Skip to content

Latest commit

 

History

History
230 lines (196 loc) · 11.8 KB

UPGRADE.md

File metadata and controls

230 lines (196 loc) · 11.8 KB
Raw

Upgrade Guide

Upgrading from 4.x to 5.x

The 5.0 release includes breaking changes to address several long-standing API issues, along with a few minor improvements. Consider following the tips below to help ensure a smooth upgrade process.

Upgrade Rust to 1.65 or newer

The minimum supported Rust version (MSRV) is now 1.65. Going forward, this crate will maintain a policy of supporting Rust releases going back at least 6 months. Changes that break compatibility with Rust releases older than 6 months will no longer be considered SemVer breaking changes and will not result in a new major version number for this crate. MSRV changes will coincide with minor version updates and will not happen in patch releases.

Add typestate generic types to Client

Each auth flow depends on one or more server endpoints. For example, the authorization code flow depends on both an authorization endpoint and a token endpoint, while the client credentials flow only depends on a token endpoint. Previously, it was possible to instantiate a Client without a token endpoint and then attempt to use an auth flow that required a token endpoint, leading to errors at runtime. Also, the authorization endpoint was always required, even for auth flows that do not use it.

In the 5.0 release, all endpoints are optional. Typestates are used to statically track, at compile time, which endpoints' setters (e.g., set_auth_uri()) have been called. Auth flows that depend on an endpoint cannot be used without first calling the corresponding setter, which is enforced by the compiler's type checker. This guarantees that certain errors will not arise at runtime.

In addition to unconditional setters (e.g., set_auth_uri()), each endpoint has a corresponding conditional setter (e.g., set_auth_uri_option()) that sets a conditional typestate (EndpointMaybeSet). When the conditional typestate is set, endpoints can be used via fallible methods that return Err(ConfigurationError::MissingUrl(_)) if an endpoint has not been set. This is useful in dynamic scenarios such as OpenID Connect Discovery, in which it cannot be determined until runtime whether an endpoint is configured.

There are three possible typestates, each implementing the EndpointState trait:

  • EndpointNotSet: the corresponding endpoint has not been set and cannot be used.
  • EndpointSet: the corresponding endpoint has been set and is ready to be used.
  • EndpointMaybeSet: the corresponding endpoint may have been set and can be used via fallible methods that return Result<_, ConfigurationError>.

The following code changes are required to support the new interface:

  1. Update calls to Client::new() to use the single-argument constructor (which accepts only a ClientId). Use the set_auth_uri(), set_token_uri(), and set_client_secret() methods to set the authorization endpoint, token endpoint, and client secret, respectively, if applicable to your application's auth flows.

  2. If required by your usage of the Client or BasicClient types (i.e., if you see related compiler errors), add the following generic parameters:

    HasAuthUrl: EndpointState,
HasDeviceAuthUrl: EndpointState,
HasIntrospectionUrl: EndpointState,
HasRevocationUrl: EndpointState,
HasTokenUrl: EndpointState,

    For example, if you store a BasicClient within another data type, you may need to annotate it as BasicClient<EndpointSet, EndpointNotSet, EndpointNotSet, EndpointNotSet, EndpointSet> if it has both an authorization endpoint and a token endpoint set. Compiler error messages will likely guide you to the appropriate combination of typestates.

    If, instead of using BasicClient, you are directly using Client with a different set of type parameters, you will need to append the five generic typestate parameters. For example, replace:

    type SpecialClient = Client<
    BasicErrorResponse,
    SpecialTokenResponse,
    BasicTokenType,
    BasicTokenIntrospectionResponse,
    StandardRevocableToken,
    BasicRevocationErrorResponse,
>;

    with:

    type SpecialClient<
    HasAuthUrl = EndpointNotSet,
    HasDeviceAuthUrl = EndpointNotSet,
    HasIntrospectionUrl = EndpointNotSet,
    HasRevocationUrl = EndpointNotSet,
    HasTokenUrl = EndpointNotSet,
> = Client<
    BasicErrorResponse,
    SpecialTokenResponse,
    BasicTokenType,
    BasicTokenIntrospectionResponse,
    StandardRevocableToken,
    BasicRevocationErrorResponse,
    HasAuthUrl,
    HasDeviceAuthUrl,
    HasIntrospectionUrl,
    HasRevocationUrl,
    HasTokenUrl,
>;

    The default values (= EndpointNotSet) are optional but often helpful since they will allow you to instantiate a client using SpecialClient::new() instead of having to specify SpecialClient::<EndpointNotSet, EndpointNotSet, EndpointNotSet, EndpointNotSet, EndpointNotSet>::new().

Rename endpoint getters and setters for consistency

The 4.0 release aimed to align the naming of each endpoint with the terminology used in the relevant RFC. For example, RFC 6749 uses the term "endpoint URI" to refer to the authorization and token endpoints, while RFC 7009 refers to the "token revocation endpoint URL," and RFC 7662 uses neither "URI" nor "URL" to describe the introspection endpoint. However, the renaming in 4.0 was both internally inconsistent, and inconsistent with the specs.

In 5.0, the Client's getters and setters for each endpoint are now named as follows:

  • Authorization endpoint: auth_uri()/set_auth_uri()
  • Token endpoint: token_uri()/set_token_uri()
  • Redirect: redirect_uri()/set_redirect_uri()
  • Revocation endpoint: revocation_url()/set_revocation_url()
  • Introspection endpoint: introspection_url()/set_introspection_url()
  • Device authorization endpoint: device_authorization_url()/set_device_authorization_url() (no change)

Use stateful HTTP clients

Previously, the HTTP clients provided by this crate were stateless. For example, the oauth2::reqwest::async_http_client() method would instantiate a new reqwest::Client for each request. This meant that TCP connections could not be reused across requests, and customizing HTTP clients (e.g., adding a custom request header to every request) was inconvenient.

The 5.0 release introduces two new traits: AsyncHttpClient and SyncHttpClient. Each request_async() and request() method now accepts a reference to a type that implements these traits, respectively, rather than a function type.

Warning

To prevent SSRF vulnerabilities, be sure to configure the HTTP client not to follow redirects. For example, use redirect::Policy::none when using reqwest, or redirects(0) when using ureq.

The AsyncHttpClient trait is implemented for the following types:

  • reqwest::Client (when the default reqwest feature is enabled)
  • Any function type that implements: 
    Fn(HttpRequest) -> F
where
  E: std::error::Error + 'static,
  F: Future<Output = Result<HttpResponse, E>>,
    To implement a custom asynchronous HTTP client, either directly implement the AsyncHttpClient trait, or use a function that implements the signature above.

The SyncHttpClient trait is implemented for the following types:

  • reqwest::blocking::Client (when the reqwest-blocking feature is enabled; see below)
  • ureq::Agent (when the ureq feature is enabled)
  • oauth2::CurlHttpClient (when the curl feature is enabled)
  • Any function type that implements: 
    Fn(HttpRequest) -> Result<HttpResponse, E>
where
  E: std::error::Error + 'static,
    To implement a custom synchronous HTTP client, either directly implement the SyncHttpClient trait, or use a function that implements the signature above.

Upgrade http to 1.0 and reqwest to 0.12

The 5.0 release of this crate depends on the new stable http 1.0 release, which affects various public interfaces. In particular, reqwest has been upgraded to 0.12, which uses http 1.0.

Enable the reqwest-blocking feature to use the synchronous reqwest HTTP client

In 4.0, enabling the (default) reqwest feature also enabled reqwest's blocking feature. To reduce dependencies and improve compilation speed, the reqwest feature now only enables reqwest's asynchronous (non-blocking) client. To use the synchronous (blocking) client, enable the reqwest-blocking feature in Cargo.toml:

oauth2 = { version = "5", features = ["reqwest-blocking" ] }

Use http::{Request, Response} for custom HTTP clients

The HttpRequest and HttpResponse structs have been replaced with type aliases to http::Request and http::Response, respectively. Custom HTTP clients will need to be updated to use the http types. See the reqwest client implementations for an example.

Import device code flow and token revocation types from the root

Previously, certain types were exported from both the root of the crate and the devicecode or revocation modules. These modules are no longer public, and their public types are exported from the root. For example, if you were previously importing oauth2::devicecode::DeviceAuthorizationResponse, instead import oauth2::DeviceAuthorizationResponse.

Replace TT generic type parameter in TokenResponse with associated type

Previously, the TokenResponse and TokenIntrospectionResponse traits had a generic type parameter TT: TokenType. This has been replaced with an associated type called TokenType. Uses of BasicTokenResponse and BasicTokenIntrospectionResponse should continue to work without changes, but custom implementations of either trait will need to be updated to replace the type parameter with an associated type.

Remove TT generic type parameter from Client and each *Request type

Removing the TT generic type parameter from TokenResponse (see above) made the TT parameters to Client and each *Request (e.g., CodeTokenRequest) redundant. Consequently, the TT parameter has been removed from each of these types. BasicClient should continue to work without any changes, but code that provides generic types for Client or any of the *Response types will need to be updated to remove the TT type parameter.

Add Display to ErrorResponse trait

To improve error messages, the RequestTokenError::ServerResponse enum variant now prints a message describing the server response using the Display trait. For most users (i.e., those using the default StandardErrorResponse), this does not require any code changes. However, users providing their own implementations of the ErrorResponse trait must now implement the Display trait. See StandardErrorResponse's Display implementation for an example.