Rutie — /ro͞oˈˌtī/rOOˈˌtI/rüˈˌtaI/
Integrate Ruby with your Rust application. Or integrate Rust with your Ruby application. This project allows you to do either with relative ease.
You are highly encouraged to read the source code for this project. Every method that has been
mapped from Ruby for public use in src/class/*
is very well documented with example code.
This is the best way to take off running with Rutie. There are also integration examples in the
examples directory which are based off of this README.
This project is a continuation of:
- Using Ruby in Rust
- Using Rust in Ruby
- Custom Ruby Objects in Rust
- Variadic Functions / Splat Operator
- Migrating from Ruru to Rutie
- Safety — The Rutie Philosophy vs The Rust Philosophy on Safety
- Troubleshooting
- It panics for some Rubies on CI server tests
- Rust signal: 11, SIGSEGV: invalid memory reference
- Error while loading shared libraries: libruby.so.#.#: cannot open shared object file: No such file or directory
- Calling methods from other methods within the
methods!
macro doesn't work - Handling exceptions raised from Ruby in Rust code
- Segfault during GC when using a Ruby method written in C
- Operating System Requirements
- Contributing
- Rutie's Future
- Additional Project History
- LICENSE
First add the dependency to your Cargo.toml
file.
[dependencies]
rutie = "0.6.1"
Then in your Rust program add VM::init()
to the beginning of its code execution path
and begin to use Rutie.
extern crate rutie;
use rutie::{Object, RString, VM};
fn try_it(s: &str) -> String {
let a = RString::new_utf8(s);
// The `send` method returns an AnyObject type.
let b = unsafe { a.send("reverse", &[]) };
// We must try to convert the AnyObject
// type back to our usable type.
match b.try_convert_to::<RString>() {
Ok(ruby_string) => ruby_string.to_string(),
Err(_) => "Fail!".to_string(),
}
}
#[test]
fn it_works() {
// Rust projects must start the Ruby VM
VM::init();
assert_eq!("selppa", try_it("apples"));
}
fn main() {}
NOTE: Currently in Linux you need to set
LD_LIBRARY_PATH
to point at the directory of your current Ruby library and in Mac you need to setDYLD_LIBRARY_PATH
with that info. You can get the path information with the following command:
ruby -e "puts RbConfig::CONFIG['libdir']"
This should let you run cargo test
and cargo run
.
Running cargo test
should have this test pass.
You can start a Ruby project with bundle gem rutie_ruby_example
and then once
you change into that directory run cargo init --lib
. Remove the TODOs from the gemspec
file. Add Rutie to the Cargo.toml
file and define the lib type.
[dependencies]
rutie = "0.6.1"
[lib]
name = "rutie_ruby_example"
crate-type = ["dylib"]
Then edit your src/lib.rs
file for your Rutie code.
#[macro_use]
extern crate rutie;
use rutie::{Class, Object, RString, VM};
class!(RutieExample);
methods!(
RutieExample,
_itself,
fn pub_reverse(input: RString) -> RString {
let ruby_string = input.
map_err(|e| VM::raise_ex(e) ).
unwrap();
RString::new_utf8(
&ruby_string.
to_string().
chars().
rev().
collect::<String>()
)
}
);
#[allow(non_snake_case)]
#[no_mangle]
pub extern "C" fn Init_rutie_ruby_example() {
Class::new("RutieExample", None).define(|itself| {
itself.def_self("reverse", pub_reverse);
});
}
And that's it for the Rust side. When using the methods!
macro or extern
functions
make sure the method name won't clash with any others. This is why this example is prefixed with pub_
.
Now you just need to load the library in Ruby. Add the rutie
gem to your gemspec or Gemfile.
# gemspec
spec.add_dependency 'rutie', '~> 0.0.3'
# Gemfile
gem 'rutie', '~> 0.0.3'
And then load the library in your main project file lib/rutie_ruby_example.rb
.
require 'rutie_ruby_example/version'
require 'rutie'
module RutieRubyExample
Rutie.new(:rutie_ruby_example).init 'Init_rutie_ruby_example', __dir__
end
That's all you need to load your Ruby things from Rust. Now to write the test in
test/rutie_ruby_example_test.rb
:
require "test_helper"
class RutieRubyExampleTest < Minitest::Test
def test_it_reverses
assert_equal "selppa", RutieExample.reverse("apples")
end
end
And to properly test it you will always need to run cargo build --release
whenever
you make any changes to the Rust code. Run the test with:
cargo build --release; rake test
Or better yet change your Rakefile
to always run the cargo build --release
before
every test suite run. Feel free to change the test input to prove it fails because
the above test works as is.
To create a Ruby object in Rust that can be returned directly to Ruby it needs just a few simple things.
Here's an example excerpt of code from FasterPath.
use rutie::types::{ Value, ValueType };
use rutie::{ RString, AnyObject, Object, Class, VerifiedObject };
pub struct Pathname {
value: Value
}
impl Pathname {
pub fn new(path: &str) -> Pathname {
let arguments = [RString::new_utf8(path).to_any_object()];
let instance = Class::from_existing("Pathname").new_instance(Some(&arguments));
Pathname { value: instance.value() }
}
pub fn to_any_object(&self) -> AnyObject {
AnyObject::from(self.value())
}
}
impl From<Value> for Pathname {
fn from(value: Value) -> Self {
Pathname { value }
}
}
impl Object for Pathname {
#[inline]
fn value(&self) -> Value {
self.value
}
}
impl VerifiedObject for Pathname {
fn is_correct_type<T: Object>(object: &T) -> bool {
Class::from_existing("Pathname").case_equals(object)
}
fn error_message() -> &'static str {
"Error converting to Pathname"
}
}
If the class does not yet exist in Ruby you'll need to account for creating it before generating a new instance of it. This object is now compatible to be returned into Ruby directly from Rust/Rutie. Note that this definition is merely a Rust compatible representation of the Ruby object and doesn't define any Ruby methods which can be used from Ruby.
A preferred way to integrate a dynamic amount of parameters has not yet been implemented in Rutie, but you can still manage to get it done in the following way.
use rutie::{AnyObject, Array};
use rutie::types::{Argc, Value};
use rutie::util::str_to_cstring;
use rutie::rubysys::class;
use std::mem;
pub extern fn example_method(argc: Argc, argv: *const AnyObject, _: AnyObject) -> AnyObject {
let args = Value::from(0);
unsafe {
let p_argv: *const Value = mem::transmute(argv);
class::rb_scan_args(
argc,
p_argv,
str_to_cstring("*").as_ptr(),
&args
)
};
let arguments = Array::from(args);
let output = // YOUR CODE HERE. Use arguments as you see fit.
output.to_any_object()
}
This style of code is meant to be used outside of the methods!
macro for now.
You may place this method on a class or module as you normally would from a methods!
macro definition.
#[macro_use]
extern crate rutie;
use rutie::{Class, Object, VM};
class!(Example);
// Code from above
fn main() {
VM::init();
Class::new("Example", None).define(|itself| {
itself.def("example_method", example_method);
});
}
The Rutie project has in its plans to remove the need for anyone to write unsafe code for
variadic support and will likely be updating the methods!
macro to support this natively.
For using Rutie versions less than 0.1 the change is simple. Replace all occurrences
of the string ruru
with rutie
in your program. And if you would like to use
ruby-sys
code from Rutie rather than requiring ruby-sys
you can change all existing
references to ruby_sys
to rutie::rubysys
.
You will have additional considerations to change like Error
being removed. For that; change instances of type ruru::result::Error
to rutie::AnyException
.
Migrated parse_arguments
from VM
to util
.
Internal changes util
from binding
and rubysys
have been replaced to reduce confusion and reduce duplication.
I'm writing this section to bring to light that, as of this writing, the safety that Rust likes to guarantee for its crates and the Rutie crate aren't currently aligned. The typical Rust safety for libraries wrapping C code is to have one unsafe crate with a -sys
extension in its name and then a crate that wraps that to make it safe.
Rutie is an official fork of the project Ruru and because of this a great deal of the decisions in design for the project remain unchanged. Rutie also brought in the ruby-sys
crate and treats it as an internal private API/module; and yet shares it openly for other developers to have full control to design their own API on top of it.
One of the glaring things that Rutie has that goes against the Rust Philosophy on Safety is that any of the methods that call Ruby code, can potentially raise an exception, and don't return the type Option<AnyObject, AnyException>
will panic when an exception is raised from Ruby… which kills the application process running. The way to avoid panics is simple: either guarantee the Ruby code you're running will never raise an exception, or Handling exceptions raised from Ruby in Rust code with "protect" methods that return the type Option<AnyObject, AnyException>
. Anyone can implement this safety by reading and understanding how the protect methods are written in this library and working with them.
The important thing to consider as to “why doesn't every method guarantee the safety as Rust would prescribe to?” is that exception handling in Ruby is not a zero cost abstraction. So there is a cost in performance when you need to implement it. One can easily argue that the guarantee of safety is far more important than leaving the risk in the hands of inexperienced developers. But one could also argue that it is better to leave the choice of performance cost, and the fact that exception capturing is occasionally unnecessary, up to the developer. Seeing how the legacy of design decisions is largely inherited this project leans towards the latter argument where the choice of being absolutely safe everywhere vs some extra speed in performance is up to the developer.
I'm not opposed to this project being 100% safe, but that will be a major change and a totally different API with many decisions that need to come into play. Also since this project doesn't strictly adhere to Rust safety principles, as a crate library would be expected to be, this project will not reach the stable 1.0 release as the idea of stability and safety are hand in hand.
I do like safety guarantees and as much as possible new features and language APIs will be built toward this. You can see what the design of a safe API would look like by examing the Enumerator features that have been implemented in this way (largely wrapping method names with calls to protect_send
).
Sometimes the Ruby binary built isn't the best for the system. Be sure to compile Ruby
for that system if this is the issue. With RVM do rvm reinstall --disable-binary
with
your choice of Ruby version.
This is an indication that you haven't started a Ruby VM in Rust yet with VM::init();
. Do this once
before using Ruby code from Rust.
Error while loading shared libraries: libruby.so.#.#: cannot open shared object file: No such file or directory
This may happen when a Ruby program is trying to link with libruby via Rutie. Simply disable linking
by setting the environment variable NO_LINK_RUTIE
before the Rust code is compiled. This is needed
to be done on the service TravisCI for example.
The way the macro is designed doesn't use the same parameter signatures you've provided and
therefore it is recommended to implement any methods you want to re-use in Rust with
functions outside of the methods!
macro. You can simply call that new external
method in the methods!
macro when defining methods for Ruby to use.
If you're using any method that doesn't return a Result<AnyObject, AnyException>
then
any exception raised from the Ruby side will interfere with that Ruby thread and cause
Rust to panic and stop. Ruby internally uses exceptions to effect the entire thread through
an internal thread global value. To handle places where Ruby may raise an exception during Rust
code execution you should use methods that are designed to handle that.
VM::eval
Object.protect_send
Object.protect_public_send
If you are writing lower level code and want to work more directly with the internal Ruby
exception you may use VM::protect
and read the source code for Object.protect_send
to
see how it's done.
One possible issue that may cause this is when you store an item in Rust in heap memory rather than the stack.
An example case that caused this issue is the following:
Class::from_existing("Pathname").new_instance(&vec![RString::new_utf8(path).to_any_object()])
Ruby's GC traces objects from the stack. Rust's Vec, on the other hand, stores elements in the heap. So Ruby's GC may not be able to find the string you created and may release it. — @irxground
To rememdy the issue it required not using Vec but rather Rust's array type to store the argument on the stack rather than the heap.
let arguments = [RString::new_utf8(path).to_any_object()];
Class::from_existing("Pathname").new_instance(&arguments)
Everything is tested against 64 bit operating systems with 64 bit Ruby & Rust builds. 32 bit isn't currently supported.
- Rust 1.26 or later
- Ruby (64 bit) 2.4 or later
- Rust 1.26 or later
- Ruby 2.5+ built with MingW (64 bit)
- MS Visual Studio (Build Tools)
Ruby needs to be compiled with the --enable shared
option. Dynamic linking to the Ruby library provides the best performance and best support. Static build support is incomplete for now.
If using RBENV then the following is recommended:
CONFIGURE_OPTS=--enable-shared rbenv install 2.6.3
You can check if your Ruby is compiled to be dynamically linked to by running the following and getting a "yes"
response.
ruby -e "pp RbConfig::CONFIG['ENABLE_SHARED']"
If you still run into ld: library not found for -lruby-static
issue, try running cargo clean
. This'll clean any artifacts from previous attempts.
If you'd like to make a pull request for adding static build support there are currently 3 methods not working with it and linking to the proper name of the ruby static lib file & path needs to be updated.
Contributors are welcome!
The code is organized in 3 main layers. The rubysys
folder is the raw mapping to Ruby C code and
all the methods from there are unsafe. The binding
folder is where we wrap those methods to abstract
away all the unsafe methods to safe methods. The class
folder is where the public API is implemented
for using Ruby with Rust code. These methods in the class
folder must all be documented and tested within
the documentation. There is a subfolder under class
for traits called traits
.
Macros for abstracting away complexity are in src/dsl.rs
.
Ruby's helper gem is in the submodule folder gem
.
Rutie will continue to be improved upon to be more and more compatible with every aspect of Ruby. It will also gradually change toward Rust safety, semantics, and best practices.
I imagine a future where Rutie is the stepping stone that helps Ruby switch from C to Rust.
As this package has taken 1.0 to mean both stable and safe and won't likely make a 1.0, then there can
be breaking changes expected in each MINOR version update. These MINOR version breaking changes will
occur in the public API of src/class/*
and src/helpers/*
. For private APIs there can be breaking
changes in each PATCH version update which includes src/rubysys/*
, src/binding/*
, and
src/util.rs
.
If you need some more examples of usage or the git blame history please look at the Ruru project as Rutie has had the README completely rewritten and this first git commit is from Ruru. Note that there are some fundamental changes which that README won't account for. This project also had ruby-sys merged in which may have some additional beneficial git history.
Both projects that were merged into this project contained identifiers under the MIT license.
This project follows with the same licensing. ruby-sys marked MIT as the license in the
Cargo.toml
file whereas ruru had that and included a LICENSE file. This projects LICENSE
has credited the original author by preserving the MIT license author line and appending new
author(s) which is permitted by the MIT LICENSE.
MIT LICENSE — see LICENSE