Blink-compatible serialization of oddball values.

src/value-serializer.cc

Index: src/value-serializer.cc
diff --git a/src/value-serializer.cc b/src/value-serializer.cc
new file mode 100644
index 0000000000000000000000000000000000000000..50fde61bdc73e7d067cc5429e55cd8f527770ae8
--- /dev/null
+++ b/src/value-serializer.cc
@@ -0,0 +1,165 @@
+// Copyright 2016 the V8 project authors. All rights reserved.
+// Use of this source code is governed by a BSD-style license that can be
+// found in the LICENSE file.
+
+#include "src/value-serializer.h"
+
+#include <type_traits>
+
+#include "src/base/logging.h"
+#include "src/factory.h"
+#include "src/handles-inl.h"
+#include "src/isolate.h"
+#include "src/objects-inl.h"
+#include "src/objects.h"
+
+namespace v8 {
+namespace internal {
+
+static const uint32_t kLatestVersion = 9;
+
+enum class SerializationTag : uint8_t {
+  kVersion = 0xFF,
+  kPadding = '\0',
+  kVerifyObjectCount = '?',
+  kUndefined = '_',
+  kNull = '0',
+  kTrue = 'T',
+  kFalse = 'F',
+};
+
+ValueSerializer::ValueSerializer() {}
+
+ValueSerializer::~ValueSerializer() {}
+
+void ValueSerializer::WriteHeader() {
+  WriteTag(SerializationTag::kVersion);
+  WriteVarint(kLatestVersion);
+}
+
+void ValueSerializer::WriteTag(SerializationTag tag) {
+  buffer_.push_back(static_cast<uint8_t>(tag));
+}
+
+template <typename T>
+void ValueSerializer::WriteVarint(T value) {
+  static_assert(std::is_integral<T>::value && std::is_unsigned<T>::value,
+                "Only unsigned integer types can be written as varints.");
+  uint8_t stack_buffer[sizeof(T) * 8 / 7 + 1];
+  uint8_t* next_byte = &stack_buffer[0];
+  do {
+    *next_byte = (value & 0x7f) | 0x80;
+    next_byte++;
+    value >>= 7;
+  } while (value);
+  *(next_byte - 1) &= 0x7f;
+  buffer_.insert(buffer_.end(), stack_buffer, next_byte);
+}
+
+Maybe<bool> ValueSerializer::WriteObject(Handle<Object> object) {
+  if (object->IsSmi()) UNIMPLEMENTED();
+
+  DCHECK(object->IsHeapObject());
+  switch (HeapObject::cast(*object)->map()->instance_type()) {

[Camillo Bruni, 2016/08/12 09:54:44]

This is much slower than directly comparing against the heap constants (e.g.
isolate()->heap()->undefined_value() and friends).
Given that you will have to do 4 checks (true, false, null undefined) it might
still be faster to read out the maps' instance_type.

I definitely like the code better how it is currently, but just wanted to add
the for future reference once you start optimizing ;)

[jbroman, 2016/08/12 15:16:34]

I'll be branching on the instance type of all objects anyhow, so I'm not sure
whether it'll be faster to check for oddballs first separately, or just have
them go through the same switch.

For what it's worth, this is how json-stringifier does it, and I've assumed that
it's fairly optimized already.
https://chromium.googlesource.com/v8/v8/+/bbd0a09/src/json-stringifier.cc#294

Nonetheless I'll definitely keep that in mind when I later try to
micro-optimize.

+    case ODDBALL_TYPE:
+      WriteOddball(Oddball::cast(*object));
+      return Just(true);
+    default:
+      UNIMPLEMENTED();
+      return Nothing<bool>();
+  }
+}
+
+void ValueSerializer::WriteOddball(Oddball* oddball) {
+  SerializationTag tag = SerializationTag::kUndefined;
+  switch (oddball->kind()) {
+    case Oddball::kUndefined:
+      tag = SerializationTag::kUndefined;
+      break;
+    case Oddball::kFalse:
+      tag = SerializationTag::kFalse;
+      break;
+    case Oddball::kTrue:
+      tag = SerializationTag::kTrue;
+      break;
+    case Oddball::kNull:
+      tag = SerializationTag::kNull;
+      break;
+    default:
+      UNREACHABLE();
+      break;
+  }
+  WriteTag(tag);
+}
+
+ValueDeserializer::ValueDeserializer(Isolate* isolate,
+                                     Vector<const uint8_t> data)
+    : isolate_(isolate),
+      position_(data.start()),
+      end_(data.start() + data.length()) {}
+
+ValueDeserializer::~ValueDeserializer() {}
+
+Maybe<bool> ValueDeserializer::ReadHeader() {
+  if (position_ < end_ &&
+      *position_ == static_cast<uint8_t>(SerializationTag::kVersion)) {
+    ReadTag().ToChecked();
+    if (!ReadVarint<uint32_t>().To(&version_)) return Nothing<bool>();
+    if (version_ > kLatestVersion) return Nothing<bool>();
+  }
+  return Just(true);
+}
+
+Maybe<SerializationTag> ValueDeserializer::ReadTag() {
+  SerializationTag tag;
+  do {
+    if (position_ >= end_) return Nothing<SerializationTag>();
+    tag = static_cast<SerializationTag>(*position_);
+    position_++;
+  } while (tag == SerializationTag::kPadding);
+  return Just(tag);
+}
+
+template <typename T>
+Maybe<T> ValueDeserializer::ReadVarint() {
+  static_assert(std::is_integral<T>::value && std::is_unsigned<T>::value,
+                "Only unsigned integer types can be read as varints.");
+  T value = 0;
+  unsigned shift = 0;
+  bool has_another_byte;
+  do {
+    if (position_ >= end_) return Nothing<T>();
+    uint8_t byte = *position_;
+    if (V8_LIKELY(shift < sizeof(T) * 8)) {
+      value |= (byte & 0x7f) << shift;
+      shift += 7;
+    }
+    has_another_byte = byte & 0x80;
+    position_++;
+  } while (has_another_byte);
+  return Just(value);
+}
+
+MaybeHandle<Object> ValueDeserializer::ReadObject() {
+  SerializationTag tag;
+  if (!ReadTag().To(&tag)) return MaybeHandle<Object>();
+  switch (tag) {
+    case SerializationTag::kVerifyObjectCount:
+      // Read the count and ignore it.
+      if (ReadVarint<uint32_t>().IsNothing()) return MaybeHandle<Object>();
+      return ReadObject();
+    case SerializationTag::kUndefined:
+      return isolate_->factory()->undefined_value();
+    case SerializationTag::kNull:
+      return isolate_->factory()->null_value();
+    case SerializationTag::kTrue:
+      return isolate_->factory()->true_value();
+    case SerializationTag::kFalse:
+      return isolate_->factory()->false_value();
+    default:
+      return MaybeHandle<Object>();
+  }
+}
+
+}  // namespace internal
+}  // namespace v8