Tapioca

Typed APIs (that Ollie Coshed into an Acronym)

tapioca is an HTTP client for rust that aims to help the compiler help you to access REST+JSON APIs in a type-safer manner.

It uses the OpenAPI Initiative's schema specification to infer types for path and query parameters, request and response bodies, et al. and then serde to de/serialise them.

infer_api!(service, "https://service.api/schema.yml")
use service::path;

fn main() {
    let auth = service::ServerAuth::new();

    match path::get(&auth) {
        Ok(response) => match response.body() {
            path::OkBody::Status200(body) => println!("Thing is: {}", body.thing),
            path::OkBody::UnspecifiedCode(body) => {
                // We're forced to handle every status code in the schema;
                //  including the possibility that the server replies off-script.
                println!("I don't know what thing is!")
            },
        },
        Err(response) => match response.body() {
            path::ErrBody::Status403(body) => println!("That's not my thing"),
            path::ErrBody::UnspecifiedCode(_)
            | path::ErrBody::MalformedJson(_)
            | path::ErrBody::NetworkFailure() => println!("Something went wrong"),
        },
    }
}

So, we can pattern-match responses by status code, and access the JSON response as a rust type.

tapioca also aims to prevent you from shooting yourself in the foot with an invalid sequence of requests, such as 'GET after DELETE' on a particular resource: this is achieved by constructing resource IDs only from responses, and static values. DELETE functions cause the resource ID argument to be moved (while other methods only borrow) preventing it from being further used.

Getting started

In order to start using tapioca in your project, the first step is to locate the OAS schema for the API you wish to use. Let's assume it's at https://example.org/schema.yml. Then, add the latest version to your Cargo.toml as usual, and import tapioca with macros:

#[macro_use]
extern crate tapioca;

and invoke the infer_api macro to build a client for the API:

infer_api!(example, "https://example.org/schema.yml");

The macro expands at compile-time, building a typed client in-place; (almost) all the code it generates will be located under a module named example, or whatever we specify in the first argument. The only exception is two crates (which must be loaded at the root level) which are needed to be externed inside your crate in order to use their macros - at least for now, the Rust's macro system is seeing a lot of change, and this may be improved. These are serde_derive and tapicoa_codegen; consequently, they also need to be in your Cargo.toml, but any other crates used (not for macros) by tapicoa will not need this treatment, or pollute your project's namespace.

Accessing the client

The module built by infer_api contains modules with the names of each of the paths available on an API, and each of those contains a function for each of the HTTP methods valid for that resource. For example, to GET /foobars, the function ident is:

example::foobars::get

In order to call this function, we might need to supply some arguments for authentication, query parameters, request body, et al. - the types for these are located inside a module of the same name as the function, for example:

example::foobars::get::QueryParams

Authentication

Before we make a request, we need to introduce authentication - currently, authentication must be specified for every request, even if null.

Authentication requirements in an OAS schema are specified at two levels: server-wide, and operation specific - GET /foobars can have different requirements to other operations, which may just inherit from the server requirement. Thus we have two enums of acceptable authentication schemes:

example::ServerAuth
example::foobars::get::OpAuth

which must be used depends on whether the operation examples::foobars::get overrides the server-wide authentication requirement - but the type-checker will tell us if we get it wrong.

If there's no authentication required at all, we can just use:

example::ServerAuth::new();

If it's HTTP Basic, then (depending on whether it's a server or operation requirement):

example::ServerAuth::Basic(username: String, password:String);
example::foobars::get::OpAuth::Basic(username: String, password:String);

If it's a custom header:

example::ServerAuth::ApiKey(api_key: String);
example::foobars::get::OpAuth::ApiKey(api_key: String);

Though note that the variant identifier, e.g. Basic or ApiKey, depends on the name used in the OAS schema. This is because there may be multiple definitions of the same type.

Making a request

Now that we've seen how to construct an authentication argument, we can actually GET some foobars!

let auth = examples::ServerAuth::new();
let response = examples::foobars::get(&auth);

response is actually a Result<Response, Response>: if the response status code is an error, we get an Err(response), otherwise it's an Ok(response). This means we can use response.is_ok, response.is_err, and pattern matching:

match examples::foobars::get(&auth) {
    Ok(response) => foobar_handler(response),
    Err(response) => err_handler(response),
}

We can use further pattern matching in each of these handlers, in order to respond differently to different status codes:

fn foobar_handler(response: Response) {
    match response.body {
        OkBody::Status200(body) => {
            for foobar in body.the_foobars {
                println!("Foobar {} is named {}", foobar.id, foobar.name);
            }
        },
        OkBody::UnspecifiedCode(body)
        | OkBody::MalformedJson(body) => something_else(),
    }
}

where we always have UnspecifiedCode (one not in the schema) and MalformedJson (invalid JSON, or did not match schema) as well as a StatusXXX for each of the possibilities specified in the schema. err_handler would look similar, with ErrBody::Status403, etc.

Request bodies

Say this example::foobars collection also supports POSTing new foobars, we can supply the request body to create one like this:

let body = example::foobars::post::RequestBody {
    name: "Foobarry".into(),
    age: 12,
    email: None,
};

The structure and field types of the body is fully defined by the schema, and may include:

• i32, i64
• bool
• String
• Option<_>
• Vec<_>
• further structs

Query parameters

Query parameters are supplied much like request bodies:

let query = example::foobars::get::QueryParams {
    age: 34,
};

Path parameters

Path parameters are slightly different. Because of the need to distinguish example::foobars::get from a GET on a single resource in that collection, the name of the path parameter is encoded in the path module name, for example:

example::foobars__id_::get

If the API specifies two resource identifiers in a row, this would be foobars__id1___id2_. This gets ugly, and may be changed in a future version.

Path parameters can be constructed from the response, for example when creating a new resource and the server generated its ID, or from a static reference:

static provisioned_id = "fea3c8e91baa1";

fn main() {
    let auth = example::ServerAuth::new();
    let resource = examples::foobars__id_::Resource_id::from_static(provisioned_id);

    example::foobars__id_::get(&resource, &auth);
}