The module graph/dependency logic for the Deno CLI.
This repository is a Rust crate which provides the foundational code to be able to build a module graph, following the Deno CLI's module resolution logic. It also provides a web assembly interface to the built code, making it possible to leverage the logic outside of the Deno CLI from JavaScript/TypeScript.
ModuleGraph::new(GraphKind::All) creates a new module graph. From there, you
can use the .build(...).await method to add roots. The build method requires
the root module specifiers/URLs for the graph and an implementation of the
source::Loader trait. It also optionally takes implementation of the
source::Resolver trait. It will load and parse the root module and recursively
all of its dependencies, returning asynchronously a resulting ModuleGraph.
Implementing this trait requires the load() method and optionally the
get_cache_into() method. The load() method returns a future with the
requested module specifier and the resulting load response. This allows the
module graph to deal with redirections, error conditions, and local and remote
content.
The get_cache_info() is the API for exposing additional meta data about a
module specifier as it resides in the cache so it can be part of a module graphs
information. When the graph is built, the method will be called with each
resolved module in the graph to see if the additional information is available.
This trait "replaces" the default resolution logic of the module graph. It is intended to allow concepts like import maps and alternative resolution logic to "plug" into the module graph.
It has two methods, resolve and resolve_types, which both have default
implementations. resolve takes the string specifier from the source and the
referring specifier and is expected to return a result with the resolved
specifier.
resolve_types takes a specifier and is expected to return a result with an
optional module specifier and optional source range of the types that should be
used. For example, if you are trying represent the "typings" field from a
package.json file, when you receive the request on resolve_types for the
main module of the package, you would respond with the absolute specifier to the
types along with a range that indicates the file URL to the package.json and
the range where it was specified. Including the range is useful to allow errors
produced from the graph to indicate "where" the dependency came from.
MemoryLoader is a structure that implements the source::Loader trait and is
designed so that a cache of modules can be stored in memory to be parsed and
retrieved when building a module graph. This is useful for testing purposes or
in situations where the module contents is already available and dynamically
loading them is not practical or desirable.
A minimal example would look like this:
use deno_graph::ModuleSpecifier;
use deno_graph::ModuleGraph;
use deno_graph::GraphKind;
use deno_graph::source::MemoryLoader;
use deno_graph::source::Source;
use futures::executor::block_on;
fn main() {
let mut loader = MemoryLoader::new(
vec![
(
"file:///test.ts",
Source::Module {
specifier: "file:///test.ts",
maybe_headers: None,
content: "import * as a from \"./a.ts\";"
}
),
(
"file:///a.ts",
Source::Module {
specifier: "file:///a.ts",
maybe_headers: None,
content: "export const a = \"a\";",
}
),
],
Vec::new(),
);
let roots = vec![ModuleSpecifier::parse("file:///test.ts").unwrap()];
let future = async move {
let mut graph = ModuleGraph::new(GraphKind::All);
graph.build(roots, &mut loader, Default::default()).await;
println!("{:#?}", graph);
};
block_on(future)
}See js/README.md.
The build script (_build.ts) requires the Deno CLI to be installed and
available in the path. If it is, the following command should just work:
> deno task build
We appreciate your help!
To contribute, please read our contributing instructions.
This repository includes .devcontainer metadata which will allow a development
container to be built which has all the development pre-requisites available to
make contribution easier.
