Hematita Da Lua is an interpreter for the scripting language Lua, written entirely in 100% safe Rust. Hematita is the portugese word for hematite, a type of iron oxide, or rust, and lua is the portugese word for moon. 'Hematita Da Lua' is a pun on what the project is, and the discovery that iron on the moon is rusting.
The purpose of the project is to provide a hardened Lua interpreter resiliant to security vulnerabilities. It accomplishes this by using no unsafe code, being compileable on stable, and by relying upon a minimal number of dependencies. With this, we can be confident we are safe from any yet-to-be-found security vulnerabilities in C code. No disrespect to the standard Lua implementation and other C projects.
That said, it is important to note that Hematita is not stable, is very early in it's development and may be buggy. It is my hope that, with enough time to mature, Hematita will be able to garuntee these things.
If you'd like to give the interpreter a test drive, run cargo install hematita_cli
, and then run hematita_cli
. You'll be placed in a basic REPL, and you can press Ctrl
+ C
at any time to exit. A large proportion of the Lua code you throw at it should work fine, but not everything, such as some features of for
loops. Please file an issue if you encounter anything that doesn't work, and it'll be fixed soon in a future version!
The command line interface has quite a few options; running it with --help
will show them all.
OPTIONS:
-h, --help Displays this and quits
-V, --version Displays version information
-v, --verbose Runs with verbose output
-i, --interactive Runs in interactive mode, after running SOURCE
-e, --evaluate Treats source as direct source code, rather than a file
-b, --byte-code Shows byte code rather than executing
-s, --ast Shows abstract syntax tree rather than executing
-t, --tokens Shows tokens rather than executing
Currently, --verbose
does nothing, and is ignored. Running with either --help
or --version
will prevent any code from being ran. --interactive
can be passed with a file, and after execution of the file ends, you'll be dropped into a REPL with all the state of the script left for you to mess with. Using --evaluate
will evaluate code passed directly on the command line, rather than loading it from a file.
The --byte-code
, --ast
, and --tokens
options all change the output of the interpreter. Using --byte-code
will print out the compiled byte code of the program rather than executing it. --ast
will print out the interpreted abstract syntax tree rather than executing, and --tokens
will print out debug views of the token stream rather than executing. Each of these options correspond to a different segment of the interpreter, see the internals section for more information.
Embedding Hematitia is fairly straight forward and only requires stringing together each segment of the interpreter. As always, require the crate in your Cargo.toml
. Then, you're just six lines of code away from running Lua code in your next big project.
use hematita::{ast::{lexer, parser}, compiler, vm, lua_lib, lua_tuple};
// Ready our Lua source code.
let source = "print(\"Hello, World!\")";
// Create a lexer (just a token iterator) from the characters of our source code.
let lexer = lexer::Lexer {source: source.chars().peekable()}.peekable();
// Parse from the lexer a block of statements.
let parsed = parser::parse_block(&mut parser::TokenIterator(lexer)).unwrap();
// Compile bytecode from the block of statements.
let compiled = compiler::compile_block(&parsed);
// Prepare the global scope.
let global = lua_lib::standard_globals();
// Create the virtual machine...
let virtual_machine = vm::VirtualMachine::new(global);
// And run the byte code.
virtual_machine.execute(&compiled.into(), lua_tuple![].arc()).unwrap();
VirtualMachine
is Send
+ Sync
, and includes a lifetime 'n
for the native functions and user data you make for your convenience. So, bust out the old crossbeam::thread::Scope
, and go wild. If you're curious about what each of the lines do, see the internals section for more information.
Creating your own native function is easy too. All it takes is to modify the global scope with any old function like type, i.e. any dynamically dispatchable Fn
.
let number = Mutex::new(0);
// Rust is bad at inferring closure paramaters, so it needs a &_. :(
let counter = move |_, _: &_| {
let mut lock = number.lock().unwrap();
let old = *lock;
*lock += 1;
Ok(lua_tuple![old].arc())
};
let global = {
let globals = standard_globals();
let mut data = globals.data.lock().unwrap();
data.insert(lua_value!("counter"), Value::NativeFunction(&counter));
drop(data);
globals
};
The same goes for creating your own user data. Make a type, implement vm::value::UserData
for it, and insert it into the global scope with Value::UserData
. (Or, if you prefer, make a native function function that returns it.) As is typical with implementing your own user data, the vast majority of your type will be implemented via the metatable. You can lock your metatable by adding a __metatable
entry into it.
Hematita is composed of four main segments. Those being, ast::lexer
, ast::parser
, compiler
, and vm
, each segment depending on the last. Each segment represents a different process, the lexer
is the lexing process, the parser
is the parsing process, so on and so forth.
Each segment can be used on it's own, but they're best used all together. If you'd like to just lex and parse lua code, the ast
module can totally handle that. If you'd like to just run hand crafted bytecode, the vm
module is well suited for it. But the real effect comes from stringing everything together, to form a complete interpreter.
The lexer just turns a stream of characters into a stream of Token
s. It's effectively just an operation over an iterator. You can read it's docs here, note that they are incomplete.
The parser takes a stream of tokens, and turns it into a Block
. A Block
is just a Vec
of Statement
s. Statement
s are an internal representation of a Lua statement. You can read it's docs here, note that they are incomplete.
The compiler takes a Block
, and produces a Chunk
. A Chunk
is just a Vec
of OpCode
s, with some metadata. It is effectively a one to one transformation, so no error handling is needed. You can read it's docs here, note that they are incomplete.
The virtual machine takes a Function
, and executes it. A Function
is just an instantiated form of a Chunk
, with associated up-values. It can be made just by calling into
on a Chunk
. The virtual machine is effectively a match statement over every OpCode
, and the code that implements it. You can read it's docs here, note that they are incomplete.