encoding/json: various minor decoder speed-ups

Reuse v.Type() and cachedTypeFields(t) when decoding maps and structs.

Always use the same data slices when in hot loops, to ensure that the
compiler generates good code. "for i < len(data) { use(d.data[i]) }"
makes it harder for the compiler.

Finally, do other minor clean-ups, such as deduplicating switch cases,
and using a switch instead of three chained ifs.

The decoder sees a noticeable speed-up, in particular when decoding
structs.

name           old time/op    new time/op    delta
CodeDecoder-4    29.8ms ± 1%    27.5ms ± 0%  -7.83%  (p=0.002 n=6+6)

name           old speed      new speed      delta
CodeDecoder-4  65.0MB/s ± 1%  70.6MB/s ± 0%  +8.49%  (p=0.002 n=6+6)

Updates #5683.

Change-Id: I9d751e22502221962da696e48996ffdeb777277d
Reviewed-on: https://go-review.googlesource.com/122468
Run-TryBot: Daniel Martí <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>
Reviewed-by: Brad Fitzpatrick <[email protected]>

Author: mvdan

src/encoding/json/decode.go

@@ -332,21 +332,20 @@ func (d *decodeState) skip() {
 
 // scanNext processes the byte at d.data[d.off].
 func (d *decodeState) scanNext() {
-	s, data, i := &d.scan, d.data, d.off
-	if i < len(data) {
-		d.opcode = s.step(s, data[i])
-		d.off = i + 1
+	if d.off < len(d.data) {
+		d.opcode = d.scan.step(&d.scan, d.data[d.off])
+		d.off++
 	} else {
-		d.opcode = s.eof()
-		d.off = len(data) + 1 // mark processed EOF with len+1
+		d.opcode = d.scan.eof()
+		d.off = len(d.data) + 1 // mark processed EOF with len+1
 	}
 }
 
 // scanWhile processes bytes in d.data[d.off:] until it
 // receives a scan code not equal to op.
 func (d *decodeState) scanWhile(op int) {
 	s, data, i := &d.scan, d.data, d.off
-	for i < len(d.data) {
+	for i < len(data) {
 		newOp := s.step(s, data[i])
 		i++
 		if newOp != op {
@@ -356,7 +355,7 @@ func (d *decodeState) scanWhile(op int) {
 		}
 	}
 
-	d.off = len(d.data) + 1 // mark processed EOF with len+1
+	d.off = len(data) + 1 // mark processed EOF with len+1
 	d.opcode = d.scan.eof()
 }
 
@@ -413,11 +412,7 @@ func (d *decodeState) valueQuoted() (interface{}, error) {
 	default:
 		return nil, errPhase
 
-	case scanBeginArray:
-		d.skip()
-		d.scanNext()
-
-	case scanBeginObject:
+	case scanBeginArray, scanBeginObject:
 		d.skip()
 		d.scanNext()
 
@@ -629,6 +624,7 @@ func (d *decodeState) object(v reflect.Value) error {
 		return nil
 	}
 	v = pv
+	t := v.Type()
 
 	// Decoding into nil interface? Switch to non-reflect code.
 	if v.Kind() == reflect.Interface && v.NumMethod() == 0 {
@@ -640,6 +636,8 @@ func (d *decodeState) object(v reflect.Value) error {
 		return nil
 	}
 
+	var fields []field
+
 	// Check type of target:
 	//   struct or
 	//   map[T1]T2 where T1 is string, an integer type,
@@ -648,14 +646,13 @@ func (d *decodeState) object(v reflect.Value) error {
 	case reflect.Map:
 		// Map key must either have string kind, have an integer kind,
 		// or be an encoding.TextUnmarshaler.
-		t := v.Type()
 		switch t.Key().Kind() {
 		case reflect.String,
 			reflect.Int, reflect.Int8, reflect.Int16, reflect.Int32, reflect.Int64,
 			reflect.Uint, reflect.Uint8, reflect.Uint16, reflect.Uint32, reflect.Uint64, reflect.Uintptr:
 		default:
 			if !reflect.PtrTo(t.Key()).Implements(textUnmarshalerType) {
-				d.saveError(&UnmarshalTypeError{Value: "object", Type: v.Type(), Offset: int64(d.off)})
+				d.saveError(&UnmarshalTypeError{Value: "object", Type: t, Offset: int64(d.off)})
 				d.skip()
 				return nil
 			}
@@ -664,9 +661,10 @@ func (d *decodeState) object(v reflect.Value) error {
 			v.Set(reflect.MakeMap(t))
 		}
 	case reflect.Struct:
+		fields = cachedTypeFields(t)
 		// ok
 	default:
-		d.saveError(&UnmarshalTypeError{Value: "object", Type: v.Type(), Offset: int64(d.off)})
+		d.saveError(&UnmarshalTypeError{Value: "object", Type: t, Offset: int64(d.off)})
 		d.skip()
 		return nil
 	}
@@ -698,7 +696,7 @@ func (d *decodeState) object(v reflect.Value) error {
 		destring := false // whether the value is wrapped in a string to be decoded first
 
 		if v.Kind() == reflect.Map {
-			elemType := v.Type().Elem()
+			elemType := t.Elem()
 			if !mapElem.IsValid() {
 				mapElem = reflect.New(elemType).Elem()
 			} else {
@@ -707,7 +705,6 @@ func (d *decodeState) object(v reflect.Value) error {
 			subv = mapElem
 		} else {
 			var f *field
-			fields := cachedTypeFields(v.Type())
 			for i := range fields {
 				ff := &fields[i]
 				if bytes.Equal(ff.nameBytes, key) {
@@ -744,7 +741,7 @@ func (d *decodeState) object(v reflect.Value) error {
 					subv = subv.Field(i)
 				}
 				d.errorContext.Field = f.name
-				d.errorContext.Struct = v.Type()
+				d.errorContext.Struct = t
 			} else if d.disallowUnknownFields {
 				d.saveError(fmt.Errorf("json: unknown field %q", key))
 			}
@@ -785,13 +782,13 @@ func (d *decodeState) object(v reflect.Value) error {
 		// Write value back to map;
 		// if using struct, subv points into struct already.
 		if v.Kind() == reflect.Map {
-			kt := v.Type().Key()
+			kt := t.Key()
 			var kv reflect.Value
 			switch {
 			case kt.Kind() == reflect.String:
 				kv = reflect.ValueOf(key).Convert(kt)
 			case reflect.PtrTo(kt).Implements(textUnmarshalerType):
-				kv = reflect.New(v.Type().Key())
+				kv = reflect.New(kt)
 				if err := d.literalStore(item, kv, true); err != nil {
 					return err
 				}

src/encoding/json/stream.go

@@ -96,19 +96,19 @@ Input:
 		// Look in the buffer for a new value.
 		for i, c := range dec.buf[scanp:] {
 			dec.scan.bytes++
-			v := dec.scan.step(&dec.scan, c)
-			if v == scanEnd {
+			switch dec.scan.step(&dec.scan, c) {
+			case scanEnd:
 				scanp += i
 				break Input
-			}
-			// scanEnd is delayed one byte.
-			// We might block trying to get that byte from src,
-			// so instead invent a space byte.
-			if (v == scanEndObject || v == scanEndArray) && dec.scan.step(&dec.scan, ' ') == scanEnd {
-				scanp += i + 1
-				break Input
-			}
-			if v == scanError {
+			case scanEndObject, scanEndArray:
+				// scanEnd is delayed one byte.
+				// We might block trying to get that byte from src,
+				// so instead invent a space byte.
+				if stateEndValue(&dec.scan, ' ') == scanEnd {
+					scanp += i + 1
+					break Input
+				}
+			case scanError:
 				dec.err = dec.scan.err
 				return 0, dec.scan.err
 			}