Stream: git-wasmtime

Topic: wasmtime / PR #7260 Add an asynchronous `wasi:io/output-s...

Wasmtime GitHub notifications bot (Oct 16 2023 at 23:05):

pchickey edited PR #7260:

Also add language to wasi:http/outgoing-body::finish explaining how it interacts with pending/in-process forwarding operations.

Together, these changes enable use cases where an instance forwards an incoming request's or response's body to an outgoing request/response and then terminates without waiting for the entire body to be streamed.

Wasmtime GitHub notifications bot (Oct 16 2023 at 23:08):

pchickey submitted PR review:

• how is write readiness of the output-stream defined - I believe it should not become ready for writing until all forwarded input-streams have been fully sent.
• if future-forwarded-result is dropped, what happens? I believe that we should continue to forward no matter what, but we need to make that explicit.
• if execution traps, what happens? I think in the case of a trap we should define forwarding to stop "at some point afterwards" - I don't think we want to be on the hook to keep forwarding after a trap, nor give a guarantee that it terminates promptly.

Wasmtime GitHub notifications bot (Oct 16 2023 at 23:08):

pchickey submitted PR review:

• how is write readiness of the output-stream defined - I believe it should not become ready for writing until all forwarded input-streams have been fully sent.
• if future-forwarded-result is dropped, what happens? I believe that we should continue to forward no matter what, but we need to make that explicit.
• if execution traps, what happens? I think in the case of a trap we should define forwarding to stop "at some point afterwards" - I don't think we want to be on the hook to keep forwarding after a trap, nor give a guarantee that it terminates promptly.

Wasmtime GitHub notifications bot (Oct 16 2023 at 23:08):

pchickey created PR review comment:

I think this should return result<future-forwarded-result, tuple<stream-error, input-stream>> - returns ownership of the input stream on failure, and says that this output-stream has already closed or somehow failed.

Wasmtime GitHub notifications bot (Oct 16 2023 at 23:08):

pchickey created PR review comment:

I don't think we should provide multiple input-streams here, there is only one that has been taken ownership of by the forward function so lets only give one back here. The caller can hold onto handles for all of their future-forward-results and unwrap each of them into an some(ok(err((_error, input-stream))))

Wasmtime GitHub notifications bot (Nov 09 2023 at 18:51):

alexcrichton closed without merge PR #7260.

Last updated: Nov 22 2024 at 16:03 UTC