cmd, core, eth, les, light: track deleted nodes (#25757)

* cmd, core, eth, les, light: track deleted nodes

* trie: add docs

* trie: address comments

* cmd, core, eth, les, light, trie: trie id

* trie: add tests

* trie, core: updates

* trie: fix imports

* trie: add utility print-method for nodeset

* trie: import err

* trie: fix go vet warnings

Co-authored-by: Martin Holst Swende <martin@swende.se>

trie/committer.go

@@ -33,13 +33,15 @@ type leaf struct {
 // insertion order.
 type committer struct {
 	nodes       *NodeSet
+	tracer      *tracer
 	collectLeaf bool
 }
 
 // newCommitter creates a new committer or picks one from the pool.
-func newCommitter(owner common.Hash, collectLeaf bool) *committer {
+func newCommitter(owner common.Hash, tracer *tracer, collectLeaf bool) *committer {
 	return &committer{
 		nodes:       NewNodeSet(owner),
+		tracer:      tracer,
 		collectLeaf: collectLeaf,
 	}
 }
@@ -51,6 +53,20 @@ func (c *committer) Commit(n node) (hashNode, *NodeSet, error) {
 	if err != nil {
 		return nil, nil, err
 	}
+	// Some nodes can be deleted from trie which can't be captured by committer
+	// itself. Iterate all deleted nodes tracked by tracer and marked them as
+	// deleted only if they are present in database previously.
+	for _, path := range c.tracer.deleteList() {
+		// There are a few possibilities for this scenario(the node is deleted
+		// but not present in database previously), for example the node was
+		// embedded in the parent and now deleted from the trie. In this case
+		// it's noop from database's perspective.
+		val := c.tracer.getPrev(path)
+		if len(val) == 0 {
+			continue
+		}
+		c.nodes.markDeleted(path, val)
+	}
 	return h.(hashNode), c.nodes, nil
 }
 
@@ -83,6 +99,12 @@ func (c *committer) commit(path []byte, n node) (node, error) {
 		if hn, ok := hashedNode.(hashNode); ok {
 			return hn, nil
 		}
+		// The short node now is embedded in its parent. Mark the node as
+		// deleted if it's present in database previously. It's equivalent
+		// as deletion from database's perspective.
+		if prev := c.tracer.getPrev(path); len(prev) != 0 {
+			c.nodes.markDeleted(path, prev)
+		}
 		return collapsed, nil
 	case *fullNode:
 		hashedKids, err := c.commitChildren(path, cn)
@@ -96,6 +118,12 @@ func (c *committer) commit(path []byte, n node) (node, error) {
 		if hn, ok := hashedNode.(hashNode); ok {
 			return hn, nil
 		}
+		// The full node now is embedded in its parent. Mark the node as
+		// deleted if it's present in database previously. It's equivalent
+		// as deletion from database's perspective.
+		if prev := c.tracer.getPrev(path); len(prev) != 0 {
+			c.nodes.markDeleted(path, prev)
+		}
 		return collapsed, nil
 	case hashNode:
 		return cn, nil
@@ -161,7 +189,7 @@ func (c *committer) store(path []byte, n node) node {
 		}
 	)
 	// Collect the dirty node to nodeset for return.
-	c.nodes.add(string(path), mnode)
+	c.nodes.markUpdated(path, mnode, c.tracer.getPrev(path))
 
 	// Collect the corresponding leaf node if it's required. We don't check
 	// full node since it's impossible to store value in fullNode. The key

trie/database.go

@@ -795,8 +795,8 @@ func (db *Database) Update(nodes *MergedNodeSet) error {
 	}
 	for _, owner := range order {
 		subset := nodes.sets[owner]
-		for _, path := range subset.paths {
-			n, ok := subset.nodes[path]
+		for _, path := range subset.updates.order {
+			n, ok := subset.updates.nodes[path]
 			if !ok {
 				return fmt.Errorf("missing node %x %v", owner, path)
 			}
@@ -837,6 +837,34 @@ func (db *Database) Size() (common.StorageSize, common.StorageSize) {
 	return db.dirtiesSize + db.childrenSize + metadataSize - metarootRefs, preimageSize
 }
 
+// GetReader retrieves a node reader belonging to the given state root.
+func (db *Database) GetReader(root common.Hash) Reader {
+	return newHashReader(db)
+}
+
+// hashReader is reader of hashDatabase which implements the Reader interface.
+type hashReader struct {
+	db *Database
+}
+
+// newHashReader initializes the hash reader.
+func newHashReader(db *Database) *hashReader {
+	return &hashReader{db: db}
+}
+
+// Node retrieves the trie node with the given node hash.
+// No error will be returned if the node is not found.
+func (reader *hashReader) Node(_ common.Hash, _ []byte, hash common.Hash) (node, error) {
+	return reader.db.node(hash), nil
+}
+
+// NodeBlob retrieves the RLP-encoded trie node blob with the given node hash.
+// No error will be returned if the node is not found.
+func (reader *hashReader) NodeBlob(_ common.Hash, _ []byte, hash common.Hash) ([]byte, error) {
+	blob, _ := reader.db.Node(hash)
+	return blob, nil
+}
+
 // saveCache saves clean state cache to given directory path
 // using specified CPU cores.
 func (db *Database) saveCache(dir string, threads int) error {

trie/iterator.go

@@ -375,7 +375,12 @@ func (it *nodeIterator) resolveHash(hash hashNode, path []byte) (node, error) {
 			}
 		}
 	}
-	return it.trie.resolveHash(hash, path)
+	// Retrieve the specified node from the underlying node reader.
+	// it.trie.resolveAndTrack is not used since in that function the
+	// loaded blob will be tracked, while it's not required here since
+	// all loaded nodes won't be linked to trie at all and track nodes
+	// may lead to out-of-memory issue.
+	return it.trie.reader.node(path, common.BytesToHash(hash))
 }
 
 func (it *nodeIterator) resolveBlob(hash hashNode, path []byte) ([]byte, error) {
@@ -384,7 +389,12 @@ func (it *nodeIterator) resolveBlob(hash hashNode, path []byte) ([]byte, error)
 			return blob, nil
 		}
 	}
-	return it.trie.resolveBlob(hash, path)
+	// Retrieve the specified node from the underlying node reader.
+	// it.trie.resolveAndTrack is not used since in that function the
+	// loaded blob will be tracked, while it's not required here since
+	// all loaded nodes won't be linked to trie at all and track nodes
+	// may lead to out-of-memory issue.
+	return it.trie.reader.nodeBlob(path, common.BytesToHash(hash))
 }
 
 func (st *nodeIteratorState) resolve(it *nodeIterator, path []byte) error {

trie/iterator_test.go

@@ -66,7 +66,7 @@ func TestIterator(t *testing.T) {
 	}
 	db.Update(NewWithNodeSet(nodes))
 
-	trie, _ = New(common.Hash{}, root, db)
+	trie, _ = New(TrieID(root), db)
 	found := make(map[string]string)
 	it := NewIterator(trie.NodeIterator(nil))
 	for it.Next() {
@@ -227,7 +227,7 @@ func TestDifferenceIterator(t *testing.T) {
 	}
 	rootA, nodesA, _ := triea.Commit(false)
 	dba.Update(NewWithNodeSet(nodesA))
-	triea, _ = New(common.Hash{}, rootA, dba)
+	triea, _ = New(TrieID(rootA), dba)
 
 	dbb := NewDatabase(rawdb.NewMemoryDatabase())
 	trieb := NewEmpty(dbb)
@@ -236,7 +236,7 @@ func TestDifferenceIterator(t *testing.T) {
 	}
 	rootB, nodesB, _ := trieb.Commit(false)
 	dbb.Update(NewWithNodeSet(nodesB))
-	trieb, _ = New(common.Hash{}, rootB, dbb)
+	trieb, _ = New(TrieID(rootB), dbb)
 
 	found := make(map[string]string)
 	di, _ := NewDifferenceIterator(triea.NodeIterator(nil), trieb.NodeIterator(nil))
@@ -269,7 +269,7 @@ func TestUnionIterator(t *testing.T) {
 	}
 	rootA, nodesA, _ := triea.Commit(false)
 	dba.Update(NewWithNodeSet(nodesA))
-	triea, _ = New(common.Hash{}, rootA, dba)
+	triea, _ = New(TrieID(rootA), dba)
 
 	dbb := NewDatabase(rawdb.NewMemoryDatabase())
 	trieb := NewEmpty(dbb)
@@ -278,7 +278,7 @@ func TestUnionIterator(t *testing.T) {
 	}
 	rootB, nodesB, _ := trieb.Commit(false)
 	dbb.Update(NewWithNodeSet(nodesB))
-	trieb, _ = New(common.Hash{}, rootB, dbb)
+	trieb, _ = New(TrieID(rootB), dbb)
 
 	di, _ := NewUnionIterator([]NodeIterator{triea.NodeIterator(nil), trieb.NodeIterator(nil)})
 	it := NewIterator(di)
@@ -356,7 +356,7 @@ func testIteratorContinueAfterError(t *testing.T, memonly bool) {
 	}
 	for i := 0; i < 20; i++ {
 		// Create trie that will load all nodes from DB.
-		tr, _ := New(common.Hash{}, tr.Hash(), triedb)
+		tr, _ := New(TrieID(tr.Hash()), triedb)
 
 		// Remove a random node from the database. It can't be the root node
 		// because that one is already loaded.
@@ -445,7 +445,7 @@ func testIteratorContinueAfterSeekError(t *testing.T, memonly bool) {
 	}
 	// Create a new iterator that seeks to "bars". Seeking can't proceed because
 	// the node is missing.
-	tr, _ := New(common.Hash{}, root, triedb)
+	tr, _ := New(TrieID(root), triedb)
 	it := tr.NodeIterator([]byte("bars"))
 	missing, ok := it.Error().(*MissingNodeError)
 	if !ok {
@@ -533,7 +533,7 @@ func makeLargeTestTrie() (*Database, *StateTrie, *loggingDb) {
 	// Create an empty trie
 	logDb := &loggingDb{0, memorydb.New()}
 	triedb := NewDatabase(logDb)
-	trie, _ := NewStateTrie(common.Hash{}, common.Hash{}, triedb)
+	trie, _ := NewStateTrie(TrieID(common.Hash{}), triedb)
 
 	// Fill it with some arbitrary data
 	for i := 0; i < 10000; i++ {

trie/nodeset.go

@@ -18,6 +18,8 @@ package trie
 
 import (
 	"fmt"
+	"reflect"
+	"strings"
 
 	"github.com/ethereum/go-ethereum/common"
 )
@@ -25,18 +27,77 @@ import (
 // memoryNode is all the information we know about a single cached trie node
 // in the memory.
 type memoryNode struct {
-	hash common.Hash // Node hash, computed by hashing rlp value
-	size uint16      // Byte size of the useful cached data
-	node node        // Cached collapsed trie node, or raw rlp data
+	hash common.Hash // Node hash, computed by hashing rlp value, empty for deleted nodes
+	size uint16      // Byte size of the useful cached data, 0 for deleted nodes
+	node node        // Cached collapsed trie node, or raw rlp data, nil for deleted nodes
+}
+
+// memoryNodeSize is the raw size of a memoryNode data structure without any
+// node data included. It's an approximate size, but should be a lot better
+// than not counting them.
+// nolint:unused
+var memoryNodeSize = int(reflect.TypeOf(memoryNode{}).Size())
+
+// memorySize returns the total memory size used by this node.
+// nolint:unused
+func (n *memoryNode) memorySize(key int) int {
+	return int(n.size) + memoryNodeSize + key
+}
+
+// rlp returns the raw rlp encoded blob of the cached trie node, either directly
+// from the cache, or by regenerating it from the collapsed node.
+// nolint:unused
+func (n *memoryNode) rlp() []byte {
+	if node, ok := n.node.(rawNode); ok {
+		return node
+	}
+	return nodeToBytes(n.node)
+}
+
+// obj returns the decoded and expanded trie node, either directly from the cache,
+// or by regenerating it from the rlp encoded blob.
+// nolint:unused
+func (n *memoryNode) obj() node {
+	if node, ok := n.node.(rawNode); ok {
+		return mustDecodeNode(n.hash[:], node)
+	}
+	return expandNode(n.hash[:], n.node)
+}
+
+// nodeWithPrev wraps the memoryNode with the previous node value.
+type nodeWithPrev struct {
+	*memoryNode
+	prev []byte // RLP-encoded previous value, nil means it's non-existent
+}
+
+// unwrap returns the internal memoryNode object.
+// nolint:unused
+func (n *nodeWithPrev) unwrap() *memoryNode {
+	return n.memoryNode
+}
+
+// memorySize returns the total memory size used by this node. It overloads
+// the function in memoryNode by counting the size of previous value as well.
+// nolint: unused
+func (n *nodeWithPrev) memorySize(key int) int {
+	return n.memoryNode.memorySize(key) + len(n.prev)
+}
+
+// nodesWithOrder represents a collection of dirty nodes which includes
+// newly-inserted and updated nodes. The modification order of all nodes
+// is represented by order list.
+type nodesWithOrder struct {
+	order []string                 // the path list of dirty nodes, sort by insertion order
+	nodes map[string]*nodeWithPrev // the map of dirty nodes, keyed by node path
 }
 
 // NodeSet contains all dirty nodes collected during the commit operation.
 // Each node is keyed by path. It's not thread-safe to use.
 type NodeSet struct {
-	owner  common.Hash            // the identifier of the trie
-	paths  []string               // the path of dirty nodes, sort by insertion order
-	nodes  map[string]*memoryNode // the map of dirty nodes, keyed by node path
-	leaves []*leaf                // the list of dirty leaves
+	owner   common.Hash       // the identifier of the trie
+	updates *nodesWithOrder   // the set of updated nodes(newly inserted, updated)
+	deletes map[string][]byte // the map of deleted nodes, keyed by node
+	leaves  []*leaf           // the list of dirty leaves
 }
 
 // NewNodeSet initializes an empty node set to be used for tracking dirty nodes
@@ -45,24 +106,78 @@ type NodeSet struct {
 func NewNodeSet(owner common.Hash) *NodeSet {
 	return &NodeSet{
 		owner: owner,
-		nodes: make(map[string]*memoryNode),
+		updates: &nodesWithOrder{
+			nodes: make(map[string]*nodeWithPrev),
+		},
+		deletes: make(map[string][]byte),
 	}
 }
 
-// add caches node with provided path and node object.
-func (set *NodeSet) add(path string, node *memoryNode) {
-	set.paths = append(set.paths, path)
-	set.nodes[path] = node
+// NewNodeSetWithDeletion initializes the nodeset with provided deletion set.
+func NewNodeSetWithDeletion(owner common.Hash, paths [][]byte, prev [][]byte) *NodeSet {
+	set := NewNodeSet(owner)
+	for i, path := range paths {
+		set.markDeleted(path, prev[i])
+	}
+	return set
 }
 
-// addLeaf caches the provided leaf node.
+// markUpdated marks the node as dirty(newly-inserted or updated) with provided
+// node path, node object along with its previous value.
+func (set *NodeSet) markUpdated(path []byte, node *memoryNode, prev []byte) {
+	set.updates.order = append(set.updates.order, string(path))
+	set.updates.nodes[string(path)] = &nodeWithPrev{
+		memoryNode: node,
+		prev:       prev,
+	}
+}
+
+// markDeleted marks the node as deleted with provided path and previous value.
+func (set *NodeSet) markDeleted(path []byte, prev []byte) {
+	set.deletes[string(path)] = prev
+}
+
+// addLeaf collects the provided leaf node into set.
 func (set *NodeSet) addLeaf(node *leaf) {
 	set.leaves = append(set.leaves, node)
 }
 
-// Len returns the number of dirty nodes contained in the set.
-func (set *NodeSet) Len() int {
-	return len(set.nodes)
+// Size returns the number of updated and deleted nodes contained in the set.
+func (set *NodeSet) Size() (int, int) {
+	return len(set.updates.order), len(set.deletes)
+}
+
+// Hashes returns the hashes of all updated nodes. TODO(rjl493456442) how can
+// we get rid of it?
+func (set *NodeSet) Hashes() []common.Hash {
+	var ret []common.Hash
+	for _, node := range set.updates.nodes {
+		ret = append(ret, node.hash)
+	}
+	return ret
+}
+
+// Summary returns a string-representation of the NodeSet.
+func (set *NodeSet) Summary() string {
+	var out = new(strings.Builder)
+	fmt.Fprintf(out, "nodeset owner: %v\n", set.owner)
+	if set.updates != nil {
+		for _, key := range set.updates.order {
+			updated := set.updates.nodes[key]
+			if updated.prev != nil {
+				fmt.Fprintf(out, "  [*]: %x -> %v prev: %x\n", key, updated.hash, updated.prev)
+			} else {
+				fmt.Fprintf(out, "  [+]: %x -> %v\n", key, updated.hash)
+			}
+		}
+	}
+	for k, n := range set.deletes {
+		fmt.Fprintf(out, "  [-]: %x -> %x\n", k, n)
+	}
+	for _, n := range set.leaves {
+		fmt.Fprintf(out, "[leaf]: %v\n", n)
+	}
+	return out.String()
 }
 
 // MergedNodeSet represents a merged dirty node set for a group of tries.

trie/proof.go

@@ -22,7 +22,6 @@ import (
 	"fmt"
 
 	"github.com/ethereum/go-ethereum/common"
-	"github.com/ethereum/go-ethereum/core/rawdb"
 	"github.com/ethereum/go-ethereum/ethdb"
 	"github.com/ethereum/go-ethereum/log"
 )
@@ -60,8 +59,13 @@ func (t *Trie) Prove(key []byte, fromLevel uint, proofDb ethdb.KeyValueWriter) e
 			key = key[1:]
 			nodes = append(nodes, n)
 		case hashNode:
+			// Retrieve the specified node from the underlying node reader.
+			// trie.resolveAndTrack is not used since in that function the
+			// loaded blob will be tracked, while it's not required here since
+			// all loaded nodes won't be linked to trie at all and track nodes
+			// may lead to out-of-memory issue.
 			var err error
-			tn, err = t.resolveHash(n, prefix)
+			tn, err = t.reader.node(prefix, common.BytesToHash(n))
 			if err != nil {
 				log.Error("Unhandled trie error in Trie.Prove", "err", err)
 				return err
@@ -559,7 +563,7 @@ func VerifyRangeProof(rootHash common.Hash, firstKey []byte, lastKey []byte, key
 	}
 	// Rebuild the trie with the leaf stream, the shape of trie
 	// should be same with the original one.
-	tr := &Trie{root: root, db: NewDatabase(rawdb.NewMemoryDatabase())}
+	tr := &Trie{root: root, reader: newEmptyReader()}
 	if empty {
 		tr.root = nil
 	}

trie/secure_trie.go

@@ -29,8 +29,13 @@ type SecureTrie = StateTrie
 
 // NewSecure creates a new StateTrie.
 // Deprecated: use NewStateTrie.
-func NewSecure(owner common.Hash, root common.Hash, db *Database) (*SecureTrie, error) {
-	return NewStateTrie(owner, root, db)
+func NewSecure(stateRoot common.Hash, owner common.Hash, root common.Hash, db *Database) (*SecureTrie, error) {
+	id := &ID{
+		StateRoot: stateRoot,
+		Owner:     owner,
+		Root:      root,
+	}
+	return NewStateTrie(id, db)
 }
 
 // StateTrie wraps a trie with key hashing. In a stateTrie trie, all
@@ -56,11 +61,11 @@ type StateTrie struct {
 // If root is the zero hash or the sha3 hash of an empty string, the
 // trie is initially empty. Otherwise, New will panic if db is nil
 // and returns MissingNodeError if the root node cannot be found.
-func NewStateTrie(owner common.Hash, root common.Hash, db *Database) (*StateTrie, error) {
+func NewStateTrie(id *ID, db *Database) (*StateTrie, error) {
 	if db == nil {
 		panic("trie.NewStateTrie called without a database")
 	}
-	trie, err := New(owner, root, db)
+	trie, err := New(id, db)
 	if err != nil {
 		return nil, err
 	}

trie/secure_trie_test.go

@@ -29,7 +29,7 @@ import (
 )
 
 func newEmptySecure() *StateTrie {
-	trie, _ := NewStateTrie(common.Hash{}, common.Hash{}, NewDatabase(memorydb.New()))
+	trie, _ := NewStateTrie(TrieID(common.Hash{}), NewDatabase(memorydb.New()))
 	return trie
 }
 
@@ -37,7 +37,7 @@ func newEmptySecure() *StateTrie {
 func makeTestStateTrie() (*Database, *StateTrie, map[string][]byte) {
 	// Create an empty trie
 	triedb := NewDatabase(memorydb.New())
-	trie, _ := NewStateTrie(common.Hash{}, common.Hash{}, triedb)
+	trie, _ := NewStateTrie(TrieID(common.Hash{}), triedb)
 
 	// Fill it with some arbitrary data
 	content := make(map[string][]byte)
@@ -66,7 +66,7 @@ func makeTestStateTrie() (*Database, *StateTrie, map[string][]byte) {
 		panic(fmt.Errorf("failed to commit db %v", err))
 	}
 	// Re-create the trie based on the new state
-	trie, _ = NewSecure(common.Hash{}, root, triedb)
+	trie, _ = NewStateTrie(TrieID(root), triedb)
 	return triedb, trie, content
 }
 

trie/sync_test.go

@@ -30,7 +30,7 @@ import (
 func makeTestTrie() (*Database, *StateTrie, map[string][]byte) {
 	// Create an empty trie
 	triedb := NewDatabase(memorydb.New())
-	trie, _ := NewStateTrie(common.Hash{}, common.Hash{}, triedb)
+	trie, _ := NewStateTrie(TrieID(common.Hash{}), triedb)
 
 	// Fill it with some arbitrary data
 	content := make(map[string][]byte)
@@ -59,7 +59,7 @@ func makeTestTrie() (*Database, *StateTrie, map[string][]byte) {
 		panic(fmt.Errorf("failed to commit db %v", err))
 	}
 	// Re-create the trie based on the new state
-	trie, _ = NewSecure(common.Hash{}, root, triedb)
+	trie, _ = NewStateTrie(TrieID(root), triedb)
 	return triedb, trie, content
 }
 
@@ -67,7 +67,7 @@ func makeTestTrie() (*Database, *StateTrie, map[string][]byte) {
 // content map.
 func checkTrieContents(t *testing.T, db *Database, root []byte, content map[string][]byte) {
 	// Check root availability and trie contents
-	trie, err := NewStateTrie(common.Hash{}, common.BytesToHash(root), db)
+	trie, err := NewStateTrie(TrieID(common.BytesToHash(root)), db)
 	if err != nil {
 		t.Fatalf("failed to create trie at %x: %v", root, err)
 	}
@@ -84,7 +84,7 @@ func checkTrieContents(t *testing.T, db *Database, root []byte, content map[stri
 // checkTrieConsistency checks that all nodes in a trie are indeed present.
 func checkTrieConsistency(db *Database, root common.Hash) error {
 	// Create and iterate a trie rooted in a subnode
-	trie, err := NewStateTrie(common.Hash{}, root, db)
+	trie, err := NewStateTrie(TrieID(root), db)
 	if err != nil {
 		return nil // Consider a non existent state consistent
 	}
@@ -105,8 +105,8 @@ type trieElement struct {
 func TestEmptySync(t *testing.T) {
 	dbA := NewDatabase(memorydb.New())
 	dbB := NewDatabase(memorydb.New())
-	emptyA := NewEmpty(dbA)
-	emptyB, _ := New(common.Hash{}, emptyRoot, dbB)
+	emptyA, _ := New(TrieID(common.Hash{}), dbA)
+	emptyB, _ := New(TrieID(emptyRoot), dbB)
 
 	for i, trie := range []*Trie{emptyA, emptyB} {
 		sync := NewSync(trie.Hash(), memorydb.New(), nil)

trie/trie.go

@@ -67,9 +67,8 @@ type Trie struct {
 	// actually unhashed nodes.
 	unhashed int
 
-	// db is the handler trie can retrieve nodes from. It's
-	// only for reading purpose and not available for writing.
-	db *Database
+	// reader is the handler trie can retrieve nodes from.
+	reader *trieReader
 
 	// tracer is the tool to track the trie changes.
 	// It will be reset after each commit operation.
@@ -87,26 +86,29 @@ func (t *Trie) Copy() *Trie {
 		root:     t.root,
 		owner:    t.owner,
 		unhashed: t.unhashed,
-		db:       t.db,
+		reader:   t.reader,
 		tracer:   t.tracer.copy(),
 	}
 }
 
-// New creates a trie with an existing root node from db and an assigned
-// owner for storage proximity.
-//
-// If root is the zero hash or the sha3 hash of an empty string, the
-// trie is initially empty and does not require a database. Otherwise,
-// New will panic if db is nil and returns a MissingNodeError if root does
-// not exist in the database. Accessing the trie loads nodes from db on demand.
-func New(owner common.Hash, root common.Hash, db *Database) (*Trie, error) {
+// New creates the trie instance with provided trie id and the read-only
+// database. The state specified by trie id must be available, otherwise
+// an error will be returned. The trie root specified by trie id can be
+// zero hash or the sha3 hash of an empty string, then trie is initially
+// empty, otherwise, the root node must be present in database or returns
+// a MissingNodeError if not.
+func New(id *ID, db NodeReader) (*Trie, error) {
+	reader, err := newTrieReader(id.StateRoot, id.Owner, db)
+	if err != nil {
+		return nil, err
+	}
 	trie := &Trie{
-		owner: owner,
-		db:    db,
+		owner:  id.Owner,
+		reader: reader,
 		//tracer: newTracer(),
 	}
-	if root != (common.Hash{}) && root != emptyRoot {
-		rootnode, err := trie.resolveHash(root[:], nil)
+	if id.Root != (common.Hash{}) && id.Root != emptyRoot {
+		rootnode, err := trie.resolveAndTrack(id.Root[:], nil)
 		if err != nil {
 			return nil, err
 		}
@@ -117,7 +119,7 @@ func New(owner common.Hash, root common.Hash, db *Database) (*Trie, error) {
 
 // NewEmpty is a shortcut to create empty tree. It's mostly used in tests.
 func NewEmpty(db *Database) *Trie {
-	tr, _ := New(common.Hash{}, common.Hash{}, db)
+	tr, _ := New(TrieID(common.Hash{}), db)
 	return tr
 }
 
@@ -173,7 +175,7 @@ func (t *Trie) tryGet(origNode node, key []byte, pos int) (value []byte, newnode
 		}
 		return value, n, didResolve, err
 	case hashNode:
-		child, err := t.resolveHash(n, key[:pos])
+		child, err := t.resolveAndTrack(n, key[:pos])
 		if err != nil {
 			return nil, n, true, err
 		}
@@ -219,7 +221,7 @@ func (t *Trie) tryGetNode(origNode node, path []byte, pos int) (item []byte, new
 		if hash == nil {
 			return nil, origNode, 0, errors.New("non-consensus node")
 		}
-		blob, err := t.db.Node(common.BytesToHash(hash))
+		blob, err := t.reader.nodeBlob(path, common.BytesToHash(hash))
 		return blob, origNode, 1, err
 	}
 	// Path still needs to be traversed, descend into children
@@ -249,7 +251,7 @@ func (t *Trie) tryGetNode(origNode node, path []byte, pos int) (item []byte, new
 		return item, n, resolved, err
 
 	case hashNode:
-		child, err := t.resolveHash(n, path[:pos])
+		child, err := t.resolveAndTrack(n, path[:pos])
 		if err != nil {
 			return nil, n, 1, err
 		}
@@ -370,7 +372,7 @@ func (t *Trie) insert(n node, prefix, key []byte, value node) (bool, node, error
 		// We've hit a part of the trie that isn't loaded yet. Load
 		// the node and insert into it. This leaves all child nodes on
 		// the path to the value in the trie.
-		rn, err := t.resolveHash(n, prefix)
+		rn, err := t.resolveAndTrack(n, prefix)
 		if err != nil {
 			return false, nil, err
 		}
@@ -524,7 +526,7 @@ func (t *Trie) delete(n node, prefix, key []byte) (bool, node, error) {
 		// We've hit a part of the trie that isn't loaded yet. Load
 		// the node and delete from it. This leaves all child nodes on
 		// the path to the value in the trie.
-		rn, err := t.resolveHash(n, prefix)
+		rn, err := t.resolveAndTrack(n, prefix)
 		if err != nil {
 			return false, nil, err
 		}
@@ -548,30 +550,22 @@ func concat(s1 []byte, s2 ...byte) []byte {
 
 func (t *Trie) resolve(n node, prefix []byte) (node, error) {
 	if n, ok := n.(hashNode); ok {
-		return t.resolveHash(n, prefix)
+		return t.resolveAndTrack(n, prefix)
 	}
 	return n, nil
 }
 
-// resolveHash loads node from the underlying database with the provided
-// node hash and path prefix.
-func (t *Trie) resolveHash(n hashNode, prefix []byte) (node, error) {
-	hash := common.BytesToHash(n)
-	if node := t.db.node(hash); node != nil {
-		return node, nil
+// resolveAndTrack loads node from the underlying store with the given node hash
+// and path prefix and also tracks the loaded node blob in tracer treated as the
+// node's original value. The rlp-encoded blob is preferred to be loaded from
+// database because it's easy to decode node while complex to encode node to blob.
+func (t *Trie) resolveAndTrack(n hashNode, prefix []byte) (node, error) {
+	blob, err := t.reader.nodeBlob(prefix, common.BytesToHash(n))
+	if err != nil {
+		return nil, err
 	}
-	return nil, &MissingNodeError{Owner: t.owner, NodeHash: hash, Path: prefix}
-}
-
-// resolveHash loads rlp-encoded node blob from the underlying database
-// with the provided node hash and path prefix.
-func (t *Trie) resolveBlob(n hashNode, prefix []byte) ([]byte, error) {
-	hash := common.BytesToHash(n)
-	blob, _ := t.db.Node(hash)
-	if len(blob) != 0 {
-		return blob, nil
-	}
-	return nil, &MissingNodeError{Owner: t.owner, NodeHash: hash, Path: prefix}
+	t.tracer.onRead(prefix, blob)
+	return mustDecodeNode(n, blob), nil
 }
 
 // Hash returns the root hash of the trie. It does not write to the
@@ -606,7 +600,7 @@ func (t *Trie) Commit(collectLeaf bool) (common.Hash, *NodeSet, error) {
 		t.root = hashedNode
 		return rootHash, nil, nil
 	}
-	h := newCommitter(t.owner, collectLeaf)
+	h := newCommitter(t.owner, t.tracer, collectLeaf)
 	newRoot, nodes, err := h.Commit(t.root)
 	if err != nil {
 		return common.Hash{}, nil, err
@@ -633,6 +627,5 @@ func (t *Trie) Reset() {
 	t.root = nil
 	t.owner = common.Hash{}
 	t.unhashed = 0
-	//t.db = nil
 	t.tracer.reset()
 }

trie/trie_id.go

@@ -0,0 +1,55 @@
+// Copyright 2022 The go-ethereum Authors
+// This file is part of the go-ethereum library.
+//
+// The go-ethereum library is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Lesser General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// The go-ethereum library is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Lesser General Public License for more details.
+//
+// You should have received a copy of the GNU Lesser General Public License
+// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>
+
+package trie
+
+import "github.com/ethereum/go-ethereum/common"
+
+// ID is the identifier for uniquely identifying a trie.
+type ID struct {
+	StateRoot common.Hash // The root of the corresponding state(block.root)
+	Owner     common.Hash // The contract address hash which the trie belongs to
+	Root      common.Hash // The root hash of trie
+}
+
+// StateTrieID constructs an identifier for state trie with the provided state root.
+func StateTrieID(root common.Hash) *ID {
+	return &ID{
+		StateRoot: root,
+		Owner:     common.Hash{},
+		Root:      root,
+	}
+}
+
+// StorageTrieID constructs an identifier for storage trie which belongs to a certain
+// state and contract specified by the stateRoot and owner.
+func StorageTrieID(stateRoot common.Hash, owner common.Hash, root common.Hash) *ID {
+	return &ID{
+		StateRoot: stateRoot,
+		Owner:     owner,
+		Root:      root,
+	}
+}
+
+// TrieID constructs an identifier for a standard trie(not a second-layer trie)
+// with provided root. It's mostly used in tests and some other tries like CHT trie.
+func TrieID(root common.Hash) *ID {
+	return &ID{
+		StateRoot: root,
+		Owner:     common.Hash{},
+		Root:      root,
+	}
+}

trie/trie_reader.go

@@ -0,0 +1,106 @@
+// Copyright 2022 The go-ethereum Authors
+// This file is part of the go-ethereum library.
+//
+// The go-ethereum library is free software: you can redistribute it and/or modify
+// it under the terms of the GNU Lesser General Public License as published by
+// the Free Software Foundation, either version 3 of the License, or
+// (at your option) any later version.
+//
+// The go-ethereum library is distributed in the hope that it will be useful,
+// but WITHOUT ANY WARRANTY; without even the implied warranty of
+// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
+// GNU Lesser General Public License for more details.
+//
+// You should have received a copy of the GNU Lesser General Public License
+// along with the go-ethereum library. If not, see <http://www.gnu.org/licenses/>.
+
+package trie
+
+import (
+	"fmt"
+
+	"github.com/ethereum/go-ethereum/common"
+)
+
+// Reader wraps the Node and NodeBlob method of a backing trie store.
+type Reader interface {
+	// Node retrieves the trie node with the provided trie identifier, hexary
+	// node path and the corresponding node hash.
+	// No error will be returned if the node is not found.
+	Node(owner common.Hash, path []byte, hash common.Hash) (node, error)
+
+	// NodeBlob retrieves the RLP-encoded trie node blob with the provided trie
+	// identifier, hexary node path and the corresponding node hash.
+	// No error will be returned if the node is not found.
+	NodeBlob(owner common.Hash, path []byte, hash common.Hash) ([]byte, error)
+}
+
+// NodeReader wraps all the necessary functions for accessing trie node.
+type NodeReader interface {
+	// GetReader returns a reader for accessing all trie nodes with provided
+	// state root. Nil is returned in case the state is not available.
+	GetReader(root common.Hash) Reader
+}
+
+// trieReader is a wrapper of the underlying node reader. It's not safe
+// for concurrent usage.
+type trieReader struct {
+	owner  common.Hash
+	reader Reader
+	banned map[string]struct{} // Marker to prevent node from being accessed, for tests
+}
+
+// newTrieReader initializes the trie reader with the given node reader.
+func newTrieReader(stateRoot, owner common.Hash, db NodeReader) (*trieReader, error) {
+	reader := db.GetReader(stateRoot)
+	if reader == nil {
+		return nil, fmt.Errorf("state not found #%x", stateRoot)
+	}
+	return &trieReader{owner: owner, reader: reader}, nil
+}
+
+// newEmptyReader initializes the pure in-memory reader. All read operations
+// should be forbidden and returns the MissingNodeError.
+func newEmptyReader() *trieReader {
+	return &trieReader{}
+}
+
+// node retrieves the trie node with the provided trie node information.
+// An MissingNodeError will be returned in case the node is not found or
+// any error is encountered.
+func (r *trieReader) node(path []byte, hash common.Hash) (node, error) {
+	// Perform the logics in tests for preventing trie node access.
+	if r.banned != nil {
+		if _, ok := r.banned[string(path)]; ok {
+			return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path}
+		}
+	}
+	if r.reader == nil {
+		return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path}
+	}
+	node, err := r.reader.Node(r.owner, path, hash)
+	if err != nil || node == nil {
+		return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path, err: err}
+	}
+	return node, nil
+}
+
+// node retrieves the rlp-encoded trie node with the provided trie node
+// information. An MissingNodeError will be returned in case the node is
+// not found or any error is encountered.
+func (r *trieReader) nodeBlob(path []byte, hash common.Hash) ([]byte, error) {
+	// Perform the logics in tests for preventing trie node access.
+	if r.banned != nil {
+		if _, ok := r.banned[string(path)]; ok {
+			return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path}
+		}
+	}
+	if r.reader == nil {
+		return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path}
+	}
+	blob, err := r.reader.NodeBlob(r.owner, path, hash)
+	if err != nil || len(blob) == 0 {
+		return nil, &MissingNodeError{Owner: r.owner, NodeHash: hash, Path: path, err: err}
+	}
+	return blob, nil
+}

trie/trie_test.go

@@ -64,7 +64,8 @@ func TestNull(t *testing.T) {
 }
 
 func TestMissingRoot(t *testing.T) {
-	trie, err := New(common.Hash{}, common.HexToHash("0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33"), NewDatabase(memorydb.New()))
+	root := common.HexToHash("0beec7b5ea3f0fdbc95d0dd47f3c5bc275da8a33")
+	trie, err := New(TrieID(root), NewDatabase(memorydb.New()))
 	if trie != nil {
 		t.Error("New returned non-nil trie for invalid root")
 	}
@@ -89,27 +90,27 @@ func testMissingNode(t *testing.T, memonly bool) {
 		triedb.Commit(root, true, nil)
 	}
 
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	_, err := trie.TryGet([]byte("120000"))
 	if err != nil {
 		t.Errorf("Unexpected error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	_, err = trie.TryGet([]byte("120099"))
 	if err != nil {
 		t.Errorf("Unexpected error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	_, err = trie.TryGet([]byte("123456"))
 	if err != nil {
 		t.Errorf("Unexpected error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	err = trie.TryUpdate([]byte("120099"), []byte("zxcvzxcvzxcvzxcvzxcvzxcvzxcvzxcv"))
 	if err != nil {
 		t.Errorf("Unexpected error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	err = trie.TryDelete([]byte("123456"))
 	if err != nil {
 		t.Errorf("Unexpected error: %v", err)
@@ -122,27 +123,27 @@ func testMissingNode(t *testing.T, memonly bool) {
 		diskdb.Delete(hash[:])
 	}
 
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	_, err = trie.TryGet([]byte("120000"))
 	if _, ok := err.(*MissingNodeError); !ok {
 		t.Errorf("Wrong error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	_, err = trie.TryGet([]byte("120099"))
 	if _, ok := err.(*MissingNodeError); !ok {
 		t.Errorf("Wrong error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	_, err = trie.TryGet([]byte("123456"))
 	if err != nil {
 		t.Errorf("Unexpected error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	err = trie.TryUpdate([]byte("120099"), []byte("zxcv"))
 	if _, ok := err.(*MissingNodeError); !ok {
 		t.Errorf("Wrong error: %v", err)
 	}
-	trie, _ = New(common.Hash{}, root, triedb)
+	trie, _ = New(TrieID(root), triedb)
 	err = trie.TryDelete([]byte("123456"))
 	if _, ok := err.(*MissingNodeError); !ok {
 		t.Errorf("Wrong error: %v", err)
@@ -196,7 +197,7 @@ func TestGet(t *testing.T) {
 		}
 		root, nodes, _ := trie.Commit(false)
 		db.Update(NewWithNodeSet(nodes))
-		trie, _ = New(common.Hash{}, root, db)
+		trie, _ = New(TrieID(root), db)
 	}
 }
 
@@ -273,7 +274,7 @@ func TestReplication(t *testing.T) {
 	triedb.Update(NewWithNodeSet(nodes))
 
 	// create a new trie on top of the database and check that lookups work.
-	trie2, err := New(common.Hash{}, exp, triedb)
+	trie2, err := New(TrieID(exp), triedb)
 	if err != nil {
 		t.Fatalf("can't recreate trie at %x: %v", exp, err)
 	}
@@ -294,7 +295,7 @@ func TestReplication(t *testing.T) {
 	if nodes != nil {
 		triedb.Update(NewWithNodeSet(nodes))
 	}
-	trie2, err = New(common.Hash{}, hash, triedb)
+	trie2, err = New(TrieID(hash), triedb)
 	if err != nil {
 		t.Fatalf("can't recreate trie at %x: %v", exp, err)
 	}
@@ -377,6 +378,7 @@ const (
 	opCommit
 	opItercheckhash
 	opNodeDiff
+	opProve
 	opMax // boundary value, not an actual op
 )
 
@@ -402,7 +404,7 @@ func (randTest) Generate(r *rand.Rand, size int) reflect.Value {
 			step.key = genKey()
 			step.value = make([]byte, 8)
 			binary.BigEndian.PutUint64(step.value, uint64(i))
-		case opGet, opDelete:
+		case opGet, opDelete, opProve:
 			step.key = genKey()
 		}
 		steps = append(steps, step)
@@ -436,24 +438,60 @@ func runRandTest(rt randTest) bool {
 			if string(v) != want {
 				rt[i].err = fmt.Errorf("mismatch for key %#x, got %#x want %#x", step.key, v, want)
 			}
+		case opProve:
+			hash := tr.Hash()
+			if hash == emptyRoot {
+				continue
+			}
+			proofDb := rawdb.NewMemoryDatabase()
+			err := tr.Prove(step.key, 0, proofDb)
+			if err != nil {
+				rt[i].err = fmt.Errorf("failed for proving key %#x, %v", step.key, err)
+			}
+			_, err = VerifyProof(hash, step.key, proofDb)
+			if err != nil {
+				rt[i].err = fmt.Errorf("failed for verifying key %#x, %v", step.key, err)
+			}
 		case opHash:
 			tr.Hash()
 		case opCommit:
-			hash, nodes, err := tr.Commit(false)
+			root, nodes, err := tr.Commit(true)
 			if err != nil {
 				rt[i].err = err
 				return false
 			}
+			// Validity the returned nodeset
+			if nodes != nil {
+				for path, node := range nodes.updates.nodes {
+					blob, _, _ := origTrie.TryGetNode(hexToCompact([]byte(path)))
+					got := node.prev
+					if !bytes.Equal(blob, got) {
+						rt[i].err = fmt.Errorf("prevalue mismatch for 0x%x, got 0x%x want 0x%x", path, got, blob)
+						panic(rt[i].err)
+					}
+				}
+				for path, prev := range nodes.deletes {
+					blob, _, _ := origTrie.TryGetNode(hexToCompact([]byte(path)))
+					if !bytes.Equal(blob, prev) {
+						rt[i].err = fmt.Errorf("prevalue mismatch for 0x%x, got 0x%x want 0x%x", path, prev, blob)
+						return false
+					}
+				}
+			}
 			if nodes != nil {
 				triedb.Update(NewWithNodeSet(nodes))
 			}
-			newtr, err := New(common.Hash{}, hash, triedb)
+			newtr, err := New(TrieID(root), triedb)
 			if err != nil {
 				rt[i].err = err
 				return false
 			}
 			tr = newtr
+
+			// Enable node tracing. Resolve the root node again explicitly
+			// since it's not captured at the beginning.
 			tr.tracer = newTracer()
+			tr.resolveAndTrack(root.Bytes(), nil)
 
 			origTrie = tr.Copy()
 		case opItercheckhash:

trie/util_test.go

@@ -17,6 +17,7 @@
 package trie
 
 import (
+	"bytes"
 	"testing"
 
 	"github.com/ethereum/go-ethereum/common"
@@ -72,7 +73,7 @@ func TestTrieTracer(t *testing.T) {
 	if err := db.Update(NewWithNodeSet(nodes)); err != nil {
 		t.Fatal(err)
 	}
-	trie, _ = New(common.Hash{}, root, db)
+	trie, _ = New(TrieID(root), db)
 	trie.tracer = newTracer()
 
 	// Delete all the elements, check deletion set
@@ -124,3 +125,120 @@ func TestTrieTracerNoop(t *testing.T) {
 		t.Fatalf("Unexpected deleted node tracked %d", len(trie.tracer.deleteList()))
 	}
 }
+
+func TestTrieTracePrevValue(t *testing.T) {
+	db := NewDatabase(rawdb.NewMemoryDatabase())
+	trie := NewEmpty(db)
+	trie.tracer = newTracer()
+
+	paths, blobs := trie.tracer.prevList()
+	if len(paths) != 0 || len(blobs) != 0 {
+		t.Fatalf("Nothing should be tracked")
+	}
+	// Insert a batch of entries, all the nodes should be marked as inserted
+	vals := []struct{ k, v string }{
+		{"do", "verb"},
+		{"ether", "wookiedoo"},
+		{"horse", "stallion"},
+		{"shaman", "horse"},
+		{"doge", "coin"},
+		{"dog", "puppy"},
+		{"somethingveryoddindeedthis is", "myothernodedata"},
+	}
+	for _, val := range vals {
+		trie.Update([]byte(val.k), []byte(val.v))
+	}
+	paths, blobs = trie.tracer.prevList()
+	if len(paths) != 0 || len(blobs) != 0 {
+		t.Fatalf("Nothing should be tracked")
+	}
+
+	// Commit the changes and re-create with new root
+	root, nodes, _ := trie.Commit(false)
+	if err := db.Update(NewWithNodeSet(nodes)); err != nil {
+		t.Fatal(err)
+	}
+	trie, _ = New(TrieID(root), db)
+	trie.tracer = newTracer()
+	trie.resolveAndTrack(root.Bytes(), nil)
+
+	// Load all nodes in trie
+	for _, val := range vals {
+		trie.TryGet([]byte(val.k))
+	}
+
+	// Ensure all nodes are tracked by tracer with correct prev-values
+	iter := trie.NodeIterator(nil)
+	seen := make(map[string][]byte)
+	for iter.Next(true) {
+		// Embedded nodes are ignored since they are not present in
+		// database.
+		if iter.Hash() == (common.Hash{}) {
+			continue
+		}
+		seen[string(iter.Path())] = common.CopyBytes(iter.NodeBlob())
+	}
+
+	paths, blobs = trie.tracer.prevList()
+	if len(paths) != len(seen) || len(blobs) != len(seen) {
+		t.Fatalf("Unexpected tracked values")
+	}
+	for i, path := range paths {
+		blob := blobs[i]
+		prev, ok := seen[string(path)]
+		if !ok {
+			t.Fatalf("Missing node %v", path)
+		}
+		if !bytes.Equal(blob, prev) {
+			t.Fatalf("Unexpected value path: %v, want: %v, got: %v", path, prev, blob)
+		}
+	}
+
+	// Re-open the trie and iterate the trie, ensure nothing will be tracked.
+	// Iterator will not link any loaded nodes to trie.
+	trie, _ = New(TrieID(root), db)
+	trie.tracer = newTracer()
+
+	iter = trie.NodeIterator(nil)
+	for iter.Next(true) {
+	}
+	paths, blobs = trie.tracer.prevList()
+	if len(paths) != 0 || len(blobs) != 0 {
+		t.Fatalf("Nothing should be tracked")
+	}
+
+	// Re-open the trie and generate proof for entries, ensure nothing will
+	// be tracked. Prover will not link any loaded nodes to trie.
+	trie, _ = New(TrieID(root), db)
+	trie.tracer = newTracer()
+	for _, val := range vals {
+		trie.Prove([]byte(val.k), 0, rawdb.NewMemoryDatabase())
+	}
+	paths, blobs = trie.tracer.prevList()
+	if len(paths) != 0 || len(blobs) != 0 {
+		t.Fatalf("Nothing should be tracked")
+	}
+
+	// Delete entries from trie, ensure all previous values are correct.
+	trie, _ = New(TrieID(root), db)
+	trie.tracer = newTracer()
+	trie.resolveAndTrack(root.Bytes(), nil)
+
+	for _, val := range vals {
+		trie.TryDelete([]byte(val.k))
+	}
+	paths, blobs = trie.tracer.prevList()
+	if len(paths) != len(seen) || len(blobs) != len(seen) {
+		t.Fatalf("Unexpected tracked values")
+	}
+	for i, path := range paths {
+		blob := blobs[i]
+		prev, ok := seen[string(path)]
+		if !ok {
+			t.Fatalf("Missing node %v", path)
+		}
+		if !bytes.Equal(blob, prev) {
+			t.Fatalf("Unexpected value path: %v, want: %v, got: %v", path, prev, blob)
+		}
+	}
+}

trie/utils.go

@@ -50,45 +50,43 @@ func newTracer() *tracer {
 	}
 }
 
-/*
 // onRead tracks the newly loaded trie node and caches the rlp-encoded blob internally.
 // Don't change the value outside of function since it's not deep-copied.
-func (t *tracer) onRead(key []byte, val []byte) {
+func (t *tracer) onRead(path []byte, val []byte) {
 	// Tracer isn't used right now, remove this check later.
 	if t == nil {
 		return
 	}
-	t.origin[string(key)] = val
+	t.origin[string(path)] = val
 }
-*/
 
 // onInsert tracks the newly inserted trie node. If it's already in the deletion set
 // (resurrected node), then just wipe it from the deletion set as the "untouched".
-func (t *tracer) onInsert(key []byte) {
+func (t *tracer) onInsert(path []byte) {
 	// Tracer isn't used right now, remove this check later.
 	if t == nil {
 		return
 	}
-	if _, present := t.delete[string(key)]; present {
-		delete(t.delete, string(key))
+	if _, present := t.delete[string(path)]; present {
+		delete(t.delete, string(path))
 		return
 	}
-	t.insert[string(key)] = struct{}{}
+	t.insert[string(path)] = struct{}{}
 }
 
 // onDelete tracks the newly deleted trie node. If it's already
 // in the addition set, then just wipe it from the addition set
 // as it's untouched.
-func (t *tracer) onDelete(key []byte) {
+func (t *tracer) onDelete(path []byte) {
 	// Tracer isn't used right now, remove this check later.
 	if t == nil {
 		return
 	}
-	if _, present := t.insert[string(key)]; present {
-		delete(t.insert, string(key))
+	if _, present := t.insert[string(path)]; present {
+		delete(t.insert, string(path))
 		return
 	}
-	t.delete[string(key)] = struct{}{}
+	t.delete[string(path)] = struct{}{}
 }
 
 // insertList returns the tracked inserted trie nodes in list format.
@@ -98,8 +96,8 @@ func (t *tracer) insertList() [][]byte {
 		return nil
 	}
 	var ret [][]byte
-	for key := range t.insert {
-		ret = append(ret, []byte(key))
+	for path := range t.insert {
+		ret = append(ret, []byte(path))
 	}
 	return ret
 }
@@ -111,22 +109,37 @@ func (t *tracer) deleteList() [][]byte {
 		return nil
 	}
 	var ret [][]byte
-	for key := range t.delete {
-		ret = append(ret, []byte(key))
+	for path := range t.delete {
+		ret = append(ret, []byte(path))
 	}
 	return ret
 }
 
-/*
+// prevList returns the tracked node blobs in list format.
+func (t *tracer) prevList() ([][]byte, [][]byte) {
+	// Tracer isn't used right now, remove this check later.
+	if t == nil {
+		return nil, nil
+	}
+	var (
+		paths [][]byte
+		blobs [][]byte
+	)
+	for path, blob := range t.origin {
+		paths = append(paths, []byte(path))
+		blobs = append(blobs, blob)
+	}
+	return paths, blobs
+}
+
 // getPrev returns the cached original value of the specified node.
-func (t *tracer) getPrev(key []byte) []byte {
-	// Don't panic on uninitialized tracer, it's possible in testing.
+func (t *tracer) getPrev(path []byte) []byte {
+	// Tracer isn't used right now, remove this check later.
 	if t == nil {
 		return nil
 	}
-	return t.origin[string(key)]
+	return t.origin[string(path)]
 }
-*/
 
 // reset clears the content tracked by tracer.
 func (t *tracer) reset() {