Stream: git-wasmtime

Topic: wasmtime / PR #11763 support continuing host-to-host stre...

Wasmtime GitHub notifications bot (Sep 30 2025 at 15:11):

dicej requested alexcrichton for a review on PR #11763.

Wasmtime GitHub notifications bot (Sep 30 2025 at 15:11):

dicej opened PR #11763 from dicej:fix-11703 to bytecodealliance:main:

This addresses #11703 by allowing the embedder to "short-circuit" host-to-host streams and keep them flowing after the guest instance and its store have been disposed. Besides fixing that bug, it makes common proxy and full-duplex streaming scenarios more efficient by getting the Wasmtime runtime out of the way as early as possible, avoiding the extra memory and indirection overhead which would otherwise be imposed.

See the individual commit messages for details. Here's a summary:

• Add StreamReader::try_into for coercing a host-created stream into the StreamProducer implementation (or possibly a subset of it) that was used to create it. wasmtime-wasi-http now uses this to unmask guest bodies which are really host bodies.
• Use tokio::task::spawn instead of Accessor::spawn to handle connection I/O for outgoing wasmtime-wasi-http requests. This ensures that the connection remains alive after the store is disposed.

Note that the implementation of StreamReader::try_into required adding a default-implemented StreamProducer::try_into function for doing the coercion. The signature of that function is a bit awkward given that we have to erase the types and do runtime downcasts internally. The upside is that it supports the scenario where the StreamProducer implementation is !Unpin (meaning it cannot be downcasted, AFAICT), in which case it may be coerced into a type representing a relevant subset which is Unpin.

<!--
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 (Sep 30 2025 at 15:11):

dicej requested wasmtime-wasi-reviewers for a review on PR #11763.

Wasmtime GitHub notifications bot (Sep 30 2025 at 15:11):

dicej requested wasmtime-core-reviewers for a review on PR #11763.

Wasmtime GitHub notifications bot (Sep 30 2025 at 15:11):

dicej requested wasmtime-default-reviewers for a review on PR #11763.

Wasmtime GitHub notifications bot (Sep 30 2025 at 15:23):

dicej commented on PR #11763:

@rvolosatovs you might have opinions about https://github.com/bytecodealliance/wasmtime/pull/11763/commits/30afd9fa9e2a7adbf768c476a6119851e229ff5b

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:41):

alexcrichton submitted PR review.

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:41):

alexcrichton created PR review comment:

This Box::new(*...) shouldn't be necessary I think? In theory Pin::into_inner(me) should suffice here I believe

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:41):

alexcrichton created PR review comment:

Could this happen during the construction of Response instead of here? Then I believe this would naturally fall through to Body::Host below?

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:48):

dicej submitted PR review.

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:48):

dicej created PR review comment:

The compiler doesn't like that:

error[E0277]: `dyn std::any::Any` cannot be unpinned
   --> crates/wasi-http/src/p3/body.rs:543:32
    |
543 |             Ok(Pin::into_inner(me))
    |                --------------- ^^ the trait `std::marker::Unpin` is not implemented for `dyn std::any::Any`
    |                |
    |                required by a bound introduced by this call

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:54):

alexcrichton created PR review comment:

Weird! let me = Pin::into_inner(me); Ok(me) works though.

I think it's because it's trying to coerce me: Pin<Box<Self>> into Pin<Box<dyn Any>> and then go through Pin::into_inner to get out the Box<dyn Any> which doesn't work because dyn Any indeed does not implement Unpin. By separating it out it guides inference correctly.

Wasmtime GitHub notifications bot (Sep 30 2025 at 19:54):

alexcrichton submitted PR review.

Wasmtime GitHub notifications bot (Sep 30 2025 at 20:02):

dicej updated PR #11763.

Wasmtime GitHub notifications bot (Sep 30 2025 at 20:02):

dicej submitted PR review.

Wasmtime GitHub notifications bot (Sep 30 2025 at 20:02):

dicej created PR review comment:

Good idea; done!

Wasmtime GitHub notifications bot (Sep 30 2025 at 20:04):

dicej has enabled auto merge for PR #11763.

Wasmtime GitHub notifications bot (Sep 30 2025 at 20:38):

dicej merged PR #11763.

Last updated: Dec 06 2025 at 07:03 UTC