|
|
package merkletree
import ( "fmt" "math/big" )
// NodeType defines the type of node in the MT.
type NodeType byte
const ( // NodeTypeMiddle indicates the type of middle Node that has children.
NodeTypeMiddle NodeType = 0 // NodeTypeLeaf indicates the type of a leaf Node that contains a key &
// value.
NodeTypeLeaf NodeType = 1 // NodeTypeEmpty indicates the type of an empty Node.
NodeTypeEmpty NodeType = 2
// DBEntryTypeRoot indicates the type of a DB entry that indicates the
// current Root of a MerkleTree
DBEntryTypeRoot NodeType = 3 )
// Node is the struct that represents a node in the MT. The node should not be
// modified after creation because the cached key won't be updated.
type Node struct { // Type is the type of node in the tree.
Type NodeType // ChildL is the left child of a middle node.
ChildL *Hash // ChildR is the right child of a middle node.
ChildR *Hash // Entry is the data stored in a leaf node.
Entry [2]*Hash // key is a cache used to avoid recalculating key
key *Hash }
// NewNodeLeaf creates a new leaf node.
func NewNodeLeaf(k, v *Hash) *Node { return &Node{Type: NodeTypeLeaf, Entry: [2]*Hash{k, v}} }
// NewNodeMiddle creates a new middle node.
func NewNodeMiddle(childL *Hash, childR *Hash) *Node { return &Node{Type: NodeTypeMiddle, ChildL: childL, ChildR: childR} }
// NewNodeEmpty creates a new empty node.
func NewNodeEmpty() *Node { return &Node{Type: NodeTypeEmpty} }
// NewNodeFromBytes creates a new node by parsing the input []byte.
func NewNodeFromBytes(b []byte) (*Node, error) { if len(b) < 1 { return nil, ErrNodeBytesBadSize } n := Node{Type: NodeType(b[0])} b = b[1:] switch n.Type { case NodeTypeMiddle: if len(b) != 2*ElemBytesLen { return nil, ErrNodeBytesBadSize } n.ChildL, n.ChildR = &Hash{}, &Hash{} copy(n.ChildL[:], b[:ElemBytesLen]) copy(n.ChildR[:], b[ElemBytesLen:ElemBytesLen*2]) case NodeTypeLeaf: if len(b) != 2*ElemBytesLen { return nil, ErrNodeBytesBadSize } n.Entry = [2]*Hash{{}, {}} copy(n.Entry[0][:], b[0:32]) copy(n.Entry[1][:], b[32:64]) case NodeTypeEmpty: break default: return nil, ErrInvalidNodeFound } return &n, nil }
// LeafKey computes the key of a leaf node given the hIndex and hValue of the
// entry of the leaf.
func LeafKey(k, v *Hash) (*Hash, error) { return HashElemsKey(big.NewInt(1), k.BigInt(), v.BigInt()) }
// Key computes the key of the node by hashing the content in a specific way
// for each type of node. This key is used as the hash of the merkle tree for
// each node.
func (n *Node) Key() (*Hash, error) { if n.key == nil { // Cache the key to avoid repeated hash computations.
// NOTE: We are not using the type to calculate the hash!
switch n.Type { case NodeTypeMiddle: // H(ChildL || ChildR)
var err error n.key, err = HashElems(n.ChildL.BigInt(), n.ChildR.BigInt()) if err != nil { return nil, err } case NodeTypeLeaf: var err error n.key, err = LeafKey(n.Entry[0], n.Entry[1]) if err != nil { return nil, err } case NodeTypeEmpty: // Zero
n.key = &HashZero default: n.key = &HashZero } } return n.key, nil }
// Value returns the value of the node. This is the content that is stored in
// the backend database.
func (n *Node) Value() []byte { switch n.Type { case NodeTypeMiddle: // {Type || ChildL || ChildR}
return append([]byte{byte(n.Type)}, append(n.ChildL[:], n.ChildR[:]...)...) case NodeTypeLeaf: // {Type || Data...}
return append([]byte{byte(n.Type)}, append(n.Entry[0][:], n.Entry[1][:]...)...) case NodeTypeEmpty: // {}
return []byte{} default: return []byte{} } }
// String outputs a string representation of a node (different for each type).
func (n *Node) String() string { switch n.Type { case NodeTypeMiddle: // {Type || ChildL || ChildR}
return fmt.Sprintf("Middle L:%s R:%s", n.ChildL, n.ChildR) case NodeTypeLeaf: // {Type || Data...}
return fmt.Sprintf("Leaf I:%v D:%v", n.Entry[0], n.Entry[1]) case NodeTypeEmpty: // {}
return "Empty" default: return "Invalid Node" } }
|