rubyx/README.md

177 lines
7.0 KiB
Markdown
Raw Normal View History

2015-07-18 12:06:42 +02:00
[![Build Status](https://travis-ci.org/salama/salama.svg?branch=master)](https://travis-ci.org/salama/salama)
[![Gem Version](https://badge.fury.io/rb/salama.svg)](http://badge.fury.io/rb/salama)
[![Code Climate](https://codeclimate.com/github/salama/salama/badges/gpa.svg)](https://codeclimate.com/github/salama/salama)
[![Test Coverage](https://codeclimate.com/github/salama/salama/badges/coverage.svg)](https://codeclimate.com/github/salama/salama)
2014-05-30 13:49:34 +02:00
2015-07-18 12:06:42 +02:00
# Salama
2014-04-14 14:51:44 +02:00
2015-03-15 12:58:17 +01:00
Salama is about native code generation in and of ruby.
2014-04-14 15:46:17 +02:00
It is probably best to read the [The Book](http://dancinglightning.gitbooks.io/the-object-machine/content/) first.
2015-03-15 12:58:17 +01:00
2015-10-07 10:32:48 +02:00
The current third rewrite adds a system language, with the idea of compiling ruby to that language, bosl.
The original ruby parser has been remodelled to parse bosl and later we will use whitequarks
parser to parse ruby. Then it will be ruby --> bosl --> assembler --> binary .
2015-03-15 12:58:17 +01:00
## Done
2015-10-07 10:32:48 +02:00
Some things that are finished (for *a* definition of finished, ie started)
### Interpreter
After doing some debugging on the generated binaries i opted to write an interpreter for the
register layer. That way test runs on the interpreter reveal most issues.
### Debugger
And after the interpreter was done, i wrote a [visual debugger](https://github.com/salama/salama-debugger).
It is a simple opal application that nevertheless has proven great help both in figuring out
what is going on, and in finding bugs.
2015-03-15 12:58:17 +01:00
### Assembly
2014-04-16 11:45:36 +02:00
Produce binary that represents code.
Traditionally called assembling, but there is no need for an external file representation.
2014-04-16 11:45:36 +02:00
Most instructions are in fact assembling correctly. Meaning i have tests, and i can use objdump to
verify the correct assembler code is disassembled
2014-04-27 21:19:32 +02:00
2014-08-22 17:00:23 +02:00
I even polished the dsl and so (from the tests), this is a valid hello world:
2014-04-27 21:19:32 +02:00
2014-09-19 18:39:08 +02:00
hello = "Hello World\n"
@program.main do
2014-04-27 21:19:32 +02:00
mov r7, 4 # 4 == write
mov r0 , 1 # stdout
add r1 , pc , hello # address of "hello World"
mov r2 , hello.length
swi 0 #software interupt, ie kernel syscall
mov r7, 1 # 1 == exit
swi 0
2014-09-19 18:39:08 +02:00
end
write(7 + hello.length/4 + 1 , 'hello')
2014-04-16 11:45:36 +02:00
2015-03-15 12:58:17 +01:00
### Linking
2014-04-16 11:45:36 +02:00
2015-03-15 12:58:17 +01:00
Package the code into an executable, currently elf, and very simple at that.
2014-04-21 16:27:05 +02:00
Above Hello World can be linked and run. And will say its thing.
2014-04-27 21:19:32 +02:00
2015-03-15 12:58:17 +01:00
There is no way to link c code currently and not planned either, for some time (see next)
### Syscalls
2014-04-21 16:27:05 +02:00
2015-03-15 12:58:17 +01:00
Some small portion of what libc usually provides is needed even right at the beginning.
Mainly file open and read, exit, that kind of thing. Looking at libc implementations and
kernel "api" docs, this is quite simple to do.
2014-04-16 11:45:36 +02:00
2015-03-15 12:58:17 +01:00
As said, "Hello world" comes out and does use syscall 4.
Also the program stops by syscall exit.
2015-03-15 12:58:17 +01:00
The full list is on the net and involves mostly just work.
2014-09-19 18:39:08 +02:00
2015-10-07 10:32:48 +02:00
### Parse Bosl
2015-10-07 10:32:48 +02:00
Parse bosl, using Parslet. This has been separated out as it's own gem, [salama-reader](https://github.com/salama/salama-reader).
2015-10-07 10:32:48 +02:00
Bosl is now fully typed (all variables, arguments and return). Also it has statements, unlike ruby
where everything is an expressions. Statements have no value. Otherwise it is quite basic, and
it's main purpose is to have an oo system language to compile to.
I spent some time on the parse testing framework, so it is safe to fiddle and add.
2015-10-07 10:32:48 +02:00
In fact it is very modular and easy to add to.
2015-03-15 12:58:17 +01:00
### Virtual: Compile the Ast
2014-08-22 17:00:23 +02:00
Since we now have an Abstact syntax tree, it needs to be compiled to a virtual machine Instruction format.
2015-03-15 12:58:17 +01:00
For the parsed subset that's almost done.
2014-08-22 17:00:23 +02:00
It took me a while to come up with a decent but simple machine model. I had tried to map straight to hardware
but failed. The current Virtual directory represent a machine with basic oo features.
### Parfait - the runtime
After an initial phase where i aimed for a **really** really small run-time, i have now started to
implement a more decent set classes and functions. This is a process off course.
I was encouraged by the thought that a large amount of the run-time code can actually be
reused at compile time, by using inlining. That off course assumes that i figure out how to do
inlining, but i have at least an idea.
### Sof
2015-10-07 10:32:48 +02:00
Salama Object File format is a yaml like format to look at code dumps and help testing.
The dumper is ok and does produce (as intended) considerably denser dumps than yaml
2014-05-30 13:49:34 +02:00
When a reader is done (not started) the idea is to use sof as pre-compiled, language independent
exchange format, have the core read that, and use the mechanism to achieve language independence.
2014-09-19 18:39:08 +02:00
2015-10-07 10:32:48 +02:00
## Status
2015-10-07 10:32:48 +02:00
Currently all the work is on the bosl front. Also documenting the *small* change of a new language.
2014-09-19 18:39:08 +02:00
2015-10-07 10:32:48 +02:00
I'll do some simple string and fibo examples in bosl next.
2014-09-19 18:39:08 +02:00
2015-10-07 10:32:48 +02:00
Next will be the multiple return feature and then to try to compile ruby to bosl.
2015-03-15 12:58:17 +01:00
## Future
#### Blocks
2014-09-19 18:39:08 +02:00
2015-03-25 16:29:39 +01:00
Implement ruby Blocks, and make new vm classes to deal with that. This is in fact a little open,
but i have a general notion that blocks are "just" methods with even more implicit arguments.
2014-09-19 18:39:08 +02:00
2015-03-15 12:58:17 +01:00
#### Exceptions
2014-04-16 11:45:36 +02:00
Implement Exceptions. Conceptually this is not so difficult in an oo machine as it would be in c.
2014-04-16 11:45:36 +02:00
I have a post [about it](http://salama.github.io/2014/06/27/an-exceptional-though.html)
2014-04-16 11:45:36 +02:00
which boils down to the fact that we can treat the address to return to in an exception quite
2015-03-25 16:29:39 +01:00
like a return address from a function. Ie just another implicit parameter
(as return is really an implicit parameter, a little like self for oo)
2014-04-16 11:45:36 +02:00
2015-03-15 12:58:17 +01:00
### C linking
2014-04-16 11:45:36 +02:00
Implement a way to call libc and other c libraries. I am not placing a large emphasis on this personally,
but expect somebody will come along and have library they want to use so much they can't stop themselves.
2014-09-19 18:39:08 +02:00
Personally i think a fresh start is what we need much more. I once counted the call chain from a simple
printf to the actual kernel invocation in some libc, and it was getting to 10!
I hope with dynamic (re)compiling and intelligent inlining, we can do better than that.
2014-05-27 18:19:55 +02:00
2015-03-15 12:58:17 +01:00
### Stary sky
2014-05-30 13:49:34 +02:00
2015-03-15 12:58:17 +01:00
Iterate:
2014-05-30 13:49:34 +02:00
1. more cpus (ie intel)
2. more systems (ie mac)
3. more syscalls, there are after all some hundreds
4. Ruby is full of niceties that are not done, also negative tests are non existant
2014-05-30 13:49:34 +02:00
5. A lot of modern cpu's functionality has to be mapped to ruby and implemented in assembler to be useful
6. Different sized machines, with different register types ?
7. on 64bit, there would be 8 bits for types and thus allow for rational, complex, and whatnot
8. Housekeeping (the superset of gc) is abundant
2014-09-19 18:39:08 +02:00
9. Any amount of time could be spent on a decent digital tree (see judy). Or possibly Dr.Cliffs hash.
10. Also better string/arrays would be good.
11. The minor point of threads and hopefully lock free primitives to deal with that.
12. Other languages, python at least, maybe others
13. translation of the vm instructions to another vm, say js
2014-05-30 13:49:34 +02:00
And generally optimize and work towards that perfect world (we never seem to be able to attain).
2014-04-16 11:45:36 +02:00
2014-04-14 15:46:17 +02:00
2014-07-29 17:33:11 +02:00
Contributing to salama
2014-04-14 15:46:17 +02:00
-----------------------
2014-04-27 21:19:32 +02:00
Probably best to talk to me, if it's not a typo or so.
I do have a todo, for the adventurous.
Fork and create a branch before sending pulls.
2014-04-14 15:46:17 +02:00
== Copyright
Copyright (c) 2014 Torsten Ruger. See LICENSE.txt for
further details.