Weight V2 (Chromatic Weight)

[shawntabrizi]

This is an initial PR where we change the Weight type from an primitive u64 to:

pub struct Weight {
	/// The weight of computational time used.
	ref_time: ComputationalWeight,
	/// The weight of the storage bandwidth used.
	proof_size: StorageWeight,
}

Fields are private to allow adding of new fields without needing to touch everything, again.

The strategy here is to migrate the core weight checking logic within frame_system and frame_support. Majority of pallets will still use type ComputationWeight = u64. We will convert all the places that still use the primitive types in a follow up PR.

Right now the bandwidth parameter correctly handled in frame_executive and frame_system, and we will make sure that new extrinsics do not bypass that limit, but basically all Pallets simply return 0 bandwidth for now, so this logic is not really used.

The check_weight extension takes the bandwith into account. The transaction-payment extension only looks at the computation. We will need to come up with a way to fee PoV usage in a follow up. So right now we only make sure that not too much bandwith is used but don't incur a fee for it.

For migrating existing pallets to the new Weight definition we will simply use a Weight instance with computation: weight_v1, bandwidth: 0. The computation part can be accessed with a simple getter weight.computation().

After this PR is merged, each pallet can be migrated one by one to correctly handle the bandwidth parameter.

• Remove all notions of WeightV1 and replace it by a normal Weight with computation = 0.
• Make fields of Weight private so we can add more fields later without breaking all the users

fixes #9584

Polkadot companion: paritytech/polkadot#5435

[ggwpez]

The StorageWeight is mainly POV size I assume?

Not sure what is nicer from the naming, but when we do not want to litter the weights mod we could add a v2 mod:

use frame_support::weights::v2::Weight;
// vs
use frame_support::weights::Weight2 as Weight;

[athei]

The StorageWeight is mainly POV size I assume?

FRAME is oblivious to this concept. PoV is what a collator sends to a validator for validation. In the cumulus case this is roughly a witness + extrinsics. For the extrinsics we already account in FRAME (this is the block lenght). The witness is what is missing from the equation. This PoV metering just assumes that accessing storage will grow the witness.

[ggwpez]

Thanks for the explanation @athei !

[ggwpez]

I think we still need to run the bench bot, to see if the template still works.

[gavofyork]

does the WeightInfo not now give the new Weight back from its functions?

[gavofyork]

Shouldn't the weights.rs files be returning the full weight, not just computation weight? Where would we expect the PoV bandwidth weight dimension to come from?

[shawntabrizi]

Yes, this is a needed follow up to regenerate weights with the proper template and values. This is just a backwards compat hack for now.

[pepyakin]

Suggested change

	/// Construct with computation weight only (zero `bandwith`).
	/// Construct with computation weight only (zero `bandwidth`).

[pepyakin]

Needs explicitly mentioning what happens in case of overflow or exceeding any of the limits

[pepyakin]

Unclear what does aggressive mean here

[pepyakin]

same as aggressive

[kianenigma]

Canonical names:

• ref_time
• proof_size

Would be good to make a function explicitly named for backwards compatibility, like from_legacy() or from_v1() that is equivalent to from_ref_time, so that it can be easily identified and used as TODO.

[louismerlin]

Suggested change

	/// Checks if any param of `self` is less than `other`.
	/// Checks if any param of `self` is greater than `other`.