miden-crypto/src/merkle/smt/full/mod.rs

use alloc::{
    collections::{BTreeMap, BTreeSet},
    string::ToString,
    vec::Vec,
};

use super::{
    EmptySubtreeRoots, Felt, InnerNode, InnerNodeInfo, LeafIndex, MerkleError, MerklePath,
    NodeIndex, Rpo256, RpoDigest, SparseMerkleTree, Word, EMPTY_WORD,
};

mod error;
pub use error::{SmtLeafError, SmtProofError};

mod leaf;
pub use leaf::SmtLeaf;

mod proof;
pub use proof::SmtProof;
use winter_utils::{ByteReader, ByteWriter, Deserializable, DeserializationError, Serializable};

#[cfg(test)]
mod tests;

// CONSTANTS
// ================================================================================================

pub const SMT_DEPTH: u8 = 64;

// SMT
// ================================================================================================

/// Sparse Merkle tree mapping 256-bit keys to 256-bit values. Both keys and values are represented
/// by 4 field elements.
///
/// All leaves sit at depth 64. The most significant element of the key is used to identify the leaf
/// to which the key maps.
///
/// A leaf is either empty, or holds one or more key-value pairs. An empty leaf hashes to the empty
/// word. Otherwise, a leaf hashes to the hash of its key-value pairs, ordered by key first, value
/// second.
#[derive(Debug, Clone, PartialEq, Eq)]
#[cfg_attr(feature = "serde", derive(serde::Deserialize, serde::Serialize))]
pub struct Smt {
    root: RpoDigest,
    leaves: BTreeMap<u64, SmtLeaf>,
    inner_nodes: BTreeMap<NodeIndex, InnerNode>,
}

impl Smt {
    // CONSTANTS
    // --------------------------------------------------------------------------------------------
    /// The default value used to compute the hash of empty leaves
    pub const EMPTY_VALUE: Word = <Self as SparseMerkleTree<SMT_DEPTH>>::EMPTY_VALUE;

    // CONSTRUCTORS
    // --------------------------------------------------------------------------------------------

    /// Returns a new [Smt].
    ///
    /// All leaves in the returned tree are set to [Self::EMPTY_VALUE].
    pub fn new() -> Self {
        let root = *EmptySubtreeRoots::entry(SMT_DEPTH, 0);

        Self {
            root,
            leaves: BTreeMap::new(),
            inner_nodes: BTreeMap::new(),
        }
    }

    /// Returns a new [Smt] instantiated with leaves set as specified by the provided entries.
    ///
    /// All leaves omitted from the entries list are set to [Self::EMPTY_VALUE].
    ///
    /// # Errors
    /// Returns an error if the provided entries contain multiple values for the same key.
    pub fn with_entries(
        entries: impl IntoIterator<Item = (RpoDigest, Word)>,
    ) -> Result<Self, MerkleError> {
        // create an empty tree
        let mut tree = Self::new();

        // This being a sparse data structure, the EMPTY_WORD is not assigned to the `BTreeMap`, so
        // entries with the empty value need additional tracking.
        let mut key_set_to_zero = BTreeSet::new();

        for (key, value) in entries {
            let old_value = tree.insert(key, value);

            if old_value != EMPTY_WORD || key_set_to_zero.contains(&key) {
                return Err(MerkleError::DuplicateValuesForIndex(
                    LeafIndex::<SMT_DEPTH>::from(key).value(),
                ));
            }

            if value == EMPTY_WORD {
                key_set_to_zero.insert(key);
            };
        }
        Ok(tree)
    }

    // PUBLIC ACCESSORS
    // --------------------------------------------------------------------------------------------

    /// Returns the depth of the tree
    pub const fn depth(&self) -> u8 {
        SMT_DEPTH
    }

    /// Returns the root of the tree
    pub fn root(&self) -> RpoDigest {
        <Self as SparseMerkleTree<SMT_DEPTH>>::root(self)
    }

    /// Returns the leaf to which `key` maps
    pub fn get_leaf(&self, key: &RpoDigest) -> SmtLeaf {
        <Self as SparseMerkleTree<SMT_DEPTH>>::get_leaf(self, key)
    }

    /// Returns the value associated with `key`
    pub fn get_value(&self, key: &RpoDigest) -> Word {
        <Self as SparseMerkleTree<SMT_DEPTH>>::get_value(self, key)
    }

    /// 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.
    pub fn open(&self, key: &RpoDigest) -> SmtProof {
        <Self as SparseMerkleTree<SMT_DEPTH>>::open(self, key)
    }

    // ITERATORS
    // --------------------------------------------------------------------------------------------

    /// Returns an iterator over the leaves of this [Smt].
    pub fn leaves(&self) -> impl Iterator<Item = (LeafIndex<SMT_DEPTH>, &SmtLeaf)> {
        self.leaves
            .iter()
            .map(|(leaf_index, leaf)| (LeafIndex::new_max_depth(*leaf_index), leaf))
    }

    /// Returns an iterator over the key-value pairs of this [Smt].
    pub fn entries(&self) -> impl Iterator<Item = &(RpoDigest, Word)> {
        self.leaves().flat_map(|(_, leaf)| leaf.entries())
    }

    /// Returns an iterator over the inner nodes of this [Smt].
    pub fn inner_nodes(&self) -> impl Iterator<Item = InnerNodeInfo> + '_ {
        self.inner_nodes.values().map(|e| InnerNodeInfo {
            value: e.hash(),
            left: e.left,
            right: e.right,
        })
    }

    // STATE MUTATORS
    // --------------------------------------------------------------------------------------------

    /// 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.
    pub fn insert(&mut self, key: RpoDigest, value: Word) -> Word {
        <Self as SparseMerkleTree<SMT_DEPTH>>::insert(self, key, value)
    }

    /// Like [`Self::insert()`], but only performs the insert if the the new tree's root
    /// hash would be equal to the hash given in `expected_root`.
    ///
    /// # Errors
    /// Returns [`MerkleError::ConflictingRoots`] with a two-item [Vec] if the new root of the tree
    /// is different from the expected root. The first item of the vector is the expected root,
    /// and the second is actual root.
    ///
    /// No mutations are performed if the roots do no match.
    pub fn insert_ensure_root(
        &mut self,
        key: RpoDigest,
        value: Word,
        expected_root: RpoDigest,
    ) -> Result<Word, MerkleError> {
        <Self as SparseMerkleTree<SMT_DEPTH>>::insert_ensure_root(self, key, value, expected_root)
    }

    // HELPERS
    // --------------------------------------------------------------------------------------------

    /// Inserts `value` at leaf index pointed to by `key`. `value` is guaranteed to not be the empty
    /// value, such that this is indeed an insertion.
    fn perform_insert(&mut self, key: RpoDigest, value: Word) -> Option<Word> {
        debug_assert_ne!(value, Self::EMPTY_VALUE);

        let leaf_index: LeafIndex<SMT_DEPTH> = Self::key_to_leaf_index(&key);

        match self.leaves.get_mut(&leaf_index.value()) {
            Some(leaf) => leaf.insert(key, value),
            None => {
                self.leaves.insert(leaf_index.value(), SmtLeaf::Single((key, value)));

                None
            },
        }
    }

    /// Removes key-value pair at leaf index pointed to by `key` if it exists.
    fn perform_remove(&mut self, key: RpoDigest) -> Option<Word> {
        let leaf_index: LeafIndex<SMT_DEPTH> = Self::key_to_leaf_index(&key);

        if let Some(leaf) = self.leaves.get_mut(&leaf_index.value()) {
            let (old_value, is_empty) = leaf.remove(key);
            if is_empty {
                self.leaves.remove(&leaf_index.value());
            }
            old_value
        } else {
            // there's nothing stored at the leaf; nothing to update
            None
        }
    }
}

impl SparseMerkleTree<SMT_DEPTH> for Smt {
    type Key = RpoDigest;
    type Value = Word;
    type Leaf = SmtLeaf;
    type Opening = SmtProof;

    const EMPTY_VALUE: Self::Value = EMPTY_WORD;

    fn root(&self) -> RpoDigest {
        self.root
    }

    fn set_root(&mut self, root: RpoDigest) {
        self.root = root;
    }

    fn get_inner_node(&self, index: NodeIndex) -> InnerNode {
        self.inner_nodes.get(&index).cloned().unwrap_or_else(|| {
            let node = EmptySubtreeRoots::entry(SMT_DEPTH, index.depth() + 1);

            InnerNode { left: *node, right: *node }
        })
    }

    fn insert_inner_node(&mut self, index: NodeIndex, inner_node: InnerNode) {
        self.inner_nodes.insert(index, inner_node);
    }

    fn remove_inner_node(&mut self, index: NodeIndex) {
        let _ = self.inner_nodes.remove(&index);
    }

    fn insert_value(&mut self, key: Self::Key, value: Self::Value) -> Option<Self::Value> {
        // inserting an `EMPTY_VALUE` is equivalent to removing any value associated with `key`
        if value != Self::EMPTY_VALUE {
            self.perform_insert(key, value)
        } else {
            self.perform_remove(key)
        }
    }

    fn get_value(&self, key: &Self::Key) -> Self::Value {
        let leaf_pos = LeafIndex::<SMT_DEPTH>::from(*key).value();

        match self.leaves.get(&leaf_pos) {
            Some(leaf) => leaf.get_value(key).unwrap_or_default(),
            None => EMPTY_WORD,
        }
    }

    fn get_leaf(&self, key: &RpoDigest) -> Self::Leaf {
        let leaf_pos = LeafIndex::<SMT_DEPTH>::from(*key).value();

        match self.leaves.get(&leaf_pos) {
            Some(leaf) => leaf.clone(),
            None => SmtLeaf::new_empty(key.into()),
        }
    }

    fn hash_leaf(leaf: &Self::Leaf) -> RpoDigest {
        leaf.hash()
    }

    fn hash_prospective_leaf(&self, key: &RpoDigest, value: &Word) -> RpoDigest {
        // Future work could avoid cloning the leaf by mirroring some of the insertion logic and
        // hashing without an intermediate leaf, but cloning is only expensive for multi-leaves
        // which should be really rare anyway.
        let leaf_index: LeafIndex<SMT_DEPTH> = Self::key_to_leaf_index(key);
        match self.leaves.get(&leaf_index.value()) {
            None => SmtLeaf::new_single(*key, *value).hash(),
            Some(existing_leaf) => {
                let mut new_leaf = existing_leaf.clone();
                if *value != EMPTY_WORD {
                    new_leaf.insert(*key, *value);
                } else {
                    new_leaf.remove(*key);
                }
                new_leaf.hash()
            },
        }
    }

    fn key_to_leaf_index(key: &RpoDigest) -> LeafIndex<SMT_DEPTH> {
        let most_significant_felt = key[3];
        LeafIndex::new_max_depth(most_significant_felt.as_int())
    }

    fn path_and_leaf_to_opening(path: MerklePath, leaf: SmtLeaf) -> SmtProof {
        SmtProof::new_unchecked(path, leaf)
    }
}

impl Default for Smt {
    fn default() -> Self {
        Self::new()
    }
}

// CONVERSIONS
// ================================================================================================

impl From<Word> for LeafIndex<SMT_DEPTH> {
    fn from(value: Word) -> Self {
        // We use the most significant `Felt` of a `Word` as the leaf index.
        Self::new_max_depth(value[3].as_int())
    }
}

impl From<RpoDigest> for LeafIndex<SMT_DEPTH> {
    fn from(value: RpoDigest) -> Self {
        Word::from(value).into()
    }
}

impl From<&RpoDigest> for LeafIndex<SMT_DEPTH> {
    fn from(value: &RpoDigest) -> Self {
        Word::from(value).into()
    }
}

// SERIALIZATION
// ================================================================================================

impl Serializable for Smt {
    fn write_into<W: ByteWriter>(&self, target: &mut W) {
        // Write the number of filled leaves for this Smt
        target.write_usize(self.entries().count());

        // Write each (key, value) pair
        for (key, value) in self.entries() {
            target.write(key);
            target.write(value);
        }
    }
}

impl Deserializable for Smt {
    fn read_from<R: ByteReader>(source: &mut R) -> Result<Self, DeserializationError> {
        // Read the number of filled leaves for this Smt
        let num_filled_leaves = source.read_usize()?;
        let mut entries = Vec::with_capacity(num_filled_leaves);

        for _ in 0..num_filled_leaves {
            let key = source.read()?;
            let value = source.read()?;
            entries.push((key, value));
        }

        Self::with_entries(entries)
            .map_err(|err| DeserializationError::InvalidValue(err.to_string()))
    }
}

#[test]
fn test_smt_serialization_deserialization() {
    // Smt for default types (empty map)
    let smt_default = Smt::default();
    let bytes = smt_default.to_bytes();
    assert_eq!(smt_default, Smt::read_from_bytes(&bytes).unwrap());

    // Smt with values
    let smt_leaves_2: [(RpoDigest, Word); 2] = [
        (
            RpoDigest::new([Felt::new(101), Felt::new(102), Felt::new(103), Felt::new(104)]),
            [Felt::new(1_u64), Felt::new(2_u64), Felt::new(3_u64), Felt::new(4_u64)],
        ),
        (
            RpoDigest::new([Felt::new(105), Felt::new(106), Felt::new(107), Felt::new(108)]),
            [Felt::new(5_u64), Felt::new(6_u64), Felt::new(7_u64), Felt::new(8_u64)],
        ),
    ];
    let smt = Smt::with_entries(smt_leaves_2).unwrap();

    let bytes = smt.to_bytes();
    assert_eq!(smt, Smt::read_from_bytes(&bytes).unwrap());
}