Skip to content

Refactor WASIp2 wasi:http implementation #12748

Open
alexcrichton wants to merge 7 commits intobytecodealliance:mainfrom
alexcrichton:refactor-wasi-http
Open

Refactor WASIp2 wasi:http implementation #12748
alexcrichton wants to merge 7 commits intobytecodealliance:mainfrom
alexcrichton:refactor-wasi-http

Conversation

@alexcrichton
Copy link
Member

This PR is a refactoring and reorganization mostly around WASIp2's implementation of wasi:http. Notably I am trying to get the implementations in this crate to look more like wasmtime-wasi with changes such as:

  • There's a top-level p2 and p3 module for each respective implementation.
  • There's a WasiHttpView trait which projects out WasiHttpCtxView for both WASI versions
  • Both WASI versions have access to the same WasiHttpCtx structure
  • Eventually there will be a single WasiHttpView and WasiHttpCtxView trait/struct (that's not done just yet)

This PR is a step in this direction. There are two commits here which are doing similar things but are still separate. The first moves all the WASIp2 bits under a p2 top-level module. The second applies many structural refactorings to resolve the remaining bullets above.

Future work here includes accounting for #12674, working to have a single WasiHttpHooks trait (and thus a single WasiHttpView trait/WasiHttpCtxView struct), and updating idioms around WASIp2 to look more like WASIp3 (e.g. similar-looking embedder-style APIs).

This mirrors the `wasmtime-wasi` crate's organization where there's a
`p2` module and a `p3` module at the top level.
This commit reorganizes and refactors the WASIp2 implementation of
`wasi:http` to look more like other `wasmtime-wasi`-style interfaces.
Specifically the old `WasiHttpImpl<T>` structure is removed in favor of
as `WasiHttpCtxView<'_>` type that is used to implement
bindgen-generated `Host` traits. This necessitated reorganizing the
methods of the previous `WasiHttpView` trait like so:

* The `WasiHttpView` trait is renamed to `WasiHttpHooks` to make space
  for a new `WasiHttpView` which behaves like `WasiView`, for example.

* The `ctx` and `table` methods of `WasiHttpHooks` were removed since
  they'll be fields in `WasiHttpCtxView`.

* Helper methods for WASIp2 were moved to methods on `WasiHttpCtxView`
  instead of default methods on `WasiHttpHooks`.

With these changes in place the WASIp3 organization was also updated
slightly as well. Notably WASIp3 now contains a reference to the crate's
`WasiHttpCtx` structure (which has field limits for example). WASIp3's
previous `WasiHttpCtx` trait is now renamed to `WasiHttpHooks` as well.
This means that there are two `WasiHttpHooks` traits right now, one for
WASIp2 and one for WASIp3. In the future I would like to unify these two
but that will require some more work around the default `send_request`.

A final note here is that the `WasiHttpHooks` trait previously, and
continues to be, optional for embedders to implement. Default functions
are provided as `wasmtime_wasi_http::{p2, p3}::default_hooks`.
Additionally there's a `Default for &mut dyn WasiHttpHooks`
implementation, too.

With all that outlined: the motivation for this change is to bring the
WASIp2 and WASIp3 implementations of `wasi:http` closer together. This
is inspired by refactorings I was doing for bytecodealliance#12674 to apply the same
header limitations for WASIp3 as is done for WASIp2. Prior to this
change there were a number of differences such as WASIp3 not having
`crate::WasiHttpCtx` around, WASIp2 having a different organization of
structures/borrows, etc. The goal is to bring the two implementations
closer in line with each other to make refactoring across them more
consistent and easier.
@alexcrichton alexcrichton requested a review from dicej March 9, 2026 21:24
@alexcrichton alexcrichton requested review from a team as code owners March 9, 2026 21:24
@github-actions github-actions bot added the wasi Issues pertaining to WASI label Mar 10, 2026

#[doc(hidden)]
#[cfg(feature = "default-send-request")]
impl WasiHttpHooks for [(); 0] {}
Copy link
Member

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Out of curiosity, is there a reason to use empty array here as opposed to e.g. a unit?

Copy link
Member Author

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

While I thought it was possible to return &mut ZST, apparently that's only implemented for empty arrays in rustc rather than all empty types -- hence the array here. It was the only way I could think of to safely get a &'static mut T to implement the trait on.

@alexcrichton alexcrichton added this pull request to the merge queue Mar 10, 2026
@github-merge-queue github-merge-queue bot removed this pull request from the merge queue due to failed status checks Mar 10, 2026
@alexcrichton alexcrichton enabled auto-merge March 10, 2026 16:10
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

wasi Issues pertaining to WASI

Projects

None yet

Development

Successfully merging this pull request may close these issues.

3 participants