Add a WebAssembly API

[turbolent]

Issue to be solved

As discussed in https://forum.onflow.org/t/idea-wasm-execution-engine-in-cadence/5164, add an API to Cadence that allows running WebAssembly programs.

Suggested Solution

An MVP should be able to:

• Instantiate a WebAssembly module, without imports (i.e. the WebAssembly module has no access to Cadence)
• Support function exports, where argument and return value types are i32 or i64 (i.e., not f32 or f64)
• Allow calling such function exports from Cadence, and returning the result back to Cadence
• Properly meter execution and abort execution if a given limit is reached. This requires defining how WebAssembly execution is metered in terms of the Cadence transaction's computation metering.

#2760 is a PoC implementation of the above. Remaining tasks:

Tasks

Beta Give feedback

1. Wrap user-errors produced by WebAssembly VM as Cadence user errors. Requires defining abstract user errors that are VM-agnostic. Define in interpreter/errors.go, as errors.UserError
2. Meter module instantiation, in terms of computation and memory. Requires getting remaining computation units and translating it
3. Define Limits
4. Validation?
5. Tests

[darkdrag00nv2]

Started looking into the metering aspects. Here is a brain dump for now. Might have misunderstood a few things, please correct me.

Computation

As @turbolent wrote in the poc #2760, we need a way to know the remaining computations.

FVM has ComputationAvailable(kind, amount) which returns true/false if there is enough computation units for the given kind. I think we can introduce RemainingComputation(kind) which returns the computation units remaining for the kind (or -1 for unbounded). For the current use case, the kind will just be ComputationKindWebAssemblyFuel but there can be future use-cases as well.

https://github.com/onflow/flow-go/blob/51202b7a9850a9728576355860b3279bdc1fe5bc/fvm/meter/computation_meter.go#L100-L111

Once have this, we just need a way to convert computation units to wasmtime fuel. Probably just a configurable multiplier would be sufficient?

---

Memory

We can limit the memory use via following:

1. Limit max stack size using SetMaxWasmStack

2. Limit linear memory, tables etc. using the Limiter

https://github.com/bytecodealliance/wasmtime-go/blob/da91f80fb2bec90f982107c127be18cfa328fd8e/store.go#L275-L281

Corresponding C documentation: https://docs.wasmtime.dev/c-api/store_8h.html#a0d8cfde96e219dd23d7c65f3cac2e15a.

[turbolent]

@darkdrag00nv2 I haven't looked into it myself yet, but your investigation looks good so far 👍

[darkdrag00nv2]

For tracking memory used, seems like we can just use the Size(store) function: https://github.com/bytecodealliance/wasmtime-go/blob/da91f80fb2bec90f982107c127be18cfa328fd8e/memory.go#L71

No, it gives the size of a linear memory.

[turbolent]

Opened onflow/flow-go#6215 to add a ComputationRemaining function to flow-go

[turbolent]

@onflow/flow-cadence-execution @janezpodhostnik How could we define the fuel-to-FVM-computation-usage factor/weight? How did we do this for EVM?