proto, runtime/protoiface, internal/impl: add fast-path Merge

Comparing -tags=protoreflect to fast-path:

name                              old time/op    new time/op    delta
pkg:google.golang.org/protobuf/internal/benchmarks goos:linux goarch:amd64
/Clone/google_message1_proto2-12    1.70µs ± 1%    0.30µs ± 1%  -82.64%  (p=0.001 n=7+7)
/Clone/google_message1_proto3-12    1.01µs ± 1%    0.19µs ± 1%  -80.77%  (p=0.000 n=7+8)
/Clone/google_message2-12            818µs ± 8%     141µs ± 6%  -82.78%  (p=0.000 n=8+8)
pkg:google.golang.org/protobuf/internal/benchmarks/micro goos:linux goarch:amd64
EmptyMessage/Clone-12               51.1ns ± 1%    39.3ns ± 3%  -23.03%  (p=0.000 n=7+8)
RepeatedInt32/Clone-12              24.5µs ± 1%     1.1µs ± 3%  -95.64%  (p=0.000 n=8+8)
Required/Clone-12                    978ns ± 1%     132ns ± 2%  -86.46%  (p=0.000 n=8+8)

name                              old alloc/op   new alloc/op   delta
pkg:google.golang.org/protobuf/internal/benchmarks goos:linux goarch:amd64
/Clone/google_message1_proto2-12    1.08kB ± 0%    0.74kB ± 0%  -31.85%  (p=0.000 n=8+8)
/Clone/google_message1_proto3-12      872B ± 0%      544B ± 0%  -37.61%  (p=0.000 n=8+8)
/Clone/google_message2-12            602kB ± 0%     411kB ± 0%  -31.65%  (p=0.000 n=8+8)
pkg:google.golang.org/protobuf/internal/benchmarks/micro goos:linux goarch:amd64
EmptyMessage/Clone-12                96.0B ± 0%     64.0B ± 0%  -33.33%  (p=0.000 n=8+8)
RepeatedInt32/Clone-12              25.4kB ± 0%     3.2kB ± 0%  -87.33%  (p=0.000 n=8+8)
Required/Clone-12                     416B ± 0%      256B ± 0%  -38.46%  (p=0.000 n=8+8)

name                              old allocs/op  new allocs/op  delta
pkg:google.golang.org/protobuf/internal/benchmarks goos:linux goarch:amd64
/Clone/google_message1_proto2-12      52.0 ± 0%      21.0 ± 0%  -59.62%  (p=0.000 n=8+8)
/Clone/google_message1_proto3-12      33.0 ± 0%       3.0 ± 0%  -90.91%  (p=0.000 n=8+8)
/Clone/google_message2-12            22.3k ± 0%      7.5k ± 0%  -66.41%  (p=0.000 n=8+8)
pkg:google.golang.org/protobuf/internal/benchmarks/micro goos:linux goarch:amd64
EmptyMessage/Clone-12                 3.00 ± 0%      2.00 ± 0%  -33.33%  (p=0.000 n=8+8)
RepeatedInt32/Clone-12               1.51k ± 0%     0.00k ± 0%  -99.80%  (p=0.000 n=8+8)
Required/Clone-12                     51.0 ± 0%      18.0 ± 0%  -64.71%  (p=0.000 n=8+8)

Change-Id: Ife9018097c34cb025dc9c4fdd9a61b2f947853c6
Reviewed-on: https://go-review.googlesource.com/c/protobuf/+/219147
Reviewed-by: Joe Tsai <thebrokentoaster@gmail.com>
diff --git a/internal/impl/merge.go b/internal/impl/merge.go
new file mode 100644
index 0000000..20d9dfd
--- /dev/null
+++ b/internal/impl/merge.go
@@ -0,0 +1,176 @@
+// Copyright 2020 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package impl
+
+import (
+	"fmt"
+	"reflect"
+
+	"google.golang.org/protobuf/proto"
+	pref "google.golang.org/protobuf/reflect/protoreflect"
+	piface "google.golang.org/protobuf/runtime/protoiface"
+)
+
+type mergeOptions piface.MergeOptions
+
+func (o mergeOptions) Merge(dst, src proto.Message) {
+	proto.Merge(dst, src)
+}
+
+// merge is protoreflect.Methods.Merge.
+func (mi *MessageInfo) merge(dst, src pref.Message, in piface.MergeInput, opts piface.MergeOptions) piface.MergeOutput {
+	dp, ok := mi.getPointer(dst)
+	if !ok {
+		return piface.MergeOutput{Merged: false}
+	}
+	sp, ok := mi.getPointer(src)
+	if !ok {
+		return piface.MergeOutput{Merged: false}
+	}
+	mi.mergePointer(dp, sp, opts)
+	return piface.MergeOutput{Merged: true}
+}
+
+func (mi *MessageInfo) mergePointer(dst, src pointer, opts mergeOptions) {
+	mi.init()
+	if dst.IsNil() {
+		panic(fmt.Sprintf("invalid value: merging into nil message"))
+	}
+	if src.IsNil() {
+		return
+	}
+	for _, f := range mi.orderedCoderFields {
+		if f.funcs.merge == nil {
+			continue
+		}
+		sfptr := src.Apply(f.offset)
+		if f.isPointer && sfptr.Elem().IsNil() {
+			continue
+		}
+		f.funcs.merge(dst.Apply(f.offset), sfptr, f, opts)
+	}
+	if mi.extensionOffset.IsValid() {
+		sext := src.Apply(mi.extensionOffset).Extensions()
+		dext := dst.Apply(mi.extensionOffset).Extensions()
+		if *dext == nil {
+			*dext = make(map[int32]ExtensionField)
+		}
+		for num, sx := range *sext {
+			xt := sx.Type()
+			xi := getExtensionFieldInfo(xt)
+			if xi.funcs.merge == nil {
+				continue
+			}
+			dx := (*dext)[num]
+			var dv pref.Value
+			if dx.Type() == sx.Type() {
+				dv = dx.Value()
+			}
+			if !dv.IsValid() && xi.unmarshalNeedsValue {
+				dv = xt.New()
+			}
+			dv = xi.funcs.merge(dv, sx.Value(), opts)
+			dx.Set(sx.Type(), dv)
+			(*dext)[num] = dx
+		}
+	}
+	if mi.unknownOffset.IsValid() {
+		du := dst.Apply(mi.unknownOffset).Bytes()
+		su := src.Apply(mi.unknownOffset).Bytes()
+		if len(*su) > 0 {
+			*du = append(*du, *su...)
+		}
+	}
+}
+
+func mergeScalarValue(dst, src pref.Value, opts mergeOptions) pref.Value {
+	return src
+}
+
+func mergeBytesValue(dst, src pref.Value, opts mergeOptions) pref.Value {
+	return pref.ValueOfBytes(append(emptyBuf[:], src.Bytes()...))
+}
+
+func mergeListValue(dst, src pref.Value, opts mergeOptions) pref.Value {
+	dstl := dst.List()
+	srcl := src.List()
+	for i, llen := 0, srcl.Len(); i < llen; i++ {
+		dstl.Append(srcl.Get(i))
+	}
+	return dst
+}
+
+func mergeBytesListValue(dst, src pref.Value, opts mergeOptions) pref.Value {
+	dstl := dst.List()
+	srcl := src.List()
+	for i, llen := 0, srcl.Len(); i < llen; i++ {
+		sb := srcl.Get(i).Bytes()
+		db := append(emptyBuf[:], sb...)
+		dstl.Append(pref.ValueOfBytes(db))
+	}
+	return dst
+}
+
+func mergeMessageListValue(dst, src pref.Value, opts mergeOptions) pref.Value {
+	dstl := dst.List()
+	srcl := src.List()
+	for i, llen := 0, srcl.Len(); i < llen; i++ {
+		sm := srcl.Get(i).Message()
+		dm := proto.Clone(sm.Interface()).ProtoReflect()
+		dstl.Append(pref.ValueOfMessage(dm))
+	}
+	return dst
+}
+
+func mergeMessageValue(dst, src pref.Value, opts mergeOptions) pref.Value {
+	opts.Merge(dst.Message().Interface(), src.Message().Interface())
+	return dst
+}
+
+func mergeMessage(dst, src pointer, f *coderFieldInfo, opts mergeOptions) {
+	if f.mi != nil {
+		if dst.Elem().IsNil() {
+			dst.SetPointer(pointerOfValue(reflect.New(f.mi.GoReflectType.Elem())))
+		}
+		f.mi.mergePointer(dst.Elem(), src.Elem(), opts)
+	} else {
+		dm := dst.AsValueOf(f.ft).Elem()
+		sm := src.AsValueOf(f.ft).Elem()
+		if dm.IsNil() {
+			dm.Set(reflect.New(f.ft.Elem()))
+		}
+		opts.Merge(asMessage(dm), asMessage(sm))
+	}
+}
+
+func mergeMessageSlice(dst, src pointer, f *coderFieldInfo, opts mergeOptions) {
+	for _, sp := range src.PointerSlice() {
+		dm := reflect.New(f.ft.Elem().Elem())
+		if f.mi != nil {
+			f.mi.mergePointer(pointerOfValue(dm), sp, opts)
+		} else {
+			opts.Merge(asMessage(dm), asMessage(sp.AsValueOf(f.ft.Elem().Elem())))
+		}
+		dst.AppendPointerSlice(pointerOfValue(dm))
+	}
+}
+
+func mergeBytes(dst, src pointer, _ *coderFieldInfo, _ mergeOptions) {
+	*dst.Bytes() = append(emptyBuf[:], *src.Bytes()...)
+}
+
+func mergeBytesNoZero(dst, src pointer, _ *coderFieldInfo, _ mergeOptions) {
+	v := *src.Bytes()
+	if len(v) > 0 {
+		*dst.Bytes() = append(emptyBuf[:], v...)
+	}
+}
+
+func mergeBytesSlice(dst, src pointer, _ *coderFieldInfo, _ mergeOptions) {
+	ds := dst.BytesSlice()
+	for _, v := range *src.BytesSlice() {
+		*ds = append(*ds, append(emptyBuf[:], v...))
+	}
+}