feat: implement Resource<T> -> ResourceAny conversion

[rvolosatovs]

Closes #7676
Follow-up on #7680

This feature requires changes in the resource table to allow specifying the ResourceType associated with the resource stored (since assumption that it's always equal to ResourceType::host::<T>() is no longer valid)

Add support for Resource<T> -> ResourceAny conversion to allow invoking component functions without bindings generated at compile-time.

[alexcrichton]

I think I might have missed something in this by accident perhaps. I don't think that this is the direction things should go in, for example Resource<T> is designed to have a statically known type as ResourceType::host::<T>() as opposed to a runtime-defined type. Having a runtime-defined type sort of defeats the purpose of the type parameter which at that point require a redesign of the resource-related APIs.

The title of this PR though, allowing host creation of guest resources, is not something Wasmtime can enable. Just as guests can't create host-defined resources without asking the host to do so the other way around is the same. The host can't conjure a guest resource out of thin air, it must originate in the guest.

Given that I fear I might have missed some motivation/use case when talking about this, could you perhaps clarify?

[rvolosatovs]

Given that I fear I might have missed some motivation/use case when talking about this, could you perhaps clarify?

Here's a more concrete example of the use case:

Assume the following WIT:

interface types {
    use wasi:io/poll@0.2.0-rc-2023-11-10.{pollable};

    resource future-string {
        subscribe: func() -> pollable;
        get: func() -> option<string>;
    }
}

interface example {
    use types.{future-string};

    foo: func(opaque: future-string);
}

world opaque-world {
    export example;
}

And a component exporting the example interface.

The intention is to call example.foo on the component without having the WIT at compilation time, i.e. parsing the WIT from the Wasm itself at runtime (which works already) and then invoking example.foo within the component.
The only blocker for this now is the fact that the host cannot construct future-string at runtime if it does not have the future-string WIT definition available at compilation time (and associated bindings)

The host, however, can already push a Resource to the table and link the associated methods on the linker, the only bit left is being able to turn the Resource constructed at runtime into a ResourceAny to be able to invoke the function via call or call_async

[rvolosatovs]

Another way to look at this would be the following:

• If one could pass a wasmtime::component::Val::U32, which stores the Resource::rep value as a function parameter directly where a resource is expected, that would cover the use case.

Perhaps an even more generic view on this could be allowing arbitrary impl Lower values to be passed as arguments to call/call_async (or yet another alternative)

This all really boils down to being able to pass the resource index to a component in cases where the resource type is not known at compilation time

[github-actions]

Subscribe to Label Action

cc @peterhuene

Details

This issue or pull request has been labeled: "wasi", "wasmtime:api"

Thus the following users have been cc'd because of the following labels:

• peterhuene: wasmtime:api

To subscribe or unsubscribe from this label, edit the .github/subscribe-to-label.json configuration file.

Learn more.

[rvolosatovs]

Small update: I found a way to obtain the TypeResourceTableIndex from the instance, which simplifies getting the destructor and flags, however I keep running into

    type mismatch for own, possibly due to mixing distinct composite types

I will reach-out offline

[rvolosatovs]

@alexcrichton , as discussed, I introduced a PathIndex mapping usizes to Vec<usize> import paths.
I did not attempt to work on optimizing this, but rather focused on getting something simple over the line.
I added the PathIndex to the NameMap value, since in order to use these indexes a reverse mapping from Vec<usize> import path to PathIndex is required

[alexcrichton]

Thanks again for this! A few comments which I'm happy to defer to later, but for the implementation of LinkerInstance::resource I think it'd be good to continue to rely on the insert helper. I wouldn't worry too much about the interaction of bumping the resource counter if an error is returned.

[sehz]

Any progress? This is blocker for us

[rvolosatovs]

Any progress? This is blocker for us

Hey @sehz,

There was a doc CI failure:

error: unresolved link to Linker::resource
--> crates/wasmtime/src/component/resources.rs:519:60
|
519 | /// idx is the [ResourceImportIndex] returned by [Linker::resource].
| ^^^^^^^^^^^^^^^^ no item named Linker in scope

warning: cranelift-jit (lib doc) generated 1 warning (1 duplicate)
Documenting wasmtime-cranelift-shared v17.0.0 (/home/runner/work/wasmtime/wasmtime/crates/cranelift-shared)
warning: wasmtime (lib doc) generated 1 warning (1 duplicate)
error: could not document wasmtime

Which, I believe, is now fixed. I've rebased and added prtest:full(not sure if a scoped option exists) to run all CI checks on this PR, once that succeeds it should be ready for merge.

Please note that I'm generally out of office until Monday, the 8th of January, but I may be able to address small things like this doc CI failure in the meantime