a risc architecture, and much like Vool uncrinkles Ruby, the Risc layer simplifies ARM.

The Risc layer here abstracts the Arm in a minimal and independent way. It does not model any real RISC cpu instruction set, but rather implements what is needed for rubyx.

Instructions are derived from a base class, so the instruction set is extensible. This way additional functionality may be added by external code.

Risc knows memory and has a small set of registers. It allows memory to register transfer and back, and inter register transfer. But has no memory to memory transfer like Mom.

Arm

There is a minimal Arm assembler that transforms Risc instructions to Arm instructions. This is mostly a one to one mapping, though it does introduce the quirks that ARM has and that were left out of the Risc layer.

Elf

Arm instructions assemble themselves into binary code. A minimal Elf implementation is able to create executable binaries from the assembled code and Parfait objects.

Parfait

Generating code (by descending above layers) is only half the story in an oo system. The other half is classes, types, constant objects and a minimal run-time. This is what is Parfait is.

Parfait has basic clases like string/array/hash, and also anything that is really needed to express code, ie Class/Type/Method/Block.

Parfait is used at compile time, and the objects get serialised into the executable to make up, or make up the executable, and are thus available at run time. Currently the methods are not parsed yet, so do not exist at runtime yet.

Builtin

There are a small number of methods that can not be coded in ruby. For example an integer addition, or a instance variable access. These methods exists in any compiler, and are called builtin here.

Builtin methods are coded at the risc level with a dsl. Even though basically assembler, they are (quite readable) through the ruby magic.

Types and classes, static vs dynamic

Classes in dynamic languages are open. They can change at any time, meaning you can add/remove methods and use any instance variable. This is the reason dynamic languages are interpreted.

For Types to make any sense, they have to be static, immutable.

Some people have equated Classes with Types, this is a basic mistake in dynamic languages.

In rubyx a Type implements a Class (at a certain time of that classes lifetime). It defines the methods and instance variables available. This is key to generating efficient code that uses type information to access instance variables.

When a class changes, say a new method is added that uses a new instance variable, a new Type is generated to describe the class at that point. New code is generated for this new Type.

In essence the Class always has a current Type and many Types implement (different versions of) a Class.

All Objects have a Type, as their first member (also integers!). The Type points to the Class that the object has in oo terms.

Classes are defined by ruby code, but the methods of a Type (that are executed) are defined by Mom and Risc only.

Other

Interpreter

After doing some debugging on the generated binaries i opted to write an interpreter for the risc layer. That way tests run on the interpreter reveal most issues.

Debugger

And after the interpreter was done, i wrote a visual debugger. It is a simple opal application that nevertheless has proven a great help, both in figuring out what is going on, and in finding bugs.

Status

The above architecture is implemented. At the top level the RubyXCompiler works pretty much as you'd expect, by falling down the layers. And when it get's to the Risc layer it slots the builtin in there as if is were just normal code.

Specifically here is a list of what works:

  • if (with or without else)
  • while
  • return
  • assignment (local/args/ivar)
  • static calling (where method is determined at compile time)
  • dynamic dispatch with caching
  • implicit blocks, ie the ones that ruby passes implicitly and are used in enumerating

Contributing to rubyx

Probably best to talk to me, if it's not a typo or so.

I've started to put some github issues out, some basic some not so. Also there is a todo for the adventurous (bigger things, no BIG things).

Actual tasks that result in pulls, should start their life as a github issue. There we can discuss details so that no work is done in vain. If you're interested in an existing issues, just comment on it.

Fork and create a branch before sending pulls.

Copyright (c) 2014-8 Torsten Ruger.

See LICENSE.txt for further details.