chore: post blob batching cleanup#14637
Merged
Merged
Conversation
This reverts commit 68be71e.
…atching-bls-utils
… mw/blob-batching-bls-utils-ts
…atching-bls-utils
… mw/blob-batching-bls-utils-ts
… mw/blob-batching-bls-utils-ts
…atching-bls-utils
… mw/blob-batching-bls-utils-ts
…nto mw/blob-batching-integration
…atching-final-cleanup
…atching-final-cleanup
github-merge-queue Bot
pushed a commit
that referenced
this pull request
Jun 9, 2025
## The blobs are back in town. This PR reworks blobs so that instead of calling the point evaluation precompile for each blob (currently up to 3 per block => up to 96 (?) calls per epoch), we call it once per epoch by batching blobs to a single kzg commitment, opening, challenge, and proof. How we can be sure that this one pairing check is equivalent to a check per blob is covered in the maths by @iAmMichaelConnor [here](https://hackmd.io/WUtNusQxS5KAw-af3gxycA?view) 🎉 ## Overview Instead of pushing to a long array of `BlobPublicInputs`, which are then individually checked on L1, we batch each blob together to a single set of `BlobAccumulatorPublicInputs`. The `start` accumulator state is fed into each block root circuit, where the block's blobs are accumulated and the `end` state is set. Each block merge circuit checks that the state follows on correctly and, finally, the root circuit checks that the very `start` state was empty and finalises the last `end` state. This last `end` state makes up the set of inputs for the point evaluation precompile. If the pairing check in that precompile passes, we know that all blobs for all blocks in the epoch are valid and contain only the tx effects validated by the rollup. ### Circuits Key changes: - Integrate BLS12-381 curve operations with `bignum` and `bigcurve` libraries, plus tests. - Rework the `blob` package to batch blobs and store in reworked structs, plus tests. - Rework the rollup circuits from `block_root` above to handle blob accumulation state rather than a list of individual blob inputs, plus (you guessed it) tests. ### Contracts The contracts: - No longer call the point evaluation precompile on `propose`, instead inject the blob commitments, check they correspond to the broadcast blobs, and stores them in the `blobCommitmentsHash`. - Do not store any blob public inputs apart from the `blobCommitmentsHash`. - Call the point evaluation precompile once on `submitEpochRootProof` for ALL blobs in the epoch. - Use the same precompile inputs as pubic inputs to the root proof verification along with the `blobCommitmentsHash` to link the circuit batched blob, real L1 blobs, and the batched blob verified on L1. ### TypeScript Key changes: - Edit all the structs and methods reliant on the circuits/contracts to match the above changes. - Inject the final blob challenges used on each blob into all block building methods in `orchestrator`. - Accumulate blobs in ts when building blocks and use as inputs to each rollup circuit, plus tests. - Return the blob inputs required for `submitEpochRootProof` on `finaliseEpoch()`. ### TODOs/Related Issues - Choose field for hashing challenge: #13608 - Instead of exponentiating `gamma` (expensive!), hash it for each iteration: #13740 - Number of public inputs: BLS points in noir take up 9 fields (4 for each coordinate as a limbed bignum, 1 for the is_inf flag) but can be compressed to only 2. For recursive verification in block root and above, would it be worth the gates to compress these? It depends whether the gate cost of compression is more/less than gate cost of recursively verifying 7 more public inputs. - Remove the large trusted setup file from `yarn-project/blob-lib/src/trusted_setup_bit_reversed.json`? Used in testing, but may not be worth keeping (see code comments). - Cleanup old, unused blob stuff in #14637. ## PR Stack - [x] `mw/blob-batching` <- main feature - [x] ^ `mw/blob-batching-bls-utils` <- BLS12-381 bigcurve and bignum utils (noir) (#13583) - [x] ^ `mw/blob-batching-bls-utils-ts` <- BLS12-381 bigcurve and bignum utils (ts) (#13606) - [x] ^ `mw/blob-batching-integration` <- Integrate batching into noir protocol circuits (#13817) - [x] ^ `mw/blob-batching-integration-ts-sol` <- Integrate batching into ts and solidity (#14329) - [ ] ^ `mw/blob-batching-cleanup` <- Remove old blob code --------- Co-authored-by: Tom French <15848336+TomAFrench@users.noreply.github.com>
LeilaWang
approved these changes
Jun 10, 2025
LeilaWang
left a comment
LeilaWang
left a comment
Contributor
There was a problem hiding this comment.
Nice cleanup! Just a few nits 😛
| // eslint-disable-next-line @typescript-eslint/no-this-alias | ||
| let acc: BatchedBlobAccumulator = this; // TODO(MW): this.clone() | ||
| // Initialise the acc to iterate over: | ||
| let acc: BatchedBlobAccumulator = Object.create(this); |
Contributor
There was a problem hiding this comment.
Probably better to define a clone method and avoid footgun of using Object.create. The data is not too big to be cloned and will be easier to understand.
| * @returns The point fields. | ||
| */ | ||
| toBN254Fields() { | ||
| return [new Fr(this.compress().subarray(0, 31)), new Fr(this.compress().subarray(31, 48))]; |
Contributor
There was a problem hiding this comment.
Can just call this.compress() once. It's slightly more efficient and easier to see that the fields are taken from the same buffer.
|
|
||
| // Note: we may be able to use the constant values in block_root/empty_block_root_rollup_inputs.nr, set by block_building_helpers.test.ts, but | ||
| // having this separate fn hopefully makes it clear how we treat empty blocks and their blobs, and won't break if we decide to change how | ||
| // getBlobsPerBlock() works on empty input. |
Contributor
There was a problem hiding this comment.
Let's include this in the test so we know the values match!
github-merge-queue
Bot
removed this pull request from the merge queue due to failed status checks
danielntmd
pushed a commit
to danielntmd/aztec-packages
that referenced
this pull request
Jul 16, 2025
## The blobs are back in town. This PR reworks blobs so that instead of calling the point evaluation precompile for each blob (currently up to 3 per block => up to 96 (?) calls per epoch), we call it once per epoch by batching blobs to a single kzg commitment, opening, challenge, and proof. How we can be sure that this one pairing check is equivalent to a check per blob is covered in the maths by @iAmMichaelConnor [here](https://hackmd.io/WUtNusQxS5KAw-af3gxycA?view) 🎉 ## Overview Instead of pushing to a long array of `BlobPublicInputs`, which are then individually checked on L1, we batch each blob together to a single set of `BlobAccumulatorPublicInputs`. The `start` accumulator state is fed into each block root circuit, where the block's blobs are accumulated and the `end` state is set. Each block merge circuit checks that the state follows on correctly and, finally, the root circuit checks that the very `start` state was empty and finalises the last `end` state. This last `end` state makes up the set of inputs for the point evaluation precompile. If the pairing check in that precompile passes, we know that all blobs for all blocks in the epoch are valid and contain only the tx effects validated by the rollup. ### Circuits Key changes: - Integrate BLS12-381 curve operations with `bignum` and `bigcurve` libraries, plus tests. - Rework the `blob` package to batch blobs and store in reworked structs, plus tests. - Rework the rollup circuits from `block_root` above to handle blob accumulation state rather than a list of individual blob inputs, plus (you guessed it) tests. ### Contracts The contracts: - No longer call the point evaluation precompile on `propose`, instead inject the blob commitments, check they correspond to the broadcast blobs, and stores them in the `blobCommitmentsHash`. - Do not store any blob public inputs apart from the `blobCommitmentsHash`. - Call the point evaluation precompile once on `submitEpochRootProof` for ALL blobs in the epoch. - Use the same precompile inputs as pubic inputs to the root proof verification along with the `blobCommitmentsHash` to link the circuit batched blob, real L1 blobs, and the batched blob verified on L1. ### TypeScript Key changes: - Edit all the structs and methods reliant on the circuits/contracts to match the above changes. - Inject the final blob challenges used on each blob into all block building methods in `orchestrator`. - Accumulate blobs in ts when building blocks and use as inputs to each rollup circuit, plus tests. - Return the blob inputs required for `submitEpochRootProof` on `finaliseEpoch()`. ### TODOs/Related Issues - Choose field for hashing challenge: AztecProtocol#13608 - Instead of exponentiating `gamma` (expensive!), hash it for each iteration: AztecProtocol#13740 - Number of public inputs: BLS points in noir take up 9 fields (4 for each coordinate as a limbed bignum, 1 for the is_inf flag) but can be compressed to only 2. For recursive verification in block root and above, would it be worth the gates to compress these? It depends whether the gate cost of compression is more/less than gate cost of recursively verifying 7 more public inputs. - Remove the large trusted setup file from `yarn-project/blob-lib/src/trusted_setup_bit_reversed.json`? Used in testing, but may not be worth keeping (see code comments). - Cleanup old, unused blob stuff in AztecProtocol#14637. ## PR Stack - [x] `mw/blob-batching` <- main feature - [x] ^ `mw/blob-batching-bls-utils` <- BLS12-381 bigcurve and bignum utils (noir) (AztecProtocol#13583) - [x] ^ `mw/blob-batching-bls-utils-ts` <- BLS12-381 bigcurve and bignum utils (ts) (AztecProtocol#13606) - [x] ^ `mw/blob-batching-integration` <- Integrate batching into noir protocol circuits (AztecProtocol#13817) - [x] ^ `mw/blob-batching-integration-ts-sol` <- Integrate batching into ts and solidity (AztecProtocol#14329) - [ ] ^ `mw/blob-batching-cleanup` <- Remove old blob code --------- Co-authored-by: Tom French <15848336+TomAFrench@users.noreply.github.com>
danielntmd
pushed a commit
to danielntmd/aztec-packages
that referenced
this pull request
Jul 16, 2025
## Cleanup
Added as a new branch as to not bloat the number of files\lines changed
in the core branches. Also can be merged later as (should) touch no
logic.
This PR:
- Rearranges nr methods into `blob_batching.nr`, keep `blob.nr` for
evaluation only
- Delete now unused `blob_public_inputs.nr`
- Use bigcurve 0.9.0
- Rename `Blob.getBlobs()` -> `Blob.getBlobsPerBlock()` (previously we
only dealt with blobs per block, but now we deal with them across
epochs, so `getBlobs` may be confusing)
- Remove `y` and `proof` from the `Blob` class (only used in
`BatchedBlob`, and not required to be carried around the repo)
- Remove a few now unused methods
- Remove circular dependencies and linter warnings
TODO on this branch:
- Rename `blob.nr` to `barycentric.nr` or similar?
- Rename BatchingBlobCommitment ('BatchedBlobCommitment'?)?
---------
Co-authored-by: Tom French <15848336+TomAFrench@users.noreply.github.com>
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Cleanup
Added as a new branch as to not bloat the number of files\lines changed in the core branches. Also can be merged later as (should) touch no logic.
This PR:
blob_batching.nr, keepblob.nrfor evaluation onlyblob_public_inputs.nrBlob.getBlobs()->Blob.getBlobsPerBlock()(previously we only dealt with blobs per block, but now we deal with them across epochs, sogetBlobsmay be confusing)yandprooffrom theBlobclass (only used inBatchedBlob, and not required to be carried around the repo)TODO on this branch:
blob.nrtobarycentric.nror similar?