Stream: git-wasmtime

Topic: wasmtime / PR #11238 Start to use `&Accessor<T, D>` more ...

Wasmtime GitHub notifications bot (Jul 14 2025 at 20:10):

alexcrichton opened PR #11238 from alexcrichton:accessor-apis to bytecodealliance:main:

After discussion with Joel we've concluded that while &mut Accessor<T, D> was originally added to model host functions it is also appropriate to use it to model embedder-rooted invocations of items such as wasm as well. Effectively the conclusion we reached was that *::call_concurrent should be taking &Accessor, not StoreContextMut. This has a number of benefits to it over the previous iteration:

• This makes exports behave more like imports where Accessor means "you're in the concurrent world".

• This makes exports have an async fn signature which is easier to read and understand.

• This automatically enforces the guarantee that the returned future is only polled within the main event loop because the future is always considered to close over the &Accessor provided meaning it statically cannot live outside of the event loop.

• This paves the way forward to future refactorings to avoid storing so much state within a Store<T> and instead try to store state directly in futures themselves. This should make cancellation more natural and eventually also remove 'static bounds on params/results. Furthermore this should make it easier to avoid spawning tasks internally by storing state in futures instead of spawned tasks.

In doing this one of the main questions we were faced with was what to do about &mut Accessor<T, D>, namely the mut part. With a mutable accessor it would be only possible to call one function concurrently. One option considered was to add combinators like Accessor::join and Accessor::race but in the end we decided to avoid going that direction and instead switch to &Accessor<T, D> everywhere, freely enabling aliasing of the accessor. This has the downside that Accessor::with is now a relatively dangerous function in that it can panic, but idiomatic usage of it is not expected to panic as the distinction between the async and sync boundary of Accessor vs StoreContextMut is expected to naturally make the recursive panic condition of with rare to come up in practice.

Concrete changes in this commit are:

• Accessor::with now requires &self.
• Accessor::spawn now requires &self.
• Host functions are now given &Accessor, not &mut Accessor.
• {Typed,}Func::call_concurrent is now an async fn which takes an &Accessor instead of StoreContextMut.
• Guest bindings generation for concurrent invocations now looks exactly like async bindings generation except for replacing StoreContextMut with Accessor.

Note that this commit does not yet update the internal implementations of these functions to benefit from the new abilities that taking &Accessor implies. Instead that's deferred to a future update as necessary. Instead this is only updating the public API of the wasmtime crate to enable these refactorings in the future.

Also note that this does not yet update all functions to take &Accessor. Notably futures and streams still need to be updated.

cc #11224

<!--
Please make sure you include the following information:

• If this work has been discussed elsewhere, please include a link to that
  conversation. If it was discussed in an issue, just mention "issue #...".

• Explain why this change is needed. If the details are in an issue already,
  this can be brief.

Our development process is documented in the Wasmtime book:
https://docs.wasmtime.dev/contributing-development-process.html

Please ensure all communication follows the code of conduct:
https://github.com/bytecodealliance/wasmtime/blob/main/CODE_OF_CONDUCT.md
-->

Wasmtime GitHub notifications bot (Jul 14 2025 at 20:10):

alexcrichton requested dicej for a review on PR #11238.

Wasmtime GitHub notifications bot (Jul 14 2025 at 20:10):

alexcrichton requested wasmtime-core-reviewers for a review on PR #11238.

Wasmtime GitHub notifications bot (Jul 14 2025 at 20:11):

alexcrichton commented on PR #11238:

One other update: Instance::run_with is updated to now take AyncFnOnce instead of FnOnce() -> Box<...>

Wasmtime GitHub notifications bot (Jul 15 2025 at 14:51):

dicej created PR review comment:

Could collapse the if expression here into ("async", "_async", ".await", self.generator.opts.concurrent_exports)

Wasmtime GitHub notifications bot (Jul 15 2025 at 14:51):

dicej submitted PR review.

Wasmtime GitHub notifications bot (Jul 15 2025 at 15:00):

alexcrichton updated PR #11238.

Wasmtime GitHub notifications bot (Jul 15 2025 at 15:00):

alexcrichton has enabled auto merge for PR #11238.

Wasmtime GitHub notifications bot (Jul 15 2025 at 15:38):

alexcrichton merged PR #11238.

Last updated: Dec 06 2025 at 06:05 UTC