bytes: don't compact Buffer so aggressively

benchmark                           old ns/op    new ns/op    delta
BenchmarkBufferNotEmptyWriteRead       848416       819983   -3.35%

Update #5154

R=golang-dev, gri, robryk
CC=golang-dev
https://golang.org/cl/8173043

src/pkg/bytes/buffer.go

@@ -87,9 +87,11 @@ func (b *Buffer) grow(n int) int {
 		var buf []byte
 		if b.buf == nil && n <= len(b.bootstrap) {
 			buf = b.bootstrap[0:]
-		} else if m+n <= cap(b.buf) {
-			// We can slide things down instead of
-			// allocating a new slice.
+		} else if m+n <= cap(b.buf)/2 {
+			// We can slide things down instead of allocating a new
+			// slice. We only need m+n <= cap(b.buf) to slide, but
+			// we instead let capacity get twice as large so we
+			// don't spend all our time copying.
 			copy(b.buf[:], b.buf[b.off:])
 			buf = b.buf[:m]
 		} else {

src/pkg/bytes/buffer_test.go

@@ -490,8 +490,10 @@ func TestBufferGrowth(t *testing.T) {
 		}
 	}
 	cap1 := b.Cap()
-	if cap1 > cap0 {
-		t.Errorf("buffer cap = %d; too big", cap1)
+	// (*Buffer).grow allows for 2x capacity slop before sliding,
+	// so set our error threshold at 3x.
+	if cap1 > cap0*3 {
+		t.Errorf("buffer cap = %d; too big (grew from %d)", cap1, cap0)
 	}
 }