feat: handle empty response foreign calls without an external resolver

[TomAFrench]

Description

Problem*

Resolves

Summary*

I've modified the DefaultForeignCallExecutor so that rather than panicking on an unrecognised foreign call we return an empty response.

This solves an issue experienced by @spalladino where nargo test was failing due to the use of a custom debugging oracle which doesn't return any values to the circuit. The only situation where these oracle calls not being handled would change the outcome of the test would be if they modify state in an external resolver such as to modify the response of a different oracle call (and so we'd still fail on those anyway).

Additional Context

I've cleared up some empty Prover.toml files as well.

Documentation*

Check one:

• No documentation needed.
• Documentation included in this PR.
• [For Experimental Features] Documentation to be submitted in a separate PR.

PR Checklist*

• I have tested the changes locally.
• I have formatted the changes with Prettier and/or cargo fmt on default settings.

[spalladino]

Thanks @TomAFrench! I'm worried this will make failures more obscure though, since if a user now forgets to set up a resolver, they'll get an ACVM error rather than an error related to a foreign call. Or do you think the stack trace should be good enough to identify the cause of the error?

[TomAFrench]

The callstack leading to the oracle call should be sufficient for users to determine the error imo.

[jfecher]

A few empty files were committed as well

The callstack leading to the oracle call should be sufficient for users to determine the error imo.

This doesn't sit right with me. If there's an error in the program we should tell users directly what the error is and not have them rely on context information. Wouldn't the correct fix here just be to have the custom debugging oracle be present in the test or otherwise removed from it?

[TomAFrench]

Wouldn't the correct fix here just be to have the custom debugging oracle be present in the test or otherwise removed from it?

I'm assuming you mean "not be present in the test"? I agree in general but in this case these oracle calls are spread throughout aztec-nr and so it's not easy to remove them solely from the tests without affecting the actual code.

[jfecher]

This sounds like an aztec-nr specific problem then. I wonder if we can at least issue a warning like "oracle xxx was not executed" or similar. If an oracle has no return values then it must have been placed for its side effects in general so I don't think they should necessarily be skipped without warning.
Or maybe we can introduce a boolean value used for checking if nargo test is being run? Then the library could change the calls to if !NARGO_TEST { debug(); }

[TomAFrench]

If an oracle has no return values then it must have been placed for its side effects in general so I don't think they should necessarily be skipped without warning.

That's true but these side effects are entirely optional. I should be able to execute the circuit while ignoring those side effects if I want to. If I don't specify an oracle resolver then by default I'm saying that I don't want to hook up the ACVM to an outside system

[TomAFrench]

Or maybe we can introduce a boolean value used for checking if nargo test is being run?

Conditional compilation is def something we can look at at some point. I think it's valid to want to be able to execute a circuit without tracking any purely outwards oracle calls though.

[TomAFrench]

My ideal outcome for this is if the ABI we're aware of the foreign calls which the program can make including the type information for arguments/return values. At that point we could distinguish the calls which need return values and throw an error in Nargo rather than the ACVM but I really don't see the current error message as that bad.

[TomAFrench]

For context, here's the error that's being shown.

error: Assertion failed: '0 output values were provided as a foreign call result for 1 destination slots'
   ┌─ /home/tom/Programming/aztec/noir/test_programs/noir_test_success/ignored_oracle/src/main.nr:22:13
   │
22 │     let _ = custom_getter();
   │             ---------------
   │
   = Call stack:
     1. /home/tom/Programming/aztec/noir/test_programs/noir_test_success/ignored_oracle/src/main.nr:22:13

[jfecher]

I'm still not sure about this to be honest but I think we can at least merge it to unblock @spalladino and anyone else who needs it

[TomAFrench]

I agree that it's not perfect but it's replacing a panic which is arguably worse.