A Bytecode Alliance project
A cargo subcommand for building WebAssembly components according to the component model proposal.
cargo component
is a cargo
subcommand for creating WebAssembly components
using Rust as the component's implementation language.
cargo component
is considered to be experimental and is not currently
stable in terms of the code it supports building.
Until the component model stabilizes, upgrading to a newer cargo component
may cause build errors for existing component projects.
- The
cargo component
subcommand is written in Rust, so you'll want the latest stable Rust installed. - A protobuf compiler is also required for registry support.
To install the cargo component
subcommand, run the following command:
cargo install --git https://github.com/bytecodealliance/cargo-component
The currently published crate on crates.io is a nonfunctional placeholder and these instructions will be updated to install the crates.io package once a proper release is made.
Today, developers that target WebAssembly typically compile a monolithic program written in a single source language to a WebAssembly module. The WebAssembly module can then be used in all sorts of places: from web browsers to cloud compute platforms. WebAssembly was intentionally designed to provide the portability and security properties required for such environments.
However, WebAssembly modules are not easily composed with other modules into a single program or service. WebAssembly only has a few primitive value types (integer and floating point types) and those are inadequate to describe the complex types that developers would desire to exchange between modules.
To make things even more challenging, WebAssembly modules typically define their own local linear memories, meaning one module can't access the (conceptual) address space of another. Something must sit between the two modules to facilitate communication when pointers are passed around.
While it is possible to solve these challenges with the existing WebAssembly standard, doing so is burdensome, error-prone, and requires foreknowledge of how the WebAssembly modules are implemented.
The WebAssembly component model proposal provides a way to simplify the process of building WebAssembly applications and services out of reusable pieces of functionality using a variety of source languages, all while still maintaining the portability and security properties of WebAssembly.
At its most fundamental level, WebAssembly components may be used to wrap a WebAssembly module in a way that describes how its interface, a set of functions using complex value types (e.g. strings, variants, records, lists, etc.), is translated to and from the lower-level representation required of the WebAssembly module.
This enables WebAssembly runtimes to know specifically how they must facilitate the exchange of data between the discrete linear memories of components, eliminating the need for developers to do so by hand.
Additionally, components can describe their dependencies in a way that modules simply cannot today; they can even control how their dependencies are instantiated, enabling a component to virtualize functionality needed by a dependency. And because different components might have a shared dependency, hosts may even share the same implementation of that dependency to save on host memory usage.
A primary goal of cargo component
is to try to imagine what
first-class support for WebAssembly components might look like for Rust.
That means being able to reference WebAssembly components via
Cargo.toml
and have WebAssembly component dependencies used in the
same way as Rust crate dependencies:
- add a dependency on a WebAssembly component to
Cargo.toml
- reference it like you would an external crate (via
<name>::...
) in your source code - build using
cargo component build
and out pops your component!
To be able to use a WebAssembly component from any particular programming language, bindings must be created by translating a WebAssembly component's interface to a representation that a specific programming language can understand.
Tools like wit-bindgen
exist to generate those bindings for different languages,
including Rust.
wit-bindgen
even provides procedural macros to generate the
bindings "inline" with the component's source code.
Unlike wit-bindgen
, cargo component
doesn't use procedural macros
or a build.rs
file to generate bindings. Instead, it generates them
into external crates that are automatically provided to the Rust
compiler when building your component's project.
This approach does come with some downsides, however. Commands like
cargo metadata
and cargo check
used by many tools (e.g.
rust-analyzer
) simply don't work because they aren't aware of the
generated bindings. That is why replacement commands such as
cargo component metadata
and cargo component check
exist.
The hope is that one day (in the not too distant future...) that
WebAssembly components might become an important part of the Rust
ecosystem such that cargo
itself might support them.
Until that time, there's cargo component
!
A quick note on the implementation status of the component model proposal.
At this time of this writing, no WebAssembly runtimes have fully implemented the component model proposal.
Wasmtime has implementation efforts underway to support it, but it's still a work-in-progress.
Until runtime support grows and additional tools are implemented
for linking components together, the usefulness of cargo component
today is effectively limited to creating components that runtime
and tooling developers can use to test their implementations.
Currently cargo component
targets wasm32-wasi
by default.
As this target is for a preview1 release of WASI, the WebAssembly module produced by the Rust compiler must be adapted to the preview2 version of WASI supported by the component model.
The adaptation is automatically performed when wasm32-wasi
is targeted.
To prevent this, override the target to wasm32-unknown-unknown
using the
--target
option when building. This, however, will disable WASI support.
Use the preview2 version of wasi-common
in your host to run components
produced by cargo component
.
When the Rust compiler supports a preview2 version of the WASI target,
support in cargo component
for adapting a preview1 module will be removed.
Use cargo component new <name>
to create a new component.
This will create a wit/world.wit
file describing the world that the
component will target:
default world component {
export hello-world: func() -> string
}
The component will export a hello-world
function returning a string.
The implementation of the component will be in src/lib.rs
:
struct Component;
impl bindings::Component for Component {
fn hello_world() -> String {
"Hello, World!".to_string()
}
}
bindings::export!(Component);
Here bindings
is the bindings crate that cargo component
generated for you.
The export!
macro informs the bindings that the Component
type exports
all interfaces listed in Cargo.toml
.
The cargo component
subcommand has some analogous commands to cargo itself:
cargo component new
— creates a new WebAssembly component Rust project.cargo component add
— adds a component interface dependency to a cargo manifest file.cargo component build
— builds a WebAssembly component from a Rust project using thewasm32-unknown-unknown
target by default.cargo component metadata
— prints package metadata ascargo metadata
would, except it also includes the metadata of generated bindings.cargo component check
— checks the local package and all of its dependencies (including generated bindings) for errors.cargo component clippy
— same ascargo clippy
but also checks generated bindings.cargo component registry
— a command for interacting with local component registries.cargo component update
— same ascargo update
but also updates the dependencies in the component lock file.
More commands will be added over time.
rust-analyzer is an extremely useful tool for analyzing Rust code and is used in many different editors to provide code completion and other features.
rust-analyzer depends on cargo metadata
and cargo check
to discover
workspace information and to check for errors.
Because cargo component
generates code for dependencies that cargo
itself is
unaware of, rust-analyzer will not detect or parse the generated bindings;
additionally, diagnostics will highlight any use of the generated bindings as
errors.
To solve this problem, rust-analyzer must be configured to use the
cargo-component
executable as the cargo
command. By doing so, the cargo component metadata
and cargo component check
subcommands will inform
rust-analyzer of the generated bindings as if they were normal crate
dependencies.
To configure rust-analyzer to use the cargo-component
executable, set the
rust-analyzer.server.extraEnv
setting to the following:
"rust-analyzer.server.extraEnv": { "CARGO": "cargo-component" }
By default, cargo component new
will configure Visual Studio Code to use
cargo component
by creating a .vscode/settings.json
file for you. To
prevent this, pass --editor none
to cargo component new
.
Please check the documentation for rust-analyzer regarding how to set settings for other IDEs.
cargo component
is a Bytecode Alliance
project, and follows
the Bytecode Alliance's Code of Conduct and
Organizational Code of Conduct.
You'll clone the code via git
:
git clone https://github.com/bytecodealliance/cargo-component
We'd like tests ideally to be written for all changes. Test can be run via:
cargo test
You'll be adding tests primarily to the tests/
directory.
Changes to cargo component
are managed through pull requests (PRs). Everyone
is welcome to submit a pull request! We'll try to get to reviewing it or
responding to it in at most a few days.
Code is required to be formatted with the current Rust stable's cargo fmt
command. This is checked on CI.
The CI for the cargo component
repository is relatively significant. It tests
changes on Windows, macOS, and Linux.
It also performs a "dry run" of the release process to ensure that release binaries can be built and are ready to be published (coming soon).
Publication of this crate is entirely automated via CI. A publish happens
whenever a tag is pushed to the repository, so to publish a new version you'll
want to make a PR that bumps the version numbers (see the bump.rs
scripts in
the root of the repository), merge the PR, then tag the PR and push the tag.
That should trigger all that's necessary to publish all the crates and binaries
to crates.io.