Update to Wasmtime 25 #2784
Update to Wasmtime 25 #2784
Conversation
alexcrichton
force-pushed
the
wasmtime25
branch
from
60c5ece
to
b75d39a
Compare
alexcrichton
force-pushed
the
wasmtime25
branch
from
b75d39a
to
c49e463
Compare
|
Ok Wasmtime 25 is now and this should be good to go 👍 |
| Scheme::Http, | ||
| Some(body), | ||
| )?; | ||
| let request = wasi_http.table().push(request)?; |
There was a problem hiding this comment.
Why are we manually creating the resource now instead of using WasiHttpView::new_incoming_request?
There was a problem hiding this comment.
Ah this is caused by a refactoring from a few versions ago in Wasmtime where the error type of the body is required to be hyper::Error for that helper but here it's ErrorCode from wasi-http instead. The Wasmtime helper will convert from hyper::Error to ErrorCode but there's not currently a constructor that works with ErrorCode itself. I tried to refactor Spin internals to use hyper::Error instead but ended up running into roadblocks around service chaining where hyper::Error was never materialized.
Perhaps the best solution would be to add a second helper or similar to the wasmtime-wasi-http crate though!
| .context("no fermyon:spin/inbound-redis instance found")?; | ||
| inbound_redis::Guest::new(&mut inbound_redis_export)? | ||
| }; | ||
| let guest_indices = inbound_redis::GuestIndices::new_instance(&mut store, &instance)?; |
There was a problem hiding this comment.
I'm not familiar with this GuestIndices API in wasmtime, and I'm having a hard time understanding why this is used over the Guest::new constructor. Can you explain this a bit?
There was a problem hiding this comment.
The short answer is that Guest::new doesn't exist any more. The longer answer is that Wasmtime's now refactored to better support exported interfaces at different versions than listed in the WIT (e.g. automatically handling 0.2.1-but-you-gave-me-0.2.0 and situations like that). That refactoring ended up moving some things around and this is the way this is loaded now.
The other slightly long-answer bit is that most of the bindings are intended to operate at the world level but this is operating at the individual interface level which means that the APIs used here are somewhat non-standard in the sense that the top-level world bits that bindgen! generates has all the helpers and individual interfaces are left to pull the pieces together individually. Perhaps something I can help refactor within Spin in the future?
alexcrichton
force-pushed
the
wasmtime25
branch
from
c49e463
to
08643c6
Compare
This commit updates to Wasmtime 25.0.0 and handles the various fallout and changes throughout to account for API changes made. Signed-off-by: Alex Crichton <[email protected]>
alexcrichton
force-pushed
the
wasmtime25
branch
from
08643c6
to
b151b31
Compare
This commit updates to Wasmtime 25.0.0 and handles the various fallout and changes throughout to account for API changes made.