fix: Handle &mut function in defunctionalize

[aakoshh]

Description

Problem*

Resolves #8662

Summary*

The defunctionalize pass did not handle &mut function parameters, just function. OTOH the post-check looked for function references and failed. This didn't cause a problem on CI because the assertion only works in debug mode, and CI runs in release mode. Only by accident did we find that using cargo run to try to compile one of the contracts in aztec-packages fails due to the post check.

Additional Context

The AST fuzzer did not find this case because it doesn't generate reference parameters yet.

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.

[github-actions]

⚠️ Performance Alert ⚠️

Possible performance regression was detected for benchmark 'Execution Time'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: e59311a	Previous: c37ded9	Ratio
semaphore-depth-10	0.025 s	0.019 s	1.32

This comment was automatically generated by workflow using github-action-benchmark.

CC: @TomAFrench

[aakoshh]

My changes to the defunctionalization resulted in the expected SSA string, but later during inlining it turned out that the value of it was wrong, causing a crash during Brillig gen.

The fix was this: 642e660
But the unit test doesn't capture the problem.

I tested the crash with this command:

$ cargo run -q -p nargo_cli -- --program-dir /Users/aakoshh/Work/aztec/aztec-packages/noir-projects/noir-contracts/contracts/app/easy_private_voting compile --silence-warnings --package easy_private_voting_contract --show-ssa
...
The application panicked (crashed).
Message:  internal error: entered unreachable code: unsupported function call type NumericConstant { constant: 33, typ: NativeField }
Location: compiler/noirc_evaluator/src/brillig/brillig_gen/brillig_block.rs:766

There is a part which before defunctionalization looks like this:

brillig(inline) fn add_to_tally_public f0 {
  b0(v2: Field):
    ...
    v48, v49, v50, v51 = call f10(v30, v31, v32, v33, v34, v2) -> (&mut u1, &mut Field, &mut function, Field)

brillig(inline) fn at f10 {
  b0(v2: &mut u1, v3: &mut Field, v4: &mut function, v5: Field, v6: function, v7: Field):
    v9 = call f16(v5, v7) -> Field
    v10, v11, v12, v13 = call v6(v2, v3, v4, v9) -> (&mut u1, &mut Field, &mut function, Field)
    return v10, v11, v12, v13
}

after defunctionalization:

brillig(inline) fn at f10 {
  b0(v2: &mut u1, v3: &mut Field, v4: &mut Field, v5: Field, v6: Field, v7: Field):
    v9 = call f16(v5, v7) -> Field
    v10, v11, v12, v13 = call v6(v2, v3, v4, v9) -> (&mut u1, &mut Field, Field, Field)
    return v10, v11, v12, v13
}

but after inlining:

brillig(inline) fn add_to_tally_public f0 {
  b0(v2: Field):
    ...
    v29, v30, v31, v32 = call Field 33(v3, v5, v7, v27) -> (&mut u1, &mut Field, Field, Field)

and then it crashes because it's trying to call Field 33 instead of f33 which it would if the defunctionalization worked correctly.

Unfortunately my minimised hand written SSA didn't try to call v6, even before the fix it was correctly forwarding to f33. In the interest of moving forward with the fix I gave up trying to capture the exact circumstances for now and just happy that the external tests would not have let this pass.

I'll try to extend the AST fuzzer to use references to cover this case.

[aakoshh]

I'll try to replicate it as an integration test as well next week.