clist: speedup Next by removing defers (#2511)

This change doubles the speed of the mempool's reaping.
Before:

BenchmarkReap-8   	    5000	    365390 ns/op	  122887 B/op

After:

BenchmarkReap-8   	   10000	    158274 ns/op	  122882 B/op

libs/clist/clist.go

@@ -113,9 +113,9 @@ func (e *CElement) NextWaitChan() <-chan struct{} {
 // Nonblocking, may return nil if at the end.
 func (e *CElement) Next() *CElement {
 	e.mtx.RLock()
-	defer e.mtx.RUnlock()
-
-	return e.next
+	val := e.next
+	e.mtx.RUnlock()
+	return val
 }
 
 // Nonblocking, may return nil if at the end.

mempool/bench_test.go

@@ -0,0 +1,55 @@
+package mempool
+
+import (
+	"encoding/binary"
+	"testing"
+
+	"github.com/tendermint/tendermint/abci/example/kvstore"
+	"github.com/tendermint/tendermint/proxy"
+)
+
+func BenchmarkReap(b *testing.B) {
+	app := kvstore.NewKVStoreApplication()
+	cc := proxy.NewLocalClientCreator(app)
+	mempool := newMempoolWithApp(cc)
+
+	size := 10000
+	for i := 0; i < size; i++ {
+		tx := make([]byte, 8)
+		binary.BigEndian.PutUint64(tx, uint64(i))
+		mempool.CheckTx(tx, nil)
+	}
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		mempool.ReapMaxBytesMaxGas(100000000, 10000000)
+	}
+}
+
+func BenchmarkCacheInsertTime(b *testing.B) {
+	cache := newMapTxCache(b.N)
+	txs := make([][]byte, b.N)
+	for i := 0; i < b.N; i++ {
+		txs[i] = make([]byte, 8)
+		binary.BigEndian.PutUint64(txs[i], uint64(i))
+	}
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		cache.Push(txs[i])
+	}
+}
+
+// This benchmark is probably skewed, since we actually will be removing
+// txs in parallel, which may cause some overhead due to mutex locking.
+func BenchmarkCacheRemoveTime(b *testing.B) {
+	cache := newMapTxCache(b.N)
+	txs := make([][]byte, b.N)
+	for i := 0; i < b.N; i++ {
+		txs[i] = make([]byte, 8)
+		binary.BigEndian.PutUint64(txs[i], uint64(i))
+		cache.Push(txs[i])
+	}
+	b.ResetTimer()
+	for i := 0; i < b.N; i++ {
+		cache.Remove(txs[i])
+	}
+}

mempool/mempool_test.go

@@ -399,35 +399,6 @@ func TestMempoolCloseWAL(t *testing.T) {
 	require.Equal(t, 1, len(m3), "expecting the wal match in")
 }
 
-func BenchmarkCacheInsertTime(b *testing.B) {
-	cache := newMapTxCache(b.N)
-	txs := make([][]byte, b.N)
-	for i := 0; i < b.N; i++ {
-		txs[i] = make([]byte, 8)
-		binary.BigEndian.PutUint64(txs[i], uint64(i))
-	}
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		cache.Push(txs[i])
-	}
-}
-
-// This benchmark is probably skewed, since we actually will be removing
-// txs in parallel, which may cause some overhead due to mutex locking.
-func BenchmarkCacheRemoveTime(b *testing.B) {
-	cache := newMapTxCache(b.N)
-	txs := make([][]byte, b.N)
-	for i := 0; i < b.N; i++ {
-		txs[i] = make([]byte, 8)
-		binary.BigEndian.PutUint64(txs[i], uint64(i))
-		cache.Push(txs[i])
-	}
-	b.ResetTimer()
-	for i := 0; i < b.N; i++ {
-		cache.Remove(txs[i])
-	}
-}
-
 func checksumIt(data []byte) string {
 	h := md5.New()
 	h.Write(data)