miden-crypto/src/merkle/smt/mod.rs
Qyriad 1b1c4ccedb smt: implement single subtree-8 hashing, w/ benchmarks & tests
This will be composed into depth-8-subtree-based computation of entire
sparse Merkle trees.
2024-11-14 15:34:16 -07:00

875 lines
35 KiB
Rust

use alloc::{collections::BTreeMap, vec::Vec};
use core::mem;
use num::Integer;
use super::{EmptySubtreeRoots, InnerNodeInfo, MerkleError, MerklePath, NodeIndex};
use crate::{
hash::rpo::{Rpo256, RpoDigest},
Felt, Word, EMPTY_WORD,
};
mod full;
pub use full::{Smt, SmtLeaf, SmtLeafError, SmtProof, SmtProofError, SMT_DEPTH};
mod simple;
pub use simple::SimpleSmt;
// CONSTANTS
// ================================================================================================
/// Minimum supported depth.
pub const SMT_MIN_DEPTH: u8 = 1;
/// Maximum supported depth.
pub const SMT_MAX_DEPTH: u8 = 64;
// SPARSE MERKLE TREE
// ================================================================================================
/// An abstract description of a sparse Merkle tree.
///
/// A sparse Merkle tree is a key-value map which also supports proving that a given value is indeed
/// stored at a given key in the tree. It is viewed as always being fully populated. If a leaf's
/// value was not explicitly set, then its value is the default value. Typically, the vast majority
/// of leaves will store the default value (hence it is "sparse"), and therefore the internal
/// representation of the tree will only keep track of the leaves that have a different value from
/// the default.
///
/// All leaves sit at the same depth. The deeper the tree, the more leaves it has; but also the
/// longer its proofs are - of exactly `log(depth)` size. A tree cannot have depth 0, since such a
/// tree is just a single value, and is probably a programming mistake.
///
/// Every key maps to one leaf. If there are as many keys as there are leaves, then
/// [Self::Leaf] should be the same type as [Self::Value], as is the case with
/// [crate::merkle::SimpleSmt]. However, if there are more keys than leaves, then [`Self::Leaf`]
/// must accomodate all keys that map to the same leaf.
///
/// [SparseMerkleTree] currently doesn't support optimizations that compress Merkle proofs.
pub(crate) trait SparseMerkleTree<const DEPTH: u8> {
/// The type for a key
type Key: Clone + Ord;
/// The type for a value
type Value: Clone + PartialEq;
/// The type for a leaf
type Leaf: Clone;
/// The type for an opening (i.e. a "proof") of a leaf
type Opening;
/// The default value used to compute the hash of empty leaves
const EMPTY_VALUE: Self::Value;
/// The root of the empty tree with provided DEPTH
const EMPTY_ROOT: RpoDigest;
// PROVIDED METHODS
// ---------------------------------------------------------------------------------------------
/// Returns an opening of the leaf associated with `key`. Conceptually, an opening is a Merkle
/// path to the leaf, as well as the leaf itself.
fn open(&self, key: &Self::Key) -> Self::Opening {
let leaf = self.get_leaf(key);
let mut index: NodeIndex = {
let leaf_index: LeafIndex<DEPTH> = Self::key_to_leaf_index(key);
leaf_index.into()
};
let merkle_path = {
let mut path = Vec::with_capacity(index.depth() as usize);
for _ in 0..index.depth() {
let is_right = index.is_value_odd();
index.move_up();
let InnerNode { left, right } = self.get_inner_node(index);
let value = if is_right { left } else { right };
path.push(value);
}
MerklePath::new(path)
};
Self::path_and_leaf_to_opening(merkle_path, leaf)
}
/// Inserts a value at the specified key, returning the previous value associated with that key.
/// Recall that by definition, any key that hasn't been updated is associated with
/// [`Self::EMPTY_VALUE`].
///
/// This also recomputes all hashes between the leaf (associated with the key) and the root,
/// updating the root itself.
fn insert(&mut self, key: Self::Key, value: Self::Value) -> Self::Value {
let old_value = self.insert_value(key.clone(), value.clone()).unwrap_or(Self::EMPTY_VALUE);
// if the old value and new value are the same, there is nothing to update
if value == old_value {
return value;
}
let leaf = self.get_leaf(&key);
let node_index = {
let leaf_index: LeafIndex<DEPTH> = Self::key_to_leaf_index(&key);
leaf_index.into()
};
self.recompute_nodes_from_index_to_root(node_index, Self::hash_leaf(&leaf));
old_value
}
/// Recomputes the branch nodes (including the root) from `index` all the way to the root.
/// `node_hash_at_index` is the hash of the node stored at index.
fn recompute_nodes_from_index_to_root(
&mut self,
mut index: NodeIndex,
node_hash_at_index: RpoDigest,
) {
let mut node_hash = node_hash_at_index;
for node_depth in (0..index.depth()).rev() {
let is_right = index.is_value_odd();
index.move_up();
let InnerNode { left, right } = self.get_inner_node(index);
let (left, right) = if is_right {
(left, node_hash)
} else {
(node_hash, right)
};
node_hash = Rpo256::merge(&[left, right]);
if node_hash == *EmptySubtreeRoots::entry(DEPTH, node_depth) {
// If a subtree is empty, when can remove the inner node, since it's equal to the
// default value
self.remove_inner_node(index)
} else {
self.insert_inner_node(index, InnerNode { left, right });
}
}
self.set_root(node_hash);
}
/// Computes what changes are necessary to insert the specified key-value pairs into this Merkle
/// tree, allowing for validation before applying those changes.
///
/// This method returns a [`MutationSet`], which contains all the information for inserting
/// `kv_pairs` into this Merkle tree already calculated, including the new root hash, which can
/// be queried with [`MutationSet::root()`]. Once a mutation set is returned,
/// [`SparseMerkleTree::apply_mutations()`] can be called in order to commit these changes to
/// the Merkle tree, or [`drop()`] to discard them.
fn compute_mutations(
&self,
kv_pairs: impl IntoIterator<Item = (Self::Key, Self::Value)>,
) -> MutationSet<DEPTH, Self::Key, Self::Value> {
use NodeMutation::*;
let mut new_root = self.root();
let mut new_pairs: BTreeMap<Self::Key, Self::Value> = Default::default();
let mut node_mutations: BTreeMap<NodeIndex, NodeMutation> = Default::default();
for (key, value) in kv_pairs {
// If the old value and the new value are the same, there is nothing to update.
// For the unusual case that kv_pairs has multiple values at the same key, we'll have
// to check the key-value pairs we've already seen to get the "effective" old value.
let old_value = new_pairs.get(&key).cloned().unwrap_or_else(|| self.get_value(&key));
if value == old_value {
continue;
}
let leaf_index = Self::key_to_leaf_index(&key);
let mut node_index = NodeIndex::from(leaf_index);
// We need the current leaf's hash to calculate the new leaf, but in the rare case that
// `kv_pairs` has multiple pairs that go into the same leaf, then those pairs are also
// part of the "current leaf".
let old_leaf = {
let pairs_at_index = new_pairs
.iter()
.filter(|&(new_key, _)| Self::key_to_leaf_index(new_key) == leaf_index);
pairs_at_index.fold(self.get_leaf(&key), |acc, (k, v)| {
// Most of the time `pairs_at_index` should only contain a single entry (or
// none at all), as multi-leaves should be really rare.
let existing_leaf = acc.clone();
self.construct_prospective_leaf(existing_leaf, k, v)
})
};
let new_leaf = self.construct_prospective_leaf(old_leaf, &key, &value);
let mut new_child_hash = Self::hash_leaf(&new_leaf);
for node_depth in (0..node_index.depth()).rev() {
// Whether the node we're replacing is the right child or the left child.
let is_right = node_index.is_value_odd();
node_index.move_up();
let old_node = node_mutations
.get(&node_index)
.map(|mutation| match mutation {
Addition(node) => node.clone(),
Removal => EmptySubtreeRoots::get_inner_node(DEPTH, node_depth),
})
.unwrap_or_else(|| self.get_inner_node(node_index));
let new_node = if is_right {
InnerNode {
left: old_node.left,
right: new_child_hash,
}
} else {
InnerNode {
left: new_child_hash,
right: old_node.right,
}
};
// The next iteration will operate on this new node's hash.
new_child_hash = new_node.hash();
let &equivalent_empty_hash = EmptySubtreeRoots::entry(DEPTH, node_depth);
let is_removal = new_child_hash == equivalent_empty_hash;
let new_entry = if is_removal { Removal } else { Addition(new_node) };
node_mutations.insert(node_index, new_entry);
}
// Once we're at depth 0, the last node we made is the new root.
new_root = new_child_hash;
// And then we're done with this pair; on to the next one.
new_pairs.insert(key, value);
}
MutationSet {
old_root: self.root(),
new_root,
node_mutations,
new_pairs,
}
}
/// Apply the prospective mutations computed with [`SparseMerkleTree::compute_mutations()`] to
/// this tree.
///
/// # Errors
/// If `mutations` was computed on a tree with a different root than this one, returns
/// [`MerkleError::ConflictingRoots`] with a two-item [`Vec`]. The first item is the root hash
/// the `mutations` were computed against, and the second item is the actual current root of
/// this tree.
fn apply_mutations(
&mut self,
mutations: MutationSet<DEPTH, Self::Key, Self::Value>,
) -> Result<(), MerkleError>
where
Self: Sized,
{
use NodeMutation::*;
let MutationSet {
old_root,
node_mutations,
new_pairs,
new_root,
} = mutations;
// Guard against accidentally trying to apply mutations that were computed against a
// different tree, including a stale version of this tree.
if old_root != self.root() {
return Err(MerkleError::ConflictingRoots(vec![old_root, self.root()]));
}
for (index, mutation) in node_mutations {
match mutation {
Removal => self.remove_inner_node(index),
Addition(node) => self.insert_inner_node(index, node),
}
}
for (key, value) in new_pairs {
self.insert_value(key, value);
}
self.set_root(new_root);
Ok(())
}
// REQUIRED METHODS
// ---------------------------------------------------------------------------------------------
/// The root of the tree
fn root(&self) -> RpoDigest;
/// Sets the root of the tree
fn set_root(&mut self, root: RpoDigest);
/// Retrieves an inner node at the given index
fn get_inner_node(&self, index: NodeIndex) -> InnerNode;
/// Inserts an inner node at the given index
fn insert_inner_node(&mut self, index: NodeIndex, inner_node: InnerNode);
/// Removes an inner node at the given index
fn remove_inner_node(&mut self, index: NodeIndex);
/// Inserts a leaf node, and returns the value at the key if already exists
fn insert_value(&mut self, key: Self::Key, value: Self::Value) -> Option<Self::Value>;
/// Returns the value at the specified key. Recall that by definition, any key that hasn't been
/// updated is associated with [`Self::EMPTY_VALUE`].
fn get_value(&self, key: &Self::Key) -> Self::Value;
/// Returns the leaf at the specified index.
fn get_leaf(&self, key: &Self::Key) -> Self::Leaf;
/// Returns the hash of a leaf
fn hash_leaf(leaf: &Self::Leaf) -> RpoDigest;
/// Returns what a leaf would look like if a key-value pair were inserted into the tree, without
/// mutating the tree itself. The existing leaf can be empty.
///
/// To get a prospective leaf based on the current state of the tree, use `self.get_leaf(key)`
/// as the argument for `existing_leaf`. The return value from this function can be chained back
/// into this function as the first argument to continue making prospective changes.
///
/// # Invariants
/// Because this method is for a prospective key-value insertion into a specific leaf,
/// `existing_leaf` must have the same leaf index as `key` (as determined by
/// [`SparseMerkleTree::key_to_leaf_index()`]), or the result will be meaningless.
fn construct_prospective_leaf(
&self,
existing_leaf: Self::Leaf,
key: &Self::Key,
value: &Self::Value,
) -> Self::Leaf;
/// Maps a key to a leaf index
fn key_to_leaf_index(key: &Self::Key) -> LeafIndex<DEPTH>;
/// Constructs a single leaf from an arbitrary amount of key-value pairs.
/// Those pairs must all have the same leaf index.
fn pairs_to_leaf(pairs: Vec<(Self::Key, Self::Value)>) -> Self::Leaf;
/// Maps a (MerklePath, Self::Leaf) to an opening.
///
/// The length `path` is guaranteed to be equal to `DEPTH`
fn path_and_leaf_to_opening(path: MerklePath, leaf: Self::Leaf) -> Self::Opening;
/// Performs the initial transforms for constructing a [`SparseMerkleTree`] by composing
/// subtrees. In other words, this function takes the key-value inputs to the tree, and produces
/// the inputs to feed into [`SparseMerkleTree::build_subtree()`].
///
/// `pairs` *must* already be sorted **by leaf index column**, not simply sorted by key. If
/// `pairs` is not correctly sorted, the returned computations will be incorrect.
///
/// # Panics
/// With debug assertions on, this function panics if it detects that `pairs` is not correctly
/// sorted. Without debug assertions, the returned computations will be incorrect.
fn sorted_pairs_to_leaves(
pairs: Vec<(Self::Key, Self::Value)>,
) -> PairComputations<u64, Self::Leaf> {
debug_assert!(pairs.is_sorted_by_key(|(key, _)| Self::key_to_leaf_index(key).value()));
let mut accumulator: PairComputations<u64, Self::Leaf> = Default::default();
let mut accumulated_leaves: Vec<SubtreeLeaf> = Default::default();
// As we iterate, we'll keep track of the kv-pairs we've seen so far that correspond to a
// single leaf. When we see a pair that's in a different leaf, we'll swap these pairs
// out and store them in our accumulated leaves.
let mut current_leaf_buffer: Vec<(Self::Key, Self::Value)> = Default::default();
let mut iter = pairs.into_iter().peekable();
while let Some((key, value)) = iter.next() {
let col = Self::key_to_leaf_index(&key).index.value();
let peeked_col = iter.peek().map(|(key, _v)| {
let index = Self::key_to_leaf_index(key);
let next_col = index.index.value();
// We panic if `pairs` is not sorted by column.
debug_assert!(next_col >= col);
next_col
});
current_leaf_buffer.push((key, value));
// If the next pair is the same column as this one, then we're done after adding this
// pair to the buffer.
if peeked_col == Some(col) {
continue;
}
// Otherwise, the next pair is a different column, or there is no next pair. Either way
// it's time to swap out our buffer.
let leaf_pairs = mem::take(&mut current_leaf_buffer);
let leaf = Self::pairs_to_leaf(leaf_pairs);
let hash = Self::hash_leaf(&leaf);
accumulator.nodes.insert(col, leaf);
accumulated_leaves.push(SubtreeLeaf { col, hash });
debug_assert!(current_leaf_buffer.is_empty());
}
// TODO: determine is there is any notable performance difference between computing
// subtree boundaries after the fact as an iterator adapter (like this), versus computing
// subtree boundaries as we go. Either way this function is only used at the beginning of a
// parallel construction, so it should not be a critical path.
accumulator.leaves = SubtreeLeavesIter::from_leaves(&mut accumulated_leaves).collect();
accumulator
}
/// Builds Merkle nodes from a bottom layer of "leaves" -- represented by a horizontal index and
/// the hash of the leaf at that index. `leaves` *must* be sorted by horizontal index, and
/// `leaves` must not contain more than one depth-8 subtree's worth of leaves.
///
/// This function will then calculate the inner nodes above each leaf for 8 layers, as well as
/// the "leaves" for the next 8-deep subtree, so this function can effectively be chained into
/// itself.
///
/// # Panics
/// With debug assertions on, this function panics under invalid inputs: if `leaves` contains
/// more entries than can fit in a depth-8 subtree, if `leaves` contains leaves belonging to
/// different depth-8 subtrees, if `bottom_depth` is lower in the tree than the specified
/// maximum depth (`DEPTH`), or if `leaves` is not sorted.
fn build_subtree(
mut leaves: Vec<SubtreeLeaf>,
bottom_depth: u8,
) -> (BTreeMap<NodeIndex, InnerNode>, Vec<SubtreeLeaf>) {
debug_assert!(bottom_depth <= DEPTH);
debug_assert!(Integer::is_multiple_of(&bottom_depth, &SUBTREE_DEPTH));
debug_assert!(leaves.len() <= usize::pow(2, SUBTREE_DEPTH as u32));
let subtree_root = bottom_depth - SUBTREE_DEPTH;
let mut inner_nodes: BTreeMap<NodeIndex, InnerNode> = Default::default();
let mut next_leaves: Vec<SubtreeLeaf> = Vec::with_capacity(leaves.len() / 2);
for next_depth in (subtree_root..bottom_depth).rev() {
debug_assert!(next_depth <= bottom_depth);
// `next_depth` is the stuff we're making.
// `current_depth` is the stuff we have.
let current_depth = next_depth + 1;
let mut iter = leaves.drain(..).peekable();
while let Some(first) = iter.next() {
// On non-continuous iterations, including the first iteration, `first_column` may
// be a left or right node. On subsequent continuous iterations, we will always call
// `iter.next()` twice.
// On non-continuous iterations (including the very first iteration), this column
// could be either on the left or the right. If the next iteration is not
// discontinuous with our right node, then the next iteration's
let is_right = first.col.is_odd();
let (left, right) = if is_right {
// Discontinuous iteration: we have no left node, so it must be empty.
let left = SubtreeLeaf {
col: first.col - 1,
hash: *EmptySubtreeRoots::entry(DEPTH, current_depth),
};
let right = first;
(left, right)
} else {
let left = first;
let right_col = first.col + 1;
let right = match iter.peek().copied() {
Some(SubtreeLeaf { col, .. }) if col == right_col => {
// Our inputs must be sorted.
debug_assert!(left.col <= col);
// The next leaf in the iterator is our sibling. Use it and consume it!
iter.next().unwrap()
},
// Otherwise, the leaves don't contain our sibling, so our sibling must be
// empty.
_ => SubtreeLeaf {
col: right_col,
hash: *EmptySubtreeRoots::entry(DEPTH, current_depth),
},
};
(left, right)
};
let index = NodeIndex::new_unchecked(current_depth, left.col).parent();
let node = InnerNode { left: left.hash, right: right.hash };
let hash = node.hash();
let &equivalent_empty_hash = EmptySubtreeRoots::entry(DEPTH, next_depth);
// If this hash is empty, then it doesn't become a new inner node, nor does it count
// as a leaf for the next depth.
if hash != equivalent_empty_hash {
inner_nodes.insert(index, node);
next_leaves.push(SubtreeLeaf { col: index.value(), hash });
}
}
// Stop borrowing `leaves`, so we can swap it.
// The iterator is empty at this point anyway.
drop(iter);
// After each depth, consider the stuff we just made the new "leaves", and empty the
// other collection.
mem::swap(&mut leaves, &mut next_leaves);
}
(inner_nodes, leaves)
}
}
// INNER NODE
// ================================================================================================
/// This struct is public so functions returning it can be used in `benches/`, but is otherwise not
/// part of the public API.
#[doc(hidden)]
#[derive(Debug, Default, Clone, PartialEq, Eq)]
#[cfg_attr(feature = "serde", derive(serde::Deserialize, serde::Serialize))]
pub struct InnerNode {
pub left: RpoDigest,
pub right: RpoDigest,
}
impl InnerNode {
pub fn hash(&self) -> RpoDigest {
Rpo256::merge(&[self.left, self.right])
}
}
// LEAF INDEX
// ================================================================================================
/// The index of a leaf, at a depth known at compile-time.
#[derive(Debug, Default, Copy, Clone, Eq, PartialEq, PartialOrd, Ord, Hash)]
#[cfg_attr(feature = "serde", derive(serde::Deserialize, serde::Serialize))]
pub struct LeafIndex<const DEPTH: u8> {
index: NodeIndex,
}
impl<const DEPTH: u8> LeafIndex<DEPTH> {
pub fn new(value: u64) -> Result<Self, MerkleError> {
if DEPTH < SMT_MIN_DEPTH {
return Err(MerkleError::DepthTooSmall(DEPTH));
}
Ok(LeafIndex { index: NodeIndex::new(DEPTH, value)? })
}
pub fn value(&self) -> u64 {
self.index.value()
}
}
impl LeafIndex<SMT_MAX_DEPTH> {
pub const fn new_max_depth(value: u64) -> Self {
LeafIndex {
index: NodeIndex::new_unchecked(SMT_MAX_DEPTH, value),
}
}
}
impl<const DEPTH: u8> From<LeafIndex<DEPTH>> for NodeIndex {
fn from(value: LeafIndex<DEPTH>) -> Self {
value.index
}
}
impl<const DEPTH: u8> TryFrom<NodeIndex> for LeafIndex<DEPTH> {
type Error = MerkleError;
fn try_from(node_index: NodeIndex) -> Result<Self, Self::Error> {
if node_index.depth() != DEPTH {
return Err(MerkleError::InvalidDepth {
expected: DEPTH,
provided: node_index.depth(),
});
}
Self::new(node_index.value())
}
}
// MUTATIONS
// ================================================================================================
/// A change to an inner node of a [`SparseMerkleTree`] that hasn't yet been applied.
/// [`MutationSet`] stores this type in relation to a [`NodeIndex`] to keep track of what changes
/// need to occur at which node indices.
#[derive(Debug, Clone, PartialEq, Eq)]
pub(crate) enum NodeMutation {
/// Corresponds to [`SparseMerkleTree::remove_inner_node()`].
Removal,
/// Corresponds to [`SparseMerkleTree::insert_inner_node()`].
Addition(InnerNode),
}
/// Represents a group of prospective mutations to a `SparseMerkleTree`, created by
/// `SparseMerkleTree::compute_mutations()`, and that can be applied with
/// `SparseMerkleTree::apply_mutations()`.
#[derive(Debug, Clone, PartialEq, Eq, Default)]
pub struct MutationSet<const DEPTH: u8, K, V> {
/// The root of the Merkle tree this MutationSet is for, recorded at the time
/// [`SparseMerkleTree::compute_mutations()`] was called. Exists to guard against applying
/// mutations to the wrong tree or applying stale mutations to a tree that has since changed.
old_root: RpoDigest,
/// The set of nodes that need to be removed or added. The "effective" node at an index is the
/// Merkle tree's existing node at that index, with the [`NodeMutation`] in this map at that
/// index overlayed, if any. Each [`NodeMutation::Addition`] corresponds to a
/// [`SparseMerkleTree::insert_inner_node()`] call, and each [`NodeMutation::Removal`]
/// corresponds to a [`SparseMerkleTree::remove_inner_node()`] call.
node_mutations: BTreeMap<NodeIndex, NodeMutation>,
/// The set of top-level key-value pairs we're prospectively adding to the tree, including
/// adding empty values. The "effective" value for a key is the value in this BTreeMap, falling
/// back to the existing value in the Merkle tree. Each entry corresponds to a
/// [`SparseMerkleTree::insert_value()`] call.
new_pairs: BTreeMap<K, V>,
/// The calculated root for the Merkle tree, given these mutations. Publicly retrievable with
/// [`MutationSet::root()`]. Corresponds to a [`SparseMerkleTree::set_root()`]. call.
new_root: RpoDigest,
}
impl<const DEPTH: u8, K, V> MutationSet<DEPTH, K, V> {
/// Queries the root that was calculated during `SparseMerkleTree::compute_mutations()`. See
/// that method for more information.
pub fn root(&self) -> RpoDigest {
self.new_root
}
}
// SUBTREES
// ================================================================================================
/// A subtree is of depth 8.
const SUBTREE_DEPTH: u8 = 8;
/// A depth-8 subtree contains 256 "columns" that can possibly be occupied.
const COLS_PER_SUBTREE: u64 = u64::pow(2, SUBTREE_DEPTH as u32);
/// Helper struct for organizing the data we care about when computing Merkle subtrees.
///
/// Note that these represet "conceptual" leaves of some subtree, not necessarily
/// [`SparseMerkleTree::Leaf`].
#[derive(Debug, Copy, Clone, PartialEq, Eq, PartialOrd, Ord, Default)]
pub struct SubtreeLeaf {
/// The 'value' field of [`NodeIndex`]. When computing a subtree, the depth is already known.
pub col: u64,
/// The hash of the node this `SubtreeLeaf` represents.
pub hash: RpoDigest,
}
/// Helper struct to organize the return value of [`SparseMerkleTree::sorted_pairs_to_leaves()`].
#[derive(Debug, Clone, PartialEq, Eq)]
pub(crate) struct PairComputations<K, L> {
/// Literal leaves to be added to the sparse Merkle tree's internal mapping.
pub nodes: BTreeMap<K, L>,
/// "Conceptual" leaves that will be used for computations.
pub leaves: Vec<Vec<SubtreeLeaf>>,
}
// Derive requires `L` to impl Default, even though we don't actually need that.
impl<K, L> Default for PairComputations<K, L> {
fn default() -> Self {
Self {
nodes: Default::default(),
leaves: Default::default(),
}
}
}
#[derive(Debug)]
struct SubtreeLeavesIter<'s> {
leaves: core::iter::Peekable<alloc::vec::Drain<'s, SubtreeLeaf>>,
}
impl<'s> SubtreeLeavesIter<'s> {
fn from_leaves(leaves: &'s mut Vec<SubtreeLeaf>) -> Self {
Self { leaves: leaves.drain(..).peekable() }
}
}
impl<'s> core::iter::Iterator for SubtreeLeavesIter<'s> {
type Item = Vec<SubtreeLeaf>;
/// Each `next()` collects an entire subtree.
fn next(&mut self) -> Option<Vec<SubtreeLeaf>> {
let mut subtree: Vec<SubtreeLeaf> = Default::default();
let mut last_subtree_col = 0;
while let Some(leaf) = self.leaves.peek() {
last_subtree_col = u64::max(1, last_subtree_col);
let is_exact_multiple = Integer::is_multiple_of(&last_subtree_col, &COLS_PER_SUBTREE);
let next_subtree_col = if is_exact_multiple {
u64::next_multiple_of(last_subtree_col + 1, COLS_PER_SUBTREE)
} else {
last_subtree_col.next_multiple_of(COLS_PER_SUBTREE)
};
last_subtree_col = leaf.col;
if leaf.col < next_subtree_col {
subtree.push(self.leaves.next().unwrap());
} else if subtree.is_empty() {
continue;
} else {
break;
}
}
if subtree.is_empty() {
debug_assert!(self.leaves.peek().is_none());
return None;
}
Some(subtree)
}
}
// TESTS
// ================================================================================================
#[cfg(test)]
mod test {
use alloc::{collections::BTreeMap, vec::Vec};
use super::{
PairComputations, SmtLeaf, SparseMerkleTree, SubtreeLeaf, SubtreeLeavesIter,
COLS_PER_SUBTREE, SUBTREE_DEPTH,
};
use crate::{
hash::rpo::RpoDigest,
merkle::{NodeIndex, Smt, SMT_DEPTH},
Felt, Word, ONE,
};
fn smtleaf_to_subtree_leaf(leaf: &SmtLeaf) -> SubtreeLeaf {
SubtreeLeaf {
col: leaf.index().index.value(),
hash: leaf.hash(),
}
}
#[test]
fn test_sorted_pairs_to_leaves() {
let entries: Vec<(RpoDigest, Word)> = vec![
// Subtree 0.
(RpoDigest::new([ONE, ONE, ONE, Felt::new(16)]), [ONE; 4]),
(RpoDigest::new([ONE, ONE, ONE, Felt::new(17)]), [ONE; 4]),
// Leaf index collision.
(RpoDigest::new([ONE, ONE, Felt::new(10), Felt::new(20)]), [ONE; 4]),
(RpoDigest::new([ONE, ONE, Felt::new(20), Felt::new(20)]), [ONE; 4]),
// Subtree 1. Normal single leaf again.
(RpoDigest::new([ONE, ONE, ONE, Felt::new(400)]), [ONE; 4]), // Subtree boundary.
(RpoDigest::new([ONE, ONE, ONE, Felt::new(401)]), [ONE; 4]),
// Subtree 2. Another normal leaf.
(RpoDigest::new([ONE, ONE, ONE, Felt::new(1024)]), [ONE; 4]),
];
let control = Smt::with_entries(entries.clone()).unwrap();
let control_leaves: Vec<SmtLeaf> = {
let mut entries_iter = entries.iter().cloned();
let mut next_entry = || entries_iter.next().unwrap();
let control_leaves = vec![
// Subtree 0.
SmtLeaf::Single(next_entry()),
SmtLeaf::Single(next_entry()),
SmtLeaf::new_multiple(vec![next_entry(), next_entry()]).unwrap(),
// Subtree 1.
SmtLeaf::Single(next_entry()),
SmtLeaf::Single(next_entry()),
// Subtree 2.
SmtLeaf::Single(next_entry()),
];
assert_eq!(entries_iter.next(), None);
control_leaves
};
let control_subtree_leaves: Vec<Vec<SubtreeLeaf>> = {
let mut control_leaves_iter = control_leaves.iter();
let mut next_leaf = || control_leaves_iter.next().unwrap();
let control_subtree_leaves: Vec<Vec<SubtreeLeaf>> = [
// Subtree 0.
vec![next_leaf(), next_leaf(), next_leaf()],
// Subtree 1.
vec![next_leaf(), next_leaf()],
// Subtree 2.
vec![next_leaf()],
]
.map(|subtree| subtree.into_iter().map(smtleaf_to_subtree_leaf).collect())
.to_vec();
assert_eq!(control_leaves_iter.next(), None);
control_subtree_leaves
};
let subtrees: PairComputations<u64, SmtLeaf> = Smt::sorted_pairs_to_leaves(entries);
// This will check that the hashes, columns, and subtree assignments all match.
assert_eq!(subtrees.leaves, control_subtree_leaves);
// Flattening and re-separating out the leaves into subtrees should have the same result.
let mut all_leaves: Vec<SubtreeLeaf> =
subtrees.leaves.clone().into_iter().flatten().collect();
let re_grouped: Vec<Vec<_>> = SubtreeLeavesIter::from_leaves(&mut all_leaves).collect();
assert_eq!(subtrees.leaves, re_grouped);
// Then finally we might as well check the computed leaf nodes too.
let control_leaves: BTreeMap<u64, SmtLeaf> = control
.leaves()
.map(|(index, value)| (index.index.value(), value.clone()))
.collect();
for (column, test_leaf) in subtrees.nodes {
if test_leaf.is_empty() {
continue;
}
let control_leaf = control_leaves
.get(&column)
.unwrap_or_else(|| panic!("no leaf node found for column {column}"));
assert_eq!(control_leaf, &test_leaf);
}
}
// Helper for the below tests.
fn generate_entries(pair_count: u64) -> Vec<(RpoDigest, Word)> {
(0..pair_count)
.map(|i| {
let leaf_index = ((i as f64 / pair_count as f64) * (pair_count as f64)) as u64;
let key = RpoDigest::new([ONE, ONE, Felt::new(i), Felt::new(leaf_index)]);
let value = [ONE, ONE, ONE, Felt::new(i)];
(key, value)
})
.collect()
}
#[test]
fn test_single_subtree() {
// A single subtree's worth of leaves.
const PAIR_COUNT: u64 = COLS_PER_SUBTREE;
let entries = generate_entries(PAIR_COUNT);
let control = Smt::with_entries(entries.clone()).unwrap();
// `entries` should already be sorted by nature of how we constructed it.
let leaves = Smt::sorted_pairs_to_leaves(entries).leaves;
let leaves = leaves.into_iter().next().unwrap();
let (first_subtree, next_leaves) = Smt::build_subtree(leaves, SMT_DEPTH);
assert!(!first_subtree.is_empty());
// The inner nodes computed from that subtree should match the nodes in our control tree.
for (index, node) in first_subtree.into_iter() {
let control = control.get_inner_node(index);
assert_eq!(
control, node,
"subtree-computed node at index {index:?} does not match control",
);
}
// The "next leaves" returned should also have matching hashes from the equivalent nodes in
// our control tree.
for SubtreeLeaf { col, hash } in next_leaves {
let index = NodeIndex::new(SMT_DEPTH - SUBTREE_DEPTH, col).unwrap();
let control_node = control.get_inner_node(index);
let control = control_node.hash();
assert_eq!(
control, hash,
"subtree-computed next leaf at index {index:?} does not match control",
);
}
}
}