trie: optimize node encoding using in-place RLP encoder.

go.mod

@@ -53,6 +53,7 @@ require (
 	github.com/olekukonko/tablewriter v0.0.5
 	github.com/peterh/liner v1.1.1-0.20190123174540-a2c9a5303de7
 	github.com/protolambda/bls12-381-util v0.0.0-20220416220906-d8552aa452c7
+	github.com/qianbin/drlp v0.0.0-20231230065804-db474f66bf91
 	github.com/rs/cors v1.7.0
 	github.com/shirou/gopsutil v3.21.4-0.20210419000835-c7a38de76ee5+incompatible
 	github.com/status-im/keycard-go v0.2.0

go.sum

@@ -522,6 +522,8 @@ github.com/prometheus/procfs v0.7.3 h1:4jVXhlkAyzOScmCkXBTOLRLTz8EeU+eyjrwB/EPq0
 github.com/prometheus/procfs v0.7.3/go.mod h1:cz+aTbrPOrUb4q7XlbU9ygM+/jj0fzG6c1xBZuNvfVA=
 github.com/protolambda/bls12-381-util v0.0.0-20220416220906-d8552aa452c7 h1:cZC+usqsYgHtlBaGulVnZ1hfKAi8iWtujBnRLQE698c=
 github.com/protolambda/bls12-381-util v0.0.0-20220416220906-d8552aa452c7/go.mod h1:IToEjHuttnUzwZI5KBSM/LOOW3qLbbrHOEfp3SbECGY=
+github.com/qianbin/drlp v0.0.0-20231230065804-db474f66bf91 h1:N1pcRULtPB7CndtexGDr94ai8wsJLx55AKYfo7F6qEs=
+github.com/qianbin/drlp v0.0.0-20231230065804-db474f66bf91/go.mod h1:OnClEjurpFUtR3RUCauP9HxNNl8xjfGAOv0kWYTznOc=
 github.com/rivo/uniseg v0.2.0 h1:S1pD9weZBuJdFmowNwbpi7BJ8TNftyUImj/0WQi72jY=
 github.com/rivo/uniseg v0.2.0/go.mod h1:J6wj4VEh+S6ZtnVlnTBMWIodfgj8LQOQFoIToxlJtxc=
 github.com/rogpeppe/go-internal v1.3.0/go.mod h1:M8bDsm7K2OlrFYOpmOWEs/qY81heoFRclV5y23lUDJ4=

trie/committer.go

@@ -139,7 +139,7 @@ func (c *committer) store(path []byte, n node) node {
 	}
 	// Collect the dirty node to nodeset for return.
 	nhash := common.BytesToHash(hash)
-	c.nodes.AddNode(path, trienode.New(nhash, nodeToBytes(n)))
+	c.nodes.AddNode(path, trienode.New(nhash, n.encode(nil)))
 
 	// 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/hasher.go

@@ -20,7 +20,6 @@ import (
 	"sync"
 
 	"github.com/ethereum/go-ethereum/crypto"
-	"github.com/ethereum/go-ethereum/rlp"
 	"golang.org/x/crypto/sha3"
 )
 
@@ -29,7 +28,6 @@ import (
 type hasher struct {
 	sha      crypto.KeccakState
 	tmp      []byte
-	encbuf   rlp.EncoderBuffer
 	parallel bool // Whether to use parallel threads when hashing
 }
 
@@ -37,9 +35,8 @@ type hasher struct {
 var hasherPool = sync.Pool{
 	New: func() interface{} {
 		return &hasher{
-			tmp:    make([]byte, 0, 550), // cap is as large as a full fullNode.
+			tmp: make([]byte, 0, 550), // cap is as large as a full fullNode.
-			sha:    sha3.NewLegacyKeccak256().(crypto.KeccakState),
+			sha: sha3.NewLegacyKeccak256().(crypto.KeccakState),
-			encbuf: rlp.NewEncoderBuffer(nil),
 		}
 	},
 }
@@ -143,41 +140,23 @@ func (h *hasher) hashFullNodeChildren(n *fullNode) (collapsed *fullNode, cached
 // into compact form for RLP encoding.
 // If the rlp data is smaller than 32 bytes, `nil` is returned.
 func (h *hasher) shortnodeToHash(n *shortNode, force bool) node {
-	n.encode(h.encbuf)
+	h.tmp = n.encode(h.tmp[:0])
-	enc := h.encodedBytes()
 
-	if len(enc) < 32 && !force {
+	if len(h.tmp) < 32 && !force {
 		return n // Nodes smaller than 32 bytes are stored inside their parent
 	}
-	return h.hashData(enc)
+	return h.hashData(h.tmp)
 }
 
 // fullnodeToHash is used to create a hashNode from a fullNode, (which
 // may contain nil values)
 func (h *hasher) fullnodeToHash(n *fullNode, force bool) node {
-	n.encode(h.encbuf)
+	h.tmp = n.encode(h.tmp[:0])
-	enc := h.encodedBytes()
 
-	if len(enc) < 32 && !force {
+	if len(h.tmp) < 32 && !force {
 		return n // Nodes smaller than 32 bytes are stored inside their parent
 	}
-	return h.hashData(enc)
+	return h.hashData(h.tmp)
-}
-
-// encodedBytes returns the result of the last encoding operation on h.encbuf.
-// This also resets the encoder buffer.
-//
-// All node encoding must be done like this:
-//
-//	node.encode(h.encbuf)
-//	enc := h.encodedBytes()
-//
-// This convention exists because node.encode can only be inlined/escape-analyzed when
-// called on a concrete receiver type.
-func (h *hasher) encodedBytes() []byte {
-	h.tmp = h.encbuf.AppendToBytes(h.tmp[:0])
-	h.encbuf.Reset(nil)
-	return h.tmp
 }
 
 // hashData hashes the provided data

trie/iterator.go

@@ -242,7 +242,7 @@ func (it *nodeIterator) LeafProof() [][]byte {
 				// Gather nodes that end up as hash nodes (or the root)
 				node, hashed := hasher.proofHash(item.node)
 				if _, ok := hashed.(hashNode); ok || i == 0 {
-					proofs = append(proofs, nodeToBytes(node))
+					proofs = append(proofs, node.encode(nil))
 				}
 			}
 			return proofs

trie/node.go

@@ -29,7 +29,8 @@ var indices = []string{"0", "1", "2", "3", "4", "5", "6", "7", "8", "9", "a", "b
 
 type node interface {
 	cache() (hashNode, bool)
-	encode(w rlp.EncoderBuffer)
+	// encode appends encoded bytes to buf and returns then extended buffer.
+	encode(buf []byte) []byte
 	fstring(string) string
 }
 
@@ -53,9 +54,11 @@ var nilValueNode = valueNode(nil)
 
 // EncodeRLP encodes a full node into the consensus RLP format.
 func (n *fullNode) EncodeRLP(w io.Writer) error {
-	eb := rlp.NewEncoderBuffer(w)
+	h := newHasher(false)
-	n.encode(eb)
+	defer returnHasherToPool(h)
-	return eb.Flush()
+	h.tmp = n.encode(h.tmp[:0])
+	_, err := w.Write(h.tmp)
+	return err
 }
 
 func (n *fullNode) copy() *fullNode   { copy := *n; return &copy }

trie/node_enc.go

@@ -17,48 +17,46 @@
 package trie
 
 import (
-	"github.com/ethereum/go-ethereum/rlp"
+	"github.com/qianbin/drlp"
 )
 
-func nodeToBytes(n node) []byte {
+func (n *fullNode) encode(buf []byte) []byte {
-	w := rlp.NewEncoderBuffer(nil)
+	if buf == nil {
-	n.encode(w)
+		buf = make([]byte, 0, 550)
-	result := w.ToBytes()
+	}
-	w.Flush()
+	offset := len(buf)
-	return result
-}
-
-func (n *fullNode) encode(w rlp.EncoderBuffer) {
-	offset := w.List()
 	for _, c := range n.Children {
 		if c != nil {
-			c.encode(w)
+			buf = c.encode(buf)
 		} else {
-			w.Write(rlp.EmptyString)
+			buf = drlp.AppendUint(buf, 0)
 		}
 	}
-	w.ListEnd(offset)
+	return drlp.EndList(buf, offset)
 }
 
-func (n *shortNode) encode(w rlp.EncoderBuffer) {
+func (n *shortNode) encode(buf []byte) []byte {
-	offset := w.List()
+	if buf == nil {
-	w.WriteBytes(n.Key)
+		buf = make([]byte, 0, len(n.Key)+40)
-	if n.Val != nil {
-		n.Val.encode(w)
-	} else {
-		w.Write(rlp.EmptyString)
 	}
-	w.ListEnd(offset)
+	offset := len(buf)
+	buf = drlp.AppendString(buf, n.Key)
+	if n.Val != nil {
+		buf = n.Val.encode(buf)
+	} else {
+		buf = drlp.AppendUint(buf, 0)
+	}
+	return drlp.EndList(buf, offset)
 }
 
-func (n hashNode) encode(w rlp.EncoderBuffer) {
+func (n hashNode) encode(buf []byte) []byte {
-	w.WriteBytes(n)
+	return drlp.AppendString(buf, n)
 }
 
-func (n valueNode) encode(w rlp.EncoderBuffer) {
+func (n valueNode) encode(buf []byte) []byte {
-	w.WriteBytes(n)
+	return drlp.AppendString(buf, n)
 }
 
-func (n rawNode) encode(w rlp.EncoderBuffer) {
+func (n rawNode) encode(buf []byte) []byte {
-	w.Write(n)
+	return append(buf, n...)
 }

trie/node_test.go

@@ -108,7 +108,7 @@ func BenchmarkEncodeShortNode(b *testing.B) {
 	b.ReportAllocs()
 
 	for i := 0; i < b.N; i++ {
-		nodeToBytes(node)
+		node.encode(nil)
 	}
 }
 
@@ -126,7 +126,7 @@ func BenchmarkEncodeFullNode(b *testing.B) {
 	b.ReportAllocs()
 
 	for i := 0; i < b.N; i++ {
-		nodeToBytes(node)
+		node.encode(nil)
 	}
 }
 
@@ -140,7 +140,7 @@ func BenchmarkDecodeShortNode(b *testing.B) {
 		Key: []byte{0x1, 0x2},
 		Val: hashNode(randBytes(32)),
 	}
-	blob := nodeToBytes(node)
+	blob := node.encode(nil)
 	hash := crypto.Keccak256(blob)
 
 	b.ResetTimer()
@@ -161,7 +161,7 @@ func BenchmarkDecodeShortNodeUnsafe(b *testing.B) {
 		Key: []byte{0x1, 0x2},
 		Val: hashNode(randBytes(32)),
 	}
-	blob := nodeToBytes(node)
+	blob := node.encode(nil)
 	hash := crypto.Keccak256(blob)
 
 	b.ResetTimer()
@@ -182,7 +182,7 @@ func BenchmarkDecodeFullNode(b *testing.B) {
 	for i := 0; i < 16; i++ {
 		node.Children[i] = hashNode(randBytes(32))
 	}
-	blob := nodeToBytes(node)
+	blob := node.encode(nil)
 	hash := crypto.Keccak256(blob)
 
 	b.ResetTimer()
@@ -203,7 +203,7 @@ func BenchmarkDecodeFullNodeUnsafe(b *testing.B) {
 	for i := 0; i < 16; i++ {
 		node.Children[i] = hashNode(randBytes(32))
 	}
-	blob := nodeToBytes(node)
+	blob := node.encode(nil)
 	hash := crypto.Keccak256(blob)
 
 	b.ResetTimer()

trie/proof.go

@@ -90,7 +90,7 @@ func (t *Trie) Prove(key []byte, proofDb ethdb.KeyValueWriter) error {
 		if hash, ok := hn.(hashNode); ok || i == 0 {
 			// If the node's database encoding is a hash (or is the
 			// root node), it becomes a proof element.
-			enc := nodeToBytes(n)
+			enc := n.encode(nil)
 			if !ok {
 				hash = hasher.hashData(enc)
 			}

trie/stacktrie.go

@@ -377,8 +377,8 @@ func (t *StackTrie) hash(st *stNode, path []byte) {
 			st.children[i] = nil
 			stPool.Put(child.reset()) // Release child back to pool.
 		}
-		nodes.encode(t.h.encbuf)
+		t.h.tmp = nodes.encode(t.h.tmp[:0])
-		blob = t.h.encodedBytes()
+		blob = t.h.tmp
 
 	case extNode:
 		// recursively hash and commit child as the first step
@@ -400,8 +400,8 @@ func (t *StackTrie) hash(st *stNode, path []byte) {
 		} else {
 			n.Val = hashNode(st.children[0].val)
 		}
-		n.encode(t.h.encbuf)
+		t.h.tmp = n.encode(t.h.tmp[:0])
-		blob = t.h.encodedBytes()
+		blob = t.h.tmp
 
 		stPool.Put(st.children[0].reset()) // Release child back to pool.
 		st.children[0] = nil
@@ -410,8 +410,8 @@ func (t *StackTrie) hash(st *stNode, path []byte) {
 		st.key = append(st.key, byte(16))
 		n := shortNode{Key: hexToCompactInPlace(st.key), Val: valueNode(st.val)}
 
-		n.encode(t.h.encbuf)
+		t.h.tmp = n.encode(t.h.tmp[:0])
-		blob = t.h.encodedBytes()
+		blob = t.h.tmp
 
 	default:
 		panic("invalid node type")