Make Ultra AllEntities use composition

[ludamad]

See AztecProtocol/aztec-packages#3560 (comment) for pain this incurred due to not matching order in precomputed and all entities

[ledwards2225]

@fcarreiro, sounds like @ludamad believes this could be significant in terms of compilation times for AVM

[ludamad]

I think the flavors are in an okay spot right now for us. I guess @fcarreiro composition could help by representing substructures instead of just a huge listing of member types. This is adding a fair chunk for anything that includes avm flavor.

Other things like the sumcheck tuple of tuples derivation likely need to be improved across the board

[fcarreiro]

Does this issue mean using composition vs inheritance? I think we use inheritance (see https://github.com/AztecProtocol/aztec-packages/blob/next/barretenberg/cpp/src/barretenberg/vm2/constraining/flavor.hpp#L169) and this was copied from some other bb stuff. We can of course diverge, and I might do it at some point. Probably I'll just have one big array of fields, we don't care as much about the names because we access them differently now.

@ludamad however I wonder how you are measuring that the AllEntities is taking time?

Other things like the sumcheck tuple of tuples derivation likely need to be improved across the board

Yep this one I mentioned a few times. Some things I have in mind are:

• Moving tupleoftuples (and the computation) to sumcheck (and not the flavor). It's only used there and otherwise it's instantiated for every flavor variant (field, etc) even if not used. For us it will make a diff
• Using tuplet for tuples. At least for us, made a big diff.

(but I don't think any of these will happen before we are feature complete)

[ludamad]

Closed as not planned any more. for AVM, the big array of fields approach sounds best