fix(ssa): Empty array of higher order functions

[vezenovm]

Description

Problem*

Builds upon #8674

Resolves the final panic in the linked issue that is slightly separate from lambdas in arrays:

fn main() {
    let lambdas: [fn(()) -> (); 0] = [];
    lambdas[0](());
}

Summary*

This is the SSA for the above snippet pre-DIE:

acir(inline) fn main f0 {
  b0():
    v0 = make_array [] : [Field; 0]
    constrain u1 0 == u1 1, "Index out of bounds"
    v4 = array_get v0, index u32 0 -> Field
    call v4()
    return
}

If we leave call v4() this leads to issues in the underconstrained check and then during ACIR as function pointers are not supported. Also, the fact that we still have a function as a value means that there are no function variants and we should be able to safely remove this function.

We now have this final SSA:

acir(inline) predicate_pure fn main f0 {
  b0():
    constrain u1 0 == u1 1, "Index out of bounds"
    return
}

This matches the final SSA and behavior of an empty array:

fn main() {
    let x = [];
    x[0];
}

Additional Context

I'm not really sure in what other case we may have a function value which is an instruction. This seems like the only case. However, I'm not a huge fan of this solution as it makes assumptions across passes. I think a better solution would be to simply error out on accessing an empty array at compile-time #8676.

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 'Test Suite Duration'.
Benchmark result of this commit is worse than the previous benchmark result exceeding threshold 1.20.

Benchmark suite	Current: 86b7da9	Previous: eb3eef9	Ratio
test_report_zkpassport_noir_rsa_	3 s	2 s	1.50

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

CC: @TomAFrench

[jfecher]

I'm not 100% on this. Maybe we could add an assertion that the instruction originates from an array get where the array is of length zero? Although if it is a slice maybe we won't know the length?

[vezenovm]

Yeah we could check the length of the array in an array get but it breaks with slices. With slices we would have to track backwards to see whether it originated from an empty make array (or track slice capacities as we do in RemoveIfElse), which I think would be overkill.

I'm also not a big fan of this solution. Even with #8676, if we eventually allow escaping unconditional panics this issue will arise again. I'll think of something else. Going to switch this back to a draft.

[vezenovm]

Here is an alternative #8697 which I prefer.

[vezenovm]

Superceded by #8697