Point Addition in Unentered else Block Causes Circuit Failure

[LHerskind]

Aim

A point addition inside an else that is NOT "entered" will cause the circuit to fail entirely. The simulator seems to apply the checks when generating the witness and if using 0 as the base value just fails.

I encountered this when having a computation that is only to happen when we have non-zero values, but anyway fails even if unused. A minimal test to reproduce is here.

#[test]
fn weird_add() {
    let a = EmbeddedCurvePoint {
        x: 0x1d8eb4378a3bde41e0b6a9a8dcbd21b7ff9c51bdd6ca13ce989abbbf90df3666,
        y: 0x06075b63354f2504f9cddba0b94ed0cef35fc88615e69ec1f853b51eb79a24a0
    };

    let aa = a.double();

    if aa.x == a.add(a).x {
        println(f"We are not trying to do a double of zero");
    } else {
        println("We are trying to do a double of zero");
        let d = EmbeddedCurvePoint { x: 0, y: 0 }.double();
        println(f"d*2 {d}");
    }
}

Similar failure is possible with the multi_scalar_mul and fixed_base_scalar_mul if providing fields with values > 128 bits inside the if.

Expected Behavior

I expected to see

We are not trying to do a double of zero

And have the test pass.

Bug

Instead the test fails with the following log.

We are not trying to do a double of zero
[noir_aztec] Testing encrypted_logs::payload::weird_add... FAIL
Failed to solve program: 'Failed to solve blackbox function: embedded_curve_add, reason: Point (0000000000000000000000000000000000000000000000000000000000000000, 0000000000000000000000000000000000000000000000000000000000000000) is not on curve'

To Reproduce

1. Run the test
2. Profit???

Project Impact

Blocker

Impact Context

We need to use the computation to constrain log production in Aztec.

Workaround

None

Workaround Description

No response

Additional Context

No response

Installation Method

None

Nargo Version

No response

NoirJS Version

No response

Would you like to submit a PR for this Issue?

None

Support Needs

No response

[TomAFrench]

This is addressed by AztecProtocol/aztec-packages#6384

Dupe of #4978

[TomAFrench]

Reopening as this is actually different and won't be addressed in that PR.

[TomAFrench]

We'll add some behaviour in the flattening pass to apply predicates to these inputs in if statements. This will be blocked pending being able to represent the point at infinity for now though.