Changes to circuit sizes

Generated at commit: 819490a00e5cb4de47a754b86131456f26bdeeb9, compared to commit: ade69a9e5f1546249e9b43b40e9ff0da87c4632e

🧾 Summary (10% most significant diffs)

Circuit sizes went up, as expected. In the ones mostly affected I think that's expected. For example array_dynamic_main_output is setting an array value with an unknown index. Previously there was no bounds check and now there is.

But in some cases there are checks that could be removed. In array_sort I can find this:

    v22 = lt v4, u32 3
    jmpif v22 then: b2, else: b3
  b2():
    v24 = lt v4, u32 3
    constrain v24 == u1 1 '"Index out of bounds"'
    v25 = array_get v0, index v4

It's checking the < 3 condition twice. I don't know if that's not optimized away because that's inside an unconstrained function, or because that optimization doesn't exist yet.

It's also interesting what happened with nested_array_dynamic_simple:

fn main(x: Field) {
    // x = 3
    let array: [[(Field, [Field; 1], [Field; 1]); 1]; 1] = [[(1, [2], [3])]];

    let fetched_value = array[x - 3];
    assert(fetched_value[0].0 == 1);
    assert(fetched_value[0].1[0] == 2);
    assert(fetched_value[0].2[0] == 3);
}

the compiler notices that the array only has one value, so the assert checks will always succeed as long as x - 3 is zero. The resulting program (in this PR) is this one:

acir(inline) fn main f0 {
  b0(v0: Field):
    v47 = sub v0, Field 3
    v48 = cast v47 as u32
    constrain v48 == u32 0 '"Index out of bounds"'
    return
}

The old program was much longer... I don't know if constrain enables some further optimizations.

Closing because we chose #5691 instead.