SSA Refactor: Remove tech debt when moving to the SSA refactor

[vezenovm]

Problem

1. PR feat(ssa_refactor)!: Add Slices #1728 implements syntax for slices into Noir. However, as we are moving to the new SSA pass, the PR only implements slices in the new pass while breaking changes were made to the frontend in order to get slices fully working. This means that the experimental_ssa flag is now needed not only to determine which SSA to generate but also during compilation.

2. There are features aside slices that touch the frontend but are only implemented using the new SSA. For example after the refactor in PR feat: Refactor Logging to use Brillig foreign calls #1917 we have two different implementations of println. One that uses the old refactor and one that uses Brillig foreign calls. The experimental_ssa flag is used again here to determine whether we should use one println or the other.

3. We have unneeded transformations like the array of structs to struct of arrays conversion during monomorphization that can be removed once the old ssa code is removed. Removing this would lead to a more intuitive serialization order when going back and forth between Noir and TypeScript.

Happy Case

The enable_slices flag inside of the NodeInterner should be fully removed and all of its conditional checks within the frontend. The array of structs to struct of arrays conversion should also be removed.

Alternatives Considered

N/A

Additional Context

No response

Would you like to submit a PR for this Issue?

No

Support Needs

No response

[jfecher]

We should also remove the array of structs -> structs of arrays conversion that happens in monomorphisation. Once the refactor is the default, it is unneeded complexity and it makes serialization order of arguments between Noir and TypeScript less intuitive. From user feedback:

Noir serializes arrays of structs packing the members together instead of the struct together, so if calling another function, for example with an array of points, instead of
PrivateCallStackItem::call(... [points[0].x, points[0].y, points[1].x, points[1].y...)
You have to do
PrivateCallStackItem::call(... [points[0].x, points[1].x, points[2].x, points[0].y...)
feels unnatural

[kevaundray]

We can also enable 2d arrays once the old SSA is removed

[kevaundray]

We should re-evaluate #562 when closing this too

[jfecher]

@kevaundray, #562 should be unrelated since that is a type system bug

[kevaundray]

@kevaundray, #562 should be unrelated since that is a type system bug

When we make this the default, we would also be merging in the breaking change which would force users to use [T;N] instead of [T], so I think that would issue may be fixed

[kevaundray]

This issue should also be reviewed once this is closed, since this can only be closed after dynamic indices: #793

[jfecher]

Issues that can be fixed after the old ssa is removed:

No extra work required:

• App crashes on nargo test --show-output #1982
• nargo test passes or fails arbitrarily with no changes to the code #1986

Some extra work:

• Move evaluation of the array_len builtin from monomorphization to SSA/acir-gen #1736

[kevaundray]

Also adding: #1830

[kevaundray]

Adding #1925

[kevaundray]

This was mentioned here: #1842

[TomAFrench]

I've closed the linked issues which are fixed by just using the new SSA.

[jfecher]

All bugs and changes listed here have been completed.