gob: use bufio on the decode to avoid a system call on each read.

Add a benchmark.
BenchmarkEndToEndPipe gives 14.3microseconds/op before,
13.1microseconds/op after, or about 76e3 round trips per second
through the kernel.
With a bytes buffer, and therefore no system calls for I/O, the
numbers go to 7.3microseconds/op, or about 137e3 round trips
per second.

R=rsc
CC=golang-dev
https://golang.org/cl/4279045

src/pkg/gob/decoder.go

@@ -5,6 +5,7 @@
 package gob
 
 import (
+	"bufio"
 	"bytes"
 	"io"
 	"os"
@@ -30,7 +31,7 @@ type Decoder struct {
 // NewDecoder returns a new decoder that reads from the io.Reader.
 func NewDecoder(r io.Reader) *Decoder {
 	dec := new(Decoder)
-	dec.r = r
+	dec.r = bufio.NewReader(r)
 	dec.wireType = make(map[typeId]*wireType)
 	dec.decoderCache = make(map[reflect.Type]map[typeId]**decEngine)
 	dec.ignorerCache = make(map[typeId]**decEngine)

src/pkg/gob/encoder_test.go

@@ -514,3 +514,39 @@ func TestNestedInterfaces(t *testing.T) {
 		t.Fatalf("final value %d; expected %d", inner.A, 7)
 	}
 }
+
+type Bench struct {
+	A int
+	B float64
+	C string
+	D []byte
+}
+
+func benchmarkEndToEnd(r io.Reader, w io.Writer, b *testing.B) {
+	b.StopTimer()
+	enc := NewEncoder(w)
+	dec := NewDecoder(r)
+	bench := &Bench{7, 3.2, "now is the time", []byte("for all good men")}
+	b.StartTimer()
+	for i := 0; i < b.N; i++ {
+		if enc.Encode(bench) != nil {
+			panic("encode error")
+		}
+		if dec.Decode(bench) != nil {
+			panic("decode error")
+		}
+	}
+}
+
+func BenchmarkEndToEndPipe(b *testing.B) {
+	r, w, err := os.Pipe()
+	if err != nil {
+		panic("can't get pipe:" + err.String())
+	}
+	benchmarkEndToEnd(r, w, b)
+}
+
+func BenchmarkEndToEndByteBuffer(b *testing.B) {
+	var buf bytes.Buffer
+	benchmarkEndToEnd(&buf, &buf, b)
+}