diff --git a/design/src/SUMMARY.md b/design/src/SUMMARY.md index 34933fca19701ba90c9ba902bdf65d6b53ac0739..a8277d72f382f443a39641c5cef5838951cfef69 100644 --- a/design/src/SUMMARY.md +++ b/design/src/SUMMARY.md @@ -61,6 +61,8 @@ - [RFC-0032: Better Constraint Violations](./rfcs/rfc0032_better_constraint_violations.md) - [RFC-0033: Improving access to request IDs in SDK clients](./rfcs/rfc0033_improve_sdk_request_id_access.md) - [RFC-0034: Smithy Orchestrator](./rfcs/rfc0034_smithy_orchestrator.md) + - [RFC-0035: Collection Defaults](./rfcs/rfc0035_collection_defaults.md) + - [RFC-0036: HTTP Dependency Exposure](./rfcs/rfc0036_http_dep_elimination.md) - [Contributing](./contributing/overview.md) - [Writing and debugging a low-level feature that relies on HTTP](./contributing/writing_and_debugging_a_low-level_feature_that_relies_on_HTTP.md) diff --git a/design/src/rfcs/rfc0035_collection_defaults.md b/design/src/rfcs/rfc0035_collection_defaults.md index a1aa389b4e5857c86d85091dcc0807e0cfc10cd5..218782f789489154207ea93e26a31449a4862099 100644 --- a/design/src/rfcs/rfc0035_collection_defaults.md +++ b/design/src/rfcs/rfc0035_collection_defaults.md @@ -3,7 +3,7 @@ RFC: Collection Defaults ============= -> Status: RFC +> Status: Accepted > > Applies to: client @@ -12,14 +12,14 @@ For a summarized list of proposed changes, see the [Changes Checklist](#changes- This RFC proposes a **breaking change** to how generated clients automatically provide default values for collections. Currently the SDK generated fields for `List` generate optional values: -```rust +```rust,ignore ///

Container for elements related to a particular part. pub fn parts(&self) -> Option<&[crate::types::Part]> { self.parts.as_deref() } ``` This is _almost never_ what users want and leads to code noise when using collections: -```rust +```rust,ignore async fn get_builds() { let project = codebuild .list_builds_for_project() @@ -50,7 +50,7 @@ In the current version of the SDK, users must call `.unwrap_or_default()` freque Once this RFC is implemented, users will be able to use these accessors directly. In the rare case where users need to distinguish be `None` and `[]`, we will direct users towards `model..is_some()`. -```rust +```rust,ignore async fn get_builds() { let project = codebuild .list_builds_for_project() diff --git a/design/src/rfcs/rfc0036_http_dep_elimination.md b/design/src/rfcs/rfc0036_http_dep_elimination.md index 355d0f92ddb19c57c824f041092e5c38c44d7c12..5c47d680fa5d4618f40f9eb5a30cdf4dbff654b7 100644 --- a/design/src/rfcs/rfc0036_http_dep_elimination.md +++ b/design/src/rfcs/rfc0036_http_dep_elimination.md @@ -2,7 +2,7 @@ RFC: Eliminating Public `http` dependencies ============= -> Status: RFC +> Status: Accepted > > Applies to: client @@ -24,7 +24,7 @@ Terminology to operate on frames instead of having separate methods. - `http` (crate): a low level crate of `http` primitives (no logic, just requests and responses) - ossified dependency: An ossified dependency describes a dependency that, when a new version is released, cannot be utilized without breaking changes. For example, if the `mutate_request` function on every operation operates on `&mut http::Request` where `http = 0.2`, that dependency is "ossified." Compare this to a function that offers the ability to convert something into an `http = 0.2` request—since http=1 and http=0.2 are largely equivalent, the - existence of this function does not prevent us from using http = 1 in the future. In general terms, **functions that operate on references are much more likely to ossify—There is no practical way for someone to mutate an `http = 0.2` request if you have an `http = 1` request other than a time-consuming clone, and reconversion process. + existence of this function does not prevent us from using http = 1 in the future. In general terms, **functions that operate on references are much more likely to ossify**—There is no practical way for someone to mutate an `http = 0.2` request if you have an `http = 1` request other than a time-consuming clone, and reconversion process. @@ -42,6 +42,20 @@ If we're still on `http = 0.*` and a vulnerability is identified, we may end up **API Friendliness** If we ship with an API that public exposes customers to `http = 0.*`, we have the API forever. We have to consider that we aren't shipping the Rust SDK for this month or even this year but probably the Rust SDK for the next 5-10 years. +**Future CRT Usage** +If we make this change, we enable a future where we can use the CRT HTTP request type natively without needing a last minute conversion to the CRT HTTP Request type. +```rust,ignore +struct HttpRequest { + inner: Inner +} + +enum Inner { + Httpv0(http_0::Request), + Httpv1(http_1::Request), + Crt(aws_crt_http::Request) +} +``` + The user experience if this RFC is implemented ---------------------------------------------- Customers are impacted in 3 main locations: @@ -70,7 +84,7 @@ One key mechanism that we SHOULD use for allowing our APIs to evolve in the futu In order to enable HTTP evolution, we will create a set of wrapper structures around `http::Request` and `http::Response`. These will use `http = 0` internally. Since the HTTP crate itself is quite small, including private dependencies on both versions of the crate is a workable solution. In general, we will aim for an API that is close to drop-in compatible to the HTTP crate while ensuring that a different crate could be used as the backing storage. -```rust +```rust,ignore // since it's our type, we can default `SdkBody` pub struct Request { // this uses the http = 0.2 request. In the future, we can make an internal enum to allow storing an http = 1 @@ -92,8 +106,6 @@ The SigV4 crate signs a number of `HTTP` types directly. We should change it to Generated clients currently include a public HTTP dependency in `customize`. This should be changed to accept our `HTTP` wrapper type instead or be restricted to a subset of operations (e.g. `add_header`) while forcing users to add an interceptor if they need full control. -In order to implement this feature, we need to add X and update Y... - Changes checklist ----------------- @@ -102,7 +114,7 @@ Changes checklist - [ ] Refactor currently written interceptors to use the wrapper: 2 days. - [ ] Refactor the SigV4 crate to remove the HTTP dependency from the public interface: 2 days. - [ ] Add / validate support for SdkBody `http-body = 1.0rc.2` either in a PR or behind a feature gate. Test this to - ensure it works with Hyper. + ensure it works with Hyper. Some previous work here exists: 1 week - [ ] Remove `http::Response` and `Uri` from the public exposed types in `aws-config`: 1-4 days. - [ ] Long tail of other usages: 1 week - [ ] Implement `~` versions for SDK Crate => runtime crate dependencies: 1 week