Skip to main content

Contract classes

A contract class is a collection of state variable declarations, and related unconstrained, private, and public functions. Contract classes don't have any initialized state, they just define code. A contract class cannot be called; only a contract instance can be called.

Rationale​

Contract classes simplify the process of reusing code by enshrining implementations as a first-class citizen at the protocol. Given multiple contract instances that rely on the same class, the class needs to be declared only once, reducing the deployment cost for all contract instances. Classes also simplify the process of upgradeability; classes decouple state from code, making it easier for an instance to switch to different code while retaining its state.

info

Read the following discussions for additional context:

ContractClass​

The structure of a contract class is defined as:

FieldTypeDescription
versionu8Version identifier. Initially one, bumped for any changes to the contract class struct.
artifact_hashFieldHash of the contract artifact. The specification of this hash is not enforced by the protocol. Should include commitments to unconstrained code and compilation metadata. Intended to be used by clients to verify that an off-chain fetched artifact matches a registered class.
private_functionsPrivateFunction[]List of individual private functions, constructors included.
packed_public_bytecodeField[]Packed bytecode representation of the AVM bytecode for all public functions in this contract.

The public function are sorted in ascending order by their function selector before being packed. This is to ensure consistent hashing later.

Note that individual public functions are not first-class citizens in the protocol, so the contract entire public function bytecode is stored in the class, unlike private or unconstrained functions which are differentiated individual circuits recognized by the protocol.

As for unconstrained functions, these are not used standalone within the protocol. They are either inlined within private functions, or called from a PXE as getters for a contract. Calling from a private function to an unconstrained one in a different contract is forbidden, since the caller would have no guarantee of the code run by the callee. Considering this, unconstrained functions are not part of a contract class at the protocol level.

contract_class_id​

Also known as contract_class_id, the Class Identifier is both a unique identifier and a commitment to the struct contents. It is computed as:

contract_class_id_crh(
    artifact_hash: Field
    private_functions: PrivateFunction[],
    packed_public_bytecode: bytes[],
) -> Field {
    let private_function_leaves: Field[] = private_functions.map(|f| private_function_leaf_crh(f));

    // Illustrative function, not defined. TODO.
    let private_function_tree_root: Field = merkleize(private_function_leaves);

    // Illustrative function, not defined. TODO.
    let public_bytecode_commitment: Point = calculate_commitment(packed_public_bytecode);

    let contract_class_id = poseidon2(
        be_string_to_field("az_contract_class_id"),

        artifact_hash,
        private_function_tree_root,
        public_bytecode_commitment.x,
        public_bytecode_commitment.y,
    );

    contract_class_id
}

See below for private_function_leaf_crh. Private Functions are sorted in ascending order by their selector, and then hashed into Function Leaves, before being merkleized into a tree of height PRIVATE_FUNCTION_TREE_HEIGHT. Empty leaves have value 0. The AVM public bytecode commitment is calculated as defined in the Public VM section.

PrivateFunction​

The structure of each private function within the protocol is the following:

FieldTypeDescription
function_selectoru32Selector of the function. Calculated as the hash of the method name and parameters. The specification of this is not enforced by the protocol.
vk_hashFieldHash of the verification key associated to this private function.

Note the lack of visibility modifiers. Internal functions are specified as a macro, and the check is handled at the application circuit level by verifying that the context.msg_sender equals the contract current address.

Also note the lack of commitment to the function compilation artifact. Even though a commitment to a function is required so that the PXE can verify the execution of correct unconstrained Brillig code embedded within private functions, this is handled entirely out of protocol. As such, PXEs are expected to verify it against the artifact_hash in the containing contract class.

Private Function Leaf Hash​

private_function_leaf_crh(
    f: PrivateFunction
) -> Field {
    let private_function_leaf = poseidon2(
        be_string_to_field("az_private_function_leaf"),

        be_bits_to_field(f.function_selector),
        f.vk_hash
    );

    private_function_leaf
}

Artifact Hash​

Even though not enforced by the protocol, it is suggested for the artifact_hash to follow this general structure, in order to be compatible with the definition of the broadcast function below.

Note: below, sha256_modulo(x) = sha256(x) % FIELD_MODULUS. This approach must not be used if seeking pseudo-randomness, but can be used for collision resistance.

artifact_crh(
  artifact // This type is out of protocol, e.g. the format output by Nargo
) -> Field {

  let private_functions_artifact_leaves: Field[] = artifact.private_functions.map(|f|
    sha256_modulo(
      be_string_to_bits("az_artifact_private_function_leaf"),

      f.selector, // 32-bits
      f.metadata_hash, // 256-bits
      sha256(f.private_bytecode)
    )
  );
  let private_functions_artifact_tree_root: Field = merkleize(private_functions_artifact_leaves);

  let unconstrained_functions_artifact_leaves: Field[] = artifact.unconstrained_functions.map(|f|
    sha256_modulo(
      be_string_to_bits("az_artifact_unconstrained_function_leaf"),

      f.selector, // 32-bits
      f.metadata_hash, // 256-bits
      sha256(f.unconstrained_bytecode)
    )
  );
  let unconstrained_functions_artifact_tree_root: Field = merkleize(unconstrained_functions_artifact_leaves);

  let artifact_hash: Field = sha256_modulo(
    be_string_to_field("az_artifact"),

    private_functions_artifact_tree_root, // 256-bits
    unconstrained_functions_artifact_tree_root, // 256-bits
    artifact_metadata
  );

  let artifact_hash: Field = artifact_hash_256_bit % FIELD_MODULUS;

  artifact_hash
}

For the artifact hash merkleization and hashing is done using sha256, since it is computed and verified outside of circuits and does not need to be SNARK friendly, and then wrapped around the field's maximum value. Fields are left-padded with zeros to 256 bits before being hashed. Function leaves are sorted in ascending order before being merkleized, according to their function selectors. Note that a tree with dynamic height is built instead of having a tree with a fixed height, since the merkleization is done out of a circuit.

Bytecode for private functions is a mix of ACIR and Brillig, whereas unconstrained function bytecode is Brillig exclusively, as described on the bytecode section.

The metadata hash for each function is suggested to be computed as the sha256 of all JSON-serialized fields in the function struct of the compilation artifact, except for bytecode and debug symbols. The metadata is JSON-serialized using no spaces, and sorting ascending all keys in objects before serializing them.

function_metadata_crh(
  function // This type is out of protocol, e.g. the format output by Nargo
) -> Field {
  let function_metadata = omit(function, "bytecode", "debug_symbols");

  let function_metadata_hash: Field = sha256_modulo(
    be_string_to_bits("az_function_metadata"),

    json_serialize(function_metadata)
  );

  function_metadata_hash
}

The artifact metadata stores all data that is not contained within the contract functions and is not debug specific. This includes the compiler version identifier, events interface, and name. Metadata is JSON-serialized in the same fashion as the function metadata.

artifact_metadata_crh(
  artifact // This type is out of protocol, e.g. the format output by Nargo
) -> Field {
  let artifact_metadata = omit(artifact, "functions", "file_map");

  let artifact_metadata_hash: Field = sha256_modulo(
    be_string_to_bits("az_artifact_metadata"),

    json_serialize(artifact_metadata)
  );

  artifact_metadata_hash
}

Versioning​

A contract class has an implicit version field that identifies the schema of the struct. This allows to change the shape of a contract class in future upgrades to the protocol to include new fields or change existing ones, while preserving the structure for existing classes. Supporting new types of contract classes would require introducing new kernel circuits, and a transaction proof may require switching between different kernel circuits depending on the version of the contract class used for each function call.

Note that the version field is not directly used when computing the contract class id, but is implicit in the generator index. Bumping the version of a contract class struct would involve using a different generator index for computing its id.

Canonical Contract Class Registerer​

A contract class is registered by calling a private register function in a canonical ContractClassRegisterer contract, which will emit a Registration Nullifier. The Registration Nullifier is defined as the contract_class_id itself of the class being registered. Note that the Private Kernel circuit will silo this value with the contract address of the ContractClassRegisterer, effectively storing the hash of the contract_class_id and ContractClassRegisterer address in the nullifier tree. As such, proving that a given contract class has been registered requires checking existence of this siloed nullifier.

The rationale for the Registerer contract is to guarantee that the public bytecode for a contract class is publicly available. This is a requirement for publicly deploying a contract instance, which ultimately prevents a sequencer from executing a public function for which other nodes in the network may not have the code.

Register Function​

The register function receives the artifact hash, private functions tree root, and packed public bytecode of a ContractClass struct as defined above, and performs the following steps:

  • Assert that packed_public_bytecode is valid according to the definition in the Public VM section.
  • Computes the contract_class_id as defined above.
  • Emits the resulting contract_class_id as a nullifier to prevent the same class from being registered again.
  • Emits an unencrypted event ContractClassRegistered with the contents of the contract class.

In pseudocode:

fn register(
  artifact_hash: Field,
  private_functions_root: Field,
  public_bytecode_commitment: Point,
  packed_public_bytecode: Field[],
) {
  assert(is_valid_packed_public_bytecode(packed_public_bytecode));

  let computed_bytecode_commitment: Point = calculate_commitment(packed_public_bytecode);

  assert(public_bytecode_commitment == computed_bytecode_commitment);

  let version: Field = 1;
  let contract_class_id = contract_class_id_crh(version, artifact_hash, private_functions_root, bytecode_commitment);

  emit_nullifier(contract_class_id);

  emit_unencrypted_event(ContractClassRegistered::new(
    contract_class_id,
    version,
    artifact_hash,
    private_functions_root,
    packed_public_bytecode
  ));
}

Upon seeing a ContractClassRegistered event in a mined transaction, nodes are expected to store the contract class, so they can retrieve it when executing a public function for that class. Note that a class may be used for deploying a contract within the same transaction in which it is registered.

Note that emitting the contract_class_id as a nullifier (the contract_class_id_nullifier), instead of as an entry in the note hashes tree, allows nodes to prove non-existence of a class. This is needed so a sequencer can provably revert a transaction that includes a call to an unregistered class.

Genesis​

The ContractClassRegisterer will need to exist from the genesis of the Aztec Network, otherwise nothing will ever be publicly deployable to the network. The Class Nullifier for the ContractClassRegisterer contract will be pre-inserted into the genesis nullifier tree at leaf index GENESIS_NULLIFIER_LEAF_INDEX_OF_CONTRACT_CLASS_REGISTERER_CLASS_ID_NULLIFIER. The canonical instance will be deployed at CONTRACT_CLASS_REGISTERER_ADDRESS, and its Deployment Nullifier will be inserted at GENESIS_NULLIFIER_LEAF_INDEX_OF_CONTRACT_CLASS_REGISTERER_DEPLOYMENT_NULLIFIER.

Broadcast​

The ContractClassRegisterer has an additional private broadcast functions that can be used for broadcasting on-chain the bytecode, both ACIR and Brillig, for private functions and unconstrained in the contract. Any user can freely call this function. Given that ACIR and Brillig do not have a circuit-friendly commitment, it is left up to nodes to perform this check.

Broadcasted function artifacts that do not match with their corresponding artifact_hash, or that reference a contract_class_id that has not been broadcasted, can be safely discarded.

fn broadcast_private_function(
  contract_class_id: Field,
  artifact_metadata_hash: Field,
  unconstrained_functions_artifact_tree_root: Field,
  private_function_tree_sibling_path: Field[],
  private_function_tree_leaf_index: Field,
  artifact_function_tree_sibling_path: Field[],
  artifact_function_tree_leaf_index: Field,
  function: { selector: Field, metadata_hash: Field, vk_hash: Field, bytecode: Field[] },
)
  emit_unencrypted_event ClassPrivateFunctionBroadcasted(
    contract_class_id,
    artifact_metadata_hash,
    unconstrained_functions_artifact_tree_root,
    private_function_tree_sibling_path,
    private_function_tree_leaf_index,
    artifact_function_tree_sibling_path,
    artifact_function_tree_leaf_index,
    function,
  )
fn broadcast_unconstrained_function(
  contract_class_id: Field,
  artifact_metadata_hash: Field,
  private_functions_artifact_tree_root: Field,
  artifact_function_tree_sibling_path: Field[],
  artifact_function_tree_leaf_index: Field
  function: { selector: Field, metadata_hash: Field, bytecode: Field[] }[],
)
  emit_unencrypted_event ClassUnconstrainedFunctionBroadcasted(
    contract_class_id,
    artifact_metadata_hash,
    private_functions_artifact_tree_root,
    artifact_function_tree_sibling_path,
    artifact_function_tree_leaf_index,
    function,
  )

The broadcast functions are split between private and unconstrained to allow for private bytecode to be broadcasted, which is valuable for composability purposes, without having to also include unconstrained functions, which could be costly to do due to data broadcasting costs. Additionally, note that each broadcast function must include enough information to reconstruct the artifact_hash from the Contract Class, so nodes can verify it against the one previously registered.

A node that captures a ClassPrivateFunctionBroadcasted should perform the following validation steps before storing the private function information in its database:

// Load contract class from local db
contract_class = db.get_contract_class(contract_class_id)

// Compute function leaf and assert it belongs to the private functions tree
function_leaf = pedersen([selector as Field, vk_hash], GENERATOR__FUNCTION_LEAF)
computed_private_function_tree_root = compute_root(function_leaf, private_function_tree_sibling_path, private_function_tree_leaf_index)
assert computed_private_function_tree_root == contract_class.private_function_root

// Compute artifact leaf and assert it belongs to the artifact
artifact_function_leaf = sha256(selector, metadata_hash, sha256(bytecode))
computed_artifact_private_function_tree_root = compute_root(artifact_function_leaf, artifact_function_tree_sibling_path, artifact_function_tree_leaf_index)
computed_artifact_hash = sha256(computed_artifact_private_function_tree_root, unconstrained_functions_artifact_tree_root, artifact_metadata_hash)
assert computed_artifact_hash == contract_class.artifact_hash

The check for an unconstrained function is similar:

// Load contract class from local db
contract_class = db.get_contract_class(contract_class_id)

// Compute artifact leaf and assert it belongs to the artifact
artifact_function_leaf = sha256(selector, metadata_hash, sha256(bytecode))
computed_artifact_unconstrained_function_tree_root = compute_root(artifact_function_leaf, artifact_function_tree_sibling_path, artifact_function_tree_leaf_index)
computed_artifact_hash = sha256(private_functions_artifact_tree_root, computed_artifact_unconstrained_function_tree_root, artifact_metadata_hash)
assert computed_artifact_hash == contract_class.artifact_hash

It is strongly recommended for developers registering new classes to broadcast the code for compute_hash_and_nullifier, so any private message recipients have the code available to process their incoming notes. However, the ContractClassRegisterer contract does not enforce this during registration, since it is difficult to check the multiple signatures for compute_hash_and_nullifier as they may evolve over time to account for new note sizes.

Encoding Bytecode​

The register, broadcast_unconstrained_function, and broadcast_private_function functions all receive and emit variable-length bytecode in unencrypted events. In every function, bytecode is encoded in a fixed-length array of field elements, which sets a maximum length for each:

  • MAX_PACKED_PUBLIC_BYTECODE_SIZE_IN_FIELDS: 15000 field elements, used for a contract's public bytecode in the register function.
  • MAX_PACKED_BYTECODE_SIZE_PER_PRIVATE_FUNCTION_IN_FIELDS: 3000 field elements, used for the ACIR and Brillig bytecode of a broadcasted private function in broadcast_private_function.
  • MAX_PACKED_BYTECODE_SIZE_PER_UNCONSTRAINED_FUNCTION_IN_FIELDS: 3000 field elements, used for the Brillig bytecode of a broadcasted unconstrained function in broadcast_unconstrained_function.

To encode the bytecode into a fixed-length array of Fields, the bytecode is first split into 31-byte chunks, and each chunk interpreted big-endian as a field element. The total length in bytes is then prepended as an initial element, and then right-padded with zeroes.

chunks = chunk bytecode into 31 bytes elements, last element right-padded with zeroes
fields = right-align each chunk into 32 bytes and cast to a field element
padding = repeat a zero-value field MAX_SIZE - fields.count - 1 times
encoded = [bytecode.length as field, ...fields, ...padding]

Discarded Approaches​

Bundling private function information into a single tree​

Data about private functions is split across two trees: one for the protocol, that deals only with selectors and verification keys, and one for the artifact, which deals with bytecode and metadata. While bundling together both trees would simplify the representation, it would also pollute the protocol circuits and require more hashing there. In order to minimize in-circuit hashing, we opted for keeping non-protocol info completely out of circuits.