src: allow snapshotting from the embedder API

PR-URL: https://github.com/nodejs/node/pull/45888
Reviewed-By: James M Snell <jasnell@gmail.com>
Reviewed-By: Joyee Cheung <joyeec9h3@gmail.com>

lib/internal/main/mksnapshot.js

@@ -12,6 +12,7 @@ const {
 const binding = internalBinding('mksnapshot');
 const { BuiltinModule } = require('internal/bootstrap/loaders');
 const {
+  getEmbedderEntryFunction,
   compileSerializeMain,
 } = binding;
 
@@ -119,14 +120,21 @@ function main() {
     prepareMainThreadExecution
   } = require('internal/process/pre_execution');
 
-  prepareMainThreadExecution(true, false);
+  let serializeMainFunction = getEmbedderEntryFunction();
+  const serializeMainArgs = [requireForUserSnapshot];
 
+  if (serializeMainFunction) { // embedded case
+    prepareMainThreadExecution(false, false);
+  } else {
+    prepareMainThreadExecution(true, false);
     const file = process.argv[1];
     const path = require('path');
     const filename = path.resolve(file);
     const dirname = path.dirname(filename);
     const source = readFileSync(file, 'utf-8');
-  const serializeMainFunction = compileSerializeMain(filename, source);
+    serializeMainFunction = compileSerializeMain(filename, source);
+    serializeMainArgs.push(filename, dirname);
+  }
 
   const {
     initializeCallbacks,
@@ -146,10 +154,9 @@ function main() {
 
   if (getOptionValue('--inspect-brk')) {
     internalBinding('inspector').callAndPauseOnStart(
-      serializeMainFunction, undefined,
+      serializeMainFunction, undefined, ...serializeMainArgs);
-      requireForUserSnapshot, filename, dirname);
   } else {
-    serializeMainFunction(requireForUserSnapshot, filename, dirname);
+    serializeMainFunction(...serializeMainArgs);
   }
 
   addSerializeCallback(() => {

src/api/embed_helpers.cc

@@ -14,6 +14,7 @@ using v8::Locker;
 using v8::Maybe;
 using v8::Nothing;
 using v8::SealHandleScope;
+using v8::SnapshotCreator;
 
 namespace node {
 
@@ -78,16 +79,18 @@ struct CommonEnvironmentSetup::Impl {
   MultiIsolatePlatform* platform = nullptr;
   uv_loop_t loop;
   std::shared_ptr<ArrayBufferAllocator> allocator;
+  std::optional<SnapshotCreator> snapshot_creator;
   Isolate* isolate = nullptr;
   DeleteFnPtr<IsolateData, FreeIsolateData> isolate_data;
   DeleteFnPtr<Environment, FreeEnvironment> env;
-  Global<Context> context;
+  Global<Context> main_context;
 };
 
 CommonEnvironmentSetup::CommonEnvironmentSetup(
     MultiIsolatePlatform* platform,
     std::vector<std::string>* errors,
     const EmbedderSnapshotData* snapshot_data,
+    uint32_t flags,
     std::function<Environment*(const CommonEnvironmentSetup*)> make_env)
     : impl_(new Impl()) {
   CHECK_NOT_NULL(platform);
@@ -105,28 +108,43 @@ CommonEnvironmentSetup::CommonEnvironmentSetup(
   }
   loop->data = this;
 
+  Isolate* isolate;
+  if (flags & Flags::kIsForSnapshotting) {
+    const std::vector<intptr_t>& external_references =
+        SnapshotBuilder::CollectExternalReferences();
+    isolate = impl_->isolate = Isolate::Allocate();
+    // Must be done before the SnapshotCreator creation so  that the
+    // memory reducer can be initialized.
+    platform->RegisterIsolate(isolate, loop);
+    impl_->snapshot_creator.emplace(isolate, external_references.data());
+    isolate->SetCaptureStackTraceForUncaughtExceptions(
+        true, 10, v8::StackTrace::StackTraceOptions::kDetailed);
+    SetIsolateMiscHandlers(isolate, {});
+  } else {
     impl_->allocator = ArrayBufferAllocator::Create();
-  impl_->isolate =
+    isolate = impl_->isolate =
         NewIsolate(impl_->allocator, &impl_->loop, platform, snapshot_data);
-  Isolate* isolate = impl_->isolate;
+  }
 
   {
     Locker locker(isolate);
     Isolate::Scope isolate_scope(isolate);
     impl_->isolate_data.reset(CreateIsolateData(
         isolate, loop, platform, impl_->allocator.get(), snapshot_data));
+    impl_->isolate_data->options()->build_snapshot =
+        impl_->snapshot_creator.has_value();
 
     HandleScope handle_scope(isolate);
     if (snapshot_data) {
       impl_->env.reset(make_env(this));
       if (impl_->env) {
-        impl_->context.Reset(isolate, impl_->env->context());
+        impl_->main_context.Reset(isolate, impl_->env->context());
       }
       return;
     }
 
     Local<Context> context = NewContext(isolate);
-    impl_->context.Reset(isolate, context);
+    impl_->main_context.Reset(isolate, context);
     if (context.IsEmpty()) {
       errors->push_back("Failed to initialize V8 Context");
       return;
@@ -141,7 +159,37 @@ CommonEnvironmentSetup::CommonEnvironmentSetup(
     MultiIsolatePlatform* platform,
     std::vector<std::string>* errors,
     std::function<Environment*(const CommonEnvironmentSetup*)> make_env)
-    : CommonEnvironmentSetup(platform, errors, nullptr, make_env) {}
+    : CommonEnvironmentSetup(platform, errors, nullptr, false, make_env) {}
+
+std::unique_ptr<CommonEnvironmentSetup>
+CommonEnvironmentSetup::CreateForSnapshotting(
+    MultiIsolatePlatform* platform,
+    std::vector<std::string>* errors,
+    const std::vector<std::string>& args,
+    const std::vector<std::string>& exec_args) {
+  // It's not guaranteed that a context that goes through
+  // v8_inspector::V8Inspector::contextCreated() is runtime-independent,
+  // so do not start the inspector on the main context when building
+  // the default snapshot.
+  uint64_t env_flags =
+      EnvironmentFlags::kDefaultFlags | EnvironmentFlags::kNoCreateInspector;
+
+  auto ret = std::unique_ptr<CommonEnvironmentSetup>(new CommonEnvironmentSetup(
+      platform,
+      errors,
+      nullptr,
+      true,
+      [&](const CommonEnvironmentSetup* setup) -> Environment* {
+        return CreateEnvironment(
+            setup->isolate_data(),
+            setup->context(),
+            args,
+            exec_args,
+            static_cast<EnvironmentFlags::Flags>(env_flags));
+      }));
+  if (!errors->empty()) ret.reset();
+  return ret;
+}
 
 CommonEnvironmentSetup::~CommonEnvironmentSetup() {
   if (impl_->isolate != nullptr) {
@@ -150,7 +198,7 @@ CommonEnvironmentSetup::~CommonEnvironmentSetup() {
       Locker locker(isolate);
       Isolate::Scope isolate_scope(isolate);
 
-      impl_->context.Reset();
+      impl_->main_context.Reset();
       impl_->env.reset();
       impl_->isolate_data.reset();
     }
@@ -160,6 +208,9 @@ CommonEnvironmentSetup::~CommonEnvironmentSetup() {
       *static_cast<bool*>(data) = true;
     }, &platform_finished);
     impl_->platform->UnregisterIsolate(isolate);
+    if (impl_->snapshot_creator.has_value())
+      impl_->snapshot_creator.reset();
+    else
       isolate->Dispose();
 
     // Wait until the platform has cleaned up all relevant resources.
@@ -173,6 +224,21 @@ CommonEnvironmentSetup::~CommonEnvironmentSetup() {
   delete impl_;
 }
 
+EmbedderSnapshotData::Pointer CommonEnvironmentSetup::CreateSnapshot() {
+  CHECK_NOT_NULL(snapshot_creator());
+  SnapshotData* snapshot_data = new SnapshotData();
+  EmbedderSnapshotData::Pointer result{
+      new EmbedderSnapshotData(snapshot_data, true)};
+
+  auto exit_code = SnapshotBuilder::CreateSnapshot(
+      snapshot_data,
+      this,
+      static_cast<uint8_t>(SnapshotMetadata::Type::kFullyCustomized));
+  if (exit_code != ExitCode::kNoFailure) return {};
+
+  return result;
+}
+
 Maybe<int> SpinEventLoop(Environment* env) {
   Maybe<ExitCode> result = SpinEventLoopInternal(env);
   if (result.IsNothing()) {
@@ -203,7 +269,11 @@ Environment* CommonEnvironmentSetup::env() const {
 }
 
 v8::Local<v8::Context> CommonEnvironmentSetup::context() const {
-  return impl_->context.Get(impl_->isolate);
+  return impl_->main_context.Get(impl_->isolate);
+}
+
+v8::SnapshotCreator* CommonEnvironmentSetup::snapshot_creator() {
+  return impl_->snapshot_creator ? &impl_->snapshot_creator.value() : nullptr;
 }
 
 void EmbedderSnapshotData::DeleteSnapshotData::operator()(
@@ -232,6 +302,10 @@ EmbedderSnapshotData::Pointer EmbedderSnapshotData::FromFile(FILE* in) {
   return result;
 }
 
+void EmbedderSnapshotData::ToFile(FILE* out) const {
+  impl_->ToBlob(out);
+}
+
 EmbedderSnapshotData::EmbedderSnapshotData(const SnapshotData* impl,
                                            bool owns_impl)
     : impl_(impl), owns_impl_(owns_impl) {}

src/env-inl.h

@@ -444,6 +444,16 @@ inline builtins::BuiltinLoader* Environment::builtin_loader() {
   return &builtin_loader_;
 }
 
+inline const StartExecutionCallback&
+Environment::embedder_mksnapshot_entry_point() const {
+  return embedder_mksnapshot_entry_point_;
+}
+
+inline void Environment::set_embedder_mksnapshot_entry_point(
+    StartExecutionCallback&& fn) {
+  embedder_mksnapshot_entry_point_ = std::move(fn);
+}
+
 inline double Environment::new_async_id() {
   async_hooks()->async_id_fields()[AsyncHooks::kAsyncIdCounter] += 1;
   return async_hooks()->async_id_fields()[AsyncHooks::kAsyncIdCounter];

src/env.h

@@ -961,6 +961,9 @@ class Environment : public MemoryRetainer {
 
 #endif  // HAVE_INSPECTOR
 
+  inline const StartExecutionCallback& embedder_mksnapshot_entry_point() const;
+  inline void set_embedder_mksnapshot_entry_point(StartExecutionCallback&& fn);
+
   inline void set_process_exit_handler(
       std::function<void(Environment*, ExitCode)>&& handler);
 
@@ -1144,6 +1147,7 @@ class Environment : public MemoryRetainer {
   std::unique_ptr<Realm> principal_realm_ = nullptr;
 
   builtins::BuiltinLoader builtin_loader_;
+  StartExecutionCallback embedder_mksnapshot_entry_point_;
 
   // Used by allocate_managed_buffer() and release_managed_buffer() to keep
   // track of the BackingStore for a given pointer.

src/node.cc

@@ -276,14 +276,21 @@ MaybeLocal<Value> StartExecution(Environment* env, StartExecutionCallback cb) {
   if (cb != nullptr) {
     EscapableHandleScope scope(env->isolate());
 
-    if (StartExecution(env, "internal/main/environment").IsEmpty()) return {};
+    if (env->isolate_data()->options()->build_snapshot) {
+      // TODO(addaleax): pass the callback to the main script more directly,
+      // e.g. by making StartExecution(env, builtin) parametrizable
+      env->set_embedder_mksnapshot_entry_point(std::move(cb));
+      auto reset_entry_point =
+          OnScopeLeave([&]() { env->set_embedder_mksnapshot_entry_point({}); });
 
-    StartExecutionCallbackInfo info = {
+      return StartExecution(env, "internal/main/mksnapshot");
+    }
+
+    if (StartExecution(env, "internal/main/environment").IsEmpty()) return {};
+    return scope.EscapeMaybe(cb({
         env->process_object(),
         env->builtin_module_require(),
-    };
+    }));
-
-    return scope.EscapeMaybe(cb(info));
   }
 
   // TODO(joyeecheung): move these conditions into JS land and let the

src/node.h

@@ -511,11 +511,16 @@ class EmbedderSnapshotData {
   static Pointer BuiltinSnapshotData();
 
   // Return an EmbedderSnapshotData object that is based on an input file.
-  // Calling this method will not consume but not close the FILE* handle.
+  // Calling this method will consume but not close the FILE* handle.
   // The FILE* handle can be closed immediately following this call.
   // If the snapshot is invalid, this returns an empty pointer.
   static Pointer FromFile(FILE* in);
 
+  // Write this EmbedderSnapshotData object to an output file.
+  // Calling this method will not close the FILE* handle.
+  // The FILE* handle can be closed immediately following this call.
+  void ToFile(FILE* out) const;
+
   // Returns whether custom snapshots can be used. Currently, this means
   // that V8 was configured without the shared-readonly-heap feature.
   static bool CanUseCustomSnapshotPerIsolate();
@@ -532,6 +537,7 @@ class EmbedderSnapshotData {
   const SnapshotData* impl_;
   bool owns_impl_;
   friend struct SnapshotData;
+  friend class CommonEnvironmentSetup;
 };
 
 // Overriding IsolateSettings may produce unexpected behavior
@@ -817,13 +823,35 @@ class NODE_EXTERN CommonEnvironmentSetup {
       std::vector<std::string>* errors,
       EnvironmentArgs&&... env_args);
   template <typename... EnvironmentArgs>
-  static std::unique_ptr<CommonEnvironmentSetup> CreateWithSnapshot(
+  static std::unique_ptr<CommonEnvironmentSetup> CreateFromSnapshot(
       MultiIsolatePlatform* platform,
       std::vector<std::string>* errors,
       const EmbedderSnapshotData* snapshot_data,
       EnvironmentArgs&&... env_args);
 
+  // Create an embedding setup which will be used for creating a snapshot
+  // using CreateSnapshot().
+  //
+  // This will create and attach a v8::SnapshotCreator to this instance,
+  // and the same restrictions apply to this instance that also apply to
+  // other V8 snapshotting environments.
+  // Not all Node.js APIs are supported in this case. Currently, there is
+  // no support for native/host objects other than Node.js builtins
+  // in the snapshot.
+  //
+  // Snapshots are an *experimental* feature. In particular, the embedder API
+  // exposed through this class is subject to change or removal between Node.js
+  // versions, including possible API and ABI breakage.
+  static std::unique_ptr<CommonEnvironmentSetup> CreateForSnapshotting(
+      MultiIsolatePlatform* platform,
+      std::vector<std::string>* errors,
+      const std::vector<std::string>& args = {},
+      const std::vector<std::string>& exec_args = {});
+  EmbedderSnapshotData::Pointer CreateSnapshot();
+
   struct uv_loop_s* event_loop() const;
+  v8::SnapshotCreator* snapshot_creator();
+  // Empty for snapshotting environments.
   std::shared_ptr<ArrayBufferAllocator> array_buffer_allocator() const;
   v8::Isolate* isolate() const;
   IsolateData* isolate_data() const;
@@ -836,8 +864,14 @@ class NODE_EXTERN CommonEnvironmentSetup {
   CommonEnvironmentSetup& operator=(CommonEnvironmentSetup&&) = delete;
 
  private:
+  enum Flags : uint32_t {
+    kNoFlags = 0,
+    kIsForSnapshotting = 1,
+  };
+
   struct Impl;
   Impl* impl_;
+
   CommonEnvironmentSetup(
       MultiIsolatePlatform*,
       std::vector<std::string>*,
@@ -846,6 +880,7 @@ class NODE_EXTERN CommonEnvironmentSetup {
       MultiIsolatePlatform*,
       std::vector<std::string>*,
       const EmbedderSnapshotData*,
+      uint32_t flags,
       std::function<Environment*(const CommonEnvironmentSetup*)>);
 };
 
@@ -865,11 +900,11 @@ std::unique_ptr<CommonEnvironmentSetup> CommonEnvironmentSetup::Create(
   if (!errors->empty()) ret.reset();
   return ret;
 }
-// Implementation for ::CreateWithSnapshot -- the ::Create() method
+// Implementation for ::CreateFromSnapshot -- the ::Create() method
 // could call this with a nullptr snapshot_data in a major version.
 template <typename... EnvironmentArgs>
 std::unique_ptr<CommonEnvironmentSetup>
-CommonEnvironmentSetup::CreateWithSnapshot(
+CommonEnvironmentSetup::CreateFromSnapshot(
     MultiIsolatePlatform* platform,
     std::vector<std::string>* errors,
     const EmbedderSnapshotData* snapshot_data,
@@ -878,6 +913,7 @@ CommonEnvironmentSetup::CreateWithSnapshot(
       platform,
       errors,
       snapshot_data,
+      Flags::kNoFlags,
       [&](const CommonEnvironmentSetup* setup) -> Environment* {
         return CreateEnvironment(setup->isolate_data(),
                                  setup->context(),

src/node_main_instance.cc

@@ -29,34 +29,6 @@ using v8::Isolate;
 using v8::Local;
 using v8::Locker;
 
-NodeMainInstance::NodeMainInstance(Isolate* isolate,
-                                   uv_loop_t* event_loop,
-                                   MultiIsolatePlatform* platform,
-                                   const std::vector<std::string>& args,
-                                   const std::vector<std::string>& exec_args)
-    : args_(args),
-      exec_args_(exec_args),
-      array_buffer_allocator_(nullptr),
-      isolate_(isolate),
-      platform_(platform),
-      isolate_data_(nullptr),
-      snapshot_data_(nullptr) {
-  isolate_data_ =
-      std::make_unique<IsolateData>(isolate_, event_loop, platform, nullptr);
-
-  SetIsolateMiscHandlers(isolate_, {});
-}
-
-std::unique_ptr<NodeMainInstance> NodeMainInstance::Create(
-    Isolate* isolate,
-    uv_loop_t* event_loop,
-    MultiIsolatePlatform* platform,
-    const std::vector<std::string>& args,
-    const std::vector<std::string>& exec_args) {
-  return std::unique_ptr<NodeMainInstance>(
-      new NodeMainInstance(isolate, event_loop, platform, args, exec_args));
-}
-
 NodeMainInstance::NodeMainInstance(const SnapshotData* snapshot_data,
                                    uv_loop_t* event_loop,
                                    MultiIsolatePlatform* platform,
@@ -88,13 +60,6 @@ NodeMainInstance::NodeMainInstance(const SnapshotData* snapshot_data,
       isolate_params_->constraints.max_young_generation_size_in_bytes();
 }
 
-void NodeMainInstance::Dispose() {
-  // This should only be called on a main instance that does not own its
-  // isolate.
-  CHECK_NULL(isolate_params_);
-  platform_->DrainTasks(isolate_);
-}
-
 NodeMainInstance::~NodeMainInstance() {
   if (isolate_params_ == nullptr) {
     return;

src/node_main_instance.h

@@ -22,33 +22,6 @@ struct SnapshotData;
 // We may be able to create an abstract class to reuse some of the routines.
 class NodeMainInstance {
  public:
-  // To create a main instance that does not own the isolate,
-  // The caller needs to do:
-  //
-  //   Isolate* isolate = Isolate::Allocate();
-  //   platform->RegisterIsolate(isolate, loop);
-  //   isolate->Initialize(...);
-  //   isolate->Enter();
-  //   std::unique_ptr<NodeMainInstance> main_instance =
-  //       NodeMainInstance::Create(isolate, loop, args, exec_args);
-  //
-  // When tearing it down:
-  //
-  //   main_instance->Cleanup();  // While the isolate is entered
-  //   isolate->Exit();
-  //   isolate->Dispose();
-  //   platform->UnregisterIsolate(isolate);
-  //
-  // After calling Dispose() the main_instance is no longer accessible.
-  static std::unique_ptr<NodeMainInstance> Create(
-      v8::Isolate* isolate,
-      uv_loop_t* event_loop,
-      MultiIsolatePlatform* platform,
-      const std::vector<std::string>& args,
-      const std::vector<std::string>& exec_args);
-
-  void Dispose();
-
   // Create a main instance that owns the isolate
   NodeMainInstance(const SnapshotData* snapshot_data,
                    uv_loop_t* event_loop,

src/node_options.cc

@@ -778,8 +778,7 @@ PerIsolateOptionsParser::PerIsolateOptionsParser(
   Implies("--harmony-shadow-realm", "--experimental-shadow-realm");
   ImpliesNot("--no-harmony-shadow-realm", "--experimental-shadow-realm");
   AddOption("--build-snapshot",
-            "Generate a snapshot blob when the process exits."
+            "Generate a snapshot blob when the process exits.",
-            " Currently only supported in the node_mksnapshot binary.",
             &PerIsolateOptions::build_snapshot,
             kDisallowedInEnvvar);
 

src/node_snapshot_builder.h

@@ -31,9 +31,14 @@ class NODE_EXTERN_PRIVATE SnapshotBuilder {
   static void InitializeIsolateParams(const SnapshotData* data,
                                       v8::Isolate::CreateParams* params);
 
- private:
   static const std::vector<intptr_t>& CollectExternalReferences();
 
+  static ExitCode CreateSnapshot(
+      SnapshotData* out,
+      CommonEnvironmentSetup* setup,
+      /*SnapshotMetadata::Type*/ uint8_t snapshot_type);
+
+ private:
   static std::unique_ptr<ExternalReferenceRegistry> registry_;
 };
 }  // namespace node

src/node_snapshotable.cc

@@ -33,6 +33,7 @@ using v8::FunctionCallbackInfo;
 using v8::HandleScope;
 using v8::Isolate;
 using v8::Local;
+using v8::MaybeLocal;
 using v8::Object;
 using v8::ObjectTemplate;
 using v8::ScriptCompiler;
@@ -1101,38 +1102,15 @@ void SnapshotBuilder::InitializeIsolateParams(const SnapshotData* data,
 ExitCode SnapshotBuilder::Generate(SnapshotData* out,
                                    const std::vector<std::string> args,
                                    const std::vector<std::string> exec_args) {
-  const std::vector<intptr_t>& external_references =
+  std::vector<std::string> errors;
-      CollectExternalReferences();
+  auto setup = CommonEnvironmentSetup::CreateForSnapshotting(
-  Isolate* isolate = Isolate::Allocate();
+      per_process::v8_platform.Platform(), &errors, args, exec_args);
-  // Must be done before the SnapshotCreator creation so  that the
+  if (!setup) {
-  // memory reducer can be initialized.
+    for (const std::string& err : errors)
-  per_process::v8_platform.Platform()->RegisterIsolate(isolate,
+      fprintf(stderr, "%s: %s\n", args[0].c_str(), err.c_str());
-                                                       uv_default_loop());
+    return ExitCode::kBootstrapFailure;
-
-  SnapshotCreator creator(isolate, external_references.data());
-
-  isolate->SetCaptureStackTraceForUncaughtExceptions(
-      true, 10, v8::StackTrace::StackTraceOptions::kDetailed);
-
-  Environment* env = nullptr;
-  std::unique_ptr<NodeMainInstance> main_instance =
-      NodeMainInstance::Create(isolate,
-                               uv_default_loop(),
-                               per_process::v8_platform.Platform(),
-                               args,
-                               exec_args);
-
-  // The cleanups should be done in case of an early exit due to errors.
-  auto cleanup = OnScopeLeave([&]() {
-    // Must be done while the snapshot creator isolate is entered i.e. the
-    // creator is still alive. The snapshot creator destructor will destroy
-    // the isolate.
-    if (env != nullptr) {
-      FreeEnvironment(env);
   }
-    main_instance->Dispose();
+  Isolate* isolate = setup->isolate();
-    per_process::v8_platform.Platform()->UnregisterIsolate(isolate);
-  });
 
   // It's only possible to be kDefault in node_mksnapshot.
   SnapshotMetadata::Type snapshot_type =
@@ -1151,57 +1129,11 @@ ExitCode SnapshotBuilder::Generate(SnapshotData* out,
       }
     });
 
-    // The default context with only things created by V8.
-    Local<Context> default_context = Context::New(isolate);
-
-    // The context used by the vm module.
-    Local<Context> vm_context;
-    {
-      Local<ObjectTemplate> global_template =
-          main_instance->isolate_data()->contextify_global_template();
-      CHECK(!global_template.IsEmpty());
-      if (!contextify::ContextifyContext::CreateV8Context(
-               isolate, global_template, nullptr, nullptr)
-               .ToLocal(&vm_context)) {
-        return ExitCode::kStartupSnapshotFailure;
-      }
-    }
-
-    // The Node.js-specific context with primodials, can be used by workers
-    // TODO(joyeecheung): investigate if this can be used by vm contexts
-    // without breaking compatibility.
-    Local<Context> base_context = NewContext(isolate);
-    if (base_context.IsEmpty()) {
-      return ExitCode::kBootstrapFailure;
-    }
-    ResetContextSettingsBeforeSnapshot(base_context);
-
-    Local<Context> main_context = NewContext(isolate);
-    if (main_context.IsEmpty()) {
-      return ExitCode::kBootstrapFailure;
-    }
     // Initialize the main instance context.
     {
-      Context::Scope context_scope(main_context);
+      Context::Scope context_scope(setup->context());
+      Environment* env = setup->env();
 
-      // Create the environment.
-      // It's not guaranteed that a context that goes through
-      // v8_inspector::V8Inspector::contextCreated() is runtime-independent,
-      // so do not start the inspector on the main context when building
-      // the default snapshot.
-      uint64_t env_flags = EnvironmentFlags::kDefaultFlags |
-                           EnvironmentFlags::kNoCreateInspector;
-
-      env = CreateEnvironment(main_instance->isolate_data(),
-                              main_context,
-                              args,
-                              exec_args,
-                              static_cast<EnvironmentFlags::Flags>(env_flags));
-
-      // This already ran scripts in lib/internal/bootstrap/, if it fails return
-      if (env == nullptr) {
-        return ExitCode::kBootstrapFailure;
-      }
       // If --build-snapshot is true, lib/internal/main/mksnapshot.js would be
       // loaded via LoadEnvironment() to execute process.argv[1] as the entry
       // point (we currently only support this kind of entry point, but we
@@ -1223,6 +1155,52 @@ ExitCode SnapshotBuilder::Generate(SnapshotData* out,
           return exit_code;
         }
       }
+    }
+  }
+
+  return CreateSnapshot(out, setup.get(), static_cast<uint8_t>(snapshot_type));
+}
+
+ExitCode SnapshotBuilder::CreateSnapshot(SnapshotData* out,
+                                         CommonEnvironmentSetup* setup,
+                                         uint8_t snapshot_type_u8) {
+  SnapshotMetadata::Type snapshot_type =
+      static_cast<SnapshotMetadata::Type>(snapshot_type_u8);
+  Isolate* isolate = setup->isolate();
+  Environment* env = setup->env();
+  SnapshotCreator* creator = setup->snapshot_creator();
+
+  {
+    HandleScope scope(isolate);
+    Local<Context> main_context = setup->context();
+
+    // The default context with only things created by V8.
+    Local<Context> default_context = Context::New(isolate);
+
+    // The context used by the vm module.
+    Local<Context> vm_context;
+    {
+      Local<ObjectTemplate> global_template =
+          setup->isolate_data()->contextify_global_template();
+      CHECK(!global_template.IsEmpty());
+      if (!contextify::ContextifyContext::CreateV8Context(
+               isolate, global_template, nullptr, nullptr)
+               .ToLocal(&vm_context)) {
+        return ExitCode::kStartupSnapshotFailure;
+      }
+    }
+
+    // The Node.js-specific context with primodials, can be used by workers
+    // TODO(joyeecheung): investigate if this can be used by vm contexts
+    // without breaking compatibility.
+    Local<Context> base_context = NewContext(isolate);
+    if (base_context.IsEmpty()) {
+      return ExitCode::kBootstrapFailure;
+    }
+    ResetContextSettingsBeforeSnapshot(base_context);
+
+    {
+      Context::Scope context_scope(main_context);
 
       if (per_process::enabled_debug_list.enabled(DebugCategory::MKSNAPSHOT)) {
         env->ForEachRealm([](Realm* realm) { realm->PrintInfoForSnapshot(); });
@@ -1230,9 +1208,8 @@ ExitCode SnapshotBuilder::Generate(SnapshotData* out,
       }
 
       // Serialize the native states
-      out->isolate_data_info =
+      out->isolate_data_info = setup->isolate_data()->Serialize(creator);
-          main_instance->isolate_data()->Serialize(&creator);
+      out->env_info = env->Serialize(creator);
-      out->env_info = env->Serialize(&creator);
 
 #ifdef NODE_USE_NODE_CODE_CACHE
       // Regenerate all the code cache.
@@ -1255,19 +1232,19 @@ ExitCode SnapshotBuilder::Generate(SnapshotData* out,
     // Global handles to the contexts can't be disposed before the
     // blob is created. So initialize all the contexts before adding them.
     // TODO(joyeecheung): figure out how to remove this restriction.
-    creator.SetDefaultContext(default_context);
+    creator->SetDefaultContext(default_context);
-    size_t index = creator.AddContext(vm_context);
+    size_t index = creator->AddContext(vm_context);
     CHECK_EQ(index, SnapshotData::kNodeVMContextIndex);
-    index = creator.AddContext(base_context);
+    index = creator->AddContext(base_context);
     CHECK_EQ(index, SnapshotData::kNodeBaseContextIndex);
-    index = creator.AddContext(main_context,
+    index = creator->AddContext(main_context,
                                 {SerializeNodeContextInternalFields, env});
     CHECK_EQ(index, SnapshotData::kNodeMainContextIndex);
   }
 
   // Must be out of HandleScope
   out->v8_snapshot_blob_data =
-      creator.CreateBlob(SnapshotCreator::FunctionCodeHandling::kKeep);
+      creator->CreateBlob(SnapshotCreator::FunctionCodeHandling::kKeep);
 
   // We must be able to rehash the blob when we restore it or otherwise
   // the hash seed would be fixed by V8, introducing a vulnerability.
@@ -1468,6 +1445,23 @@ void SerializeSnapshotableObjects(Realm* realm,
 
 namespace mksnapshot {
 
+void GetEmbedderEntryFunction(const FunctionCallbackInfo<Value>& args) {
+  Environment* env = Environment::GetCurrent(args);
+  Isolate* isolate = env->isolate();
+  if (!env->embedder_mksnapshot_entry_point()) return;
+  MaybeLocal<Function> jsfn =
+      Function::New(isolate->GetCurrentContext(),
+                    [](const FunctionCallbackInfo<Value>& args) {
+                      Environment* env = Environment::GetCurrent(args);
+                      Local<Value> require_fn = args[0];
+                      CHECK(require_fn->IsFunction());
+                      CHECK(env->embedder_mksnapshot_entry_point());
+                      env->embedder_mksnapshot_entry_point()(
+                          {env->process_object(), require_fn.As<Function>()});
+                    });
+  if (!jsfn.IsEmpty()) args.GetReturnValue().Set(jsfn.ToLocalChecked());
+}
+
 void CompileSerializeMain(const FunctionCallbackInfo<Value>& args) {
   CHECK(args[0]->IsString());
   Local<String> filename = args[0].As<String>();
@@ -1521,6 +1515,8 @@ void Initialize(Local<Object> target,
                 Local<Value> unused,
                 Local<Context> context,
                 void* priv) {
+  SetMethod(
+      context, target, "getEmbedderEntryFunction", GetEmbedderEntryFunction);
   SetMethod(context, target, "compileSerializeMain", CompileSerializeMain);
   SetMethod(context, target, "setSerializeCallback", SetSerializeCallback);
   SetMethod(context, target, "setDeserializeCallback", SetDeserializeCallback);
@@ -1531,6 +1527,7 @@ void Initialize(Local<Object> target,
 }
 
 void RegisterExternalReferences(ExternalReferenceRegistry* registry) {
+  registry->Register(GetEmbedderEntryFunction);
   registry->Register(CompileSerializeMain);
   registry->Register(SetSerializeCallback);
   registry->Register(SetDeserializeCallback);

test/embedding/embedtest.cc

@@ -1,3 +1,6 @@
+#ifdef NDEBUG
+#undef NDEBUG
+#endif
 #include "node.h"
 #include "uv.h"
 #include <assert.h>
@@ -58,9 +61,12 @@ int RunNodeInstance(MultiIsolatePlatform* platform,
   int exit_code = 0;
 
   node::EmbedderSnapshotData::Pointer snapshot;
+  auto snapshot_build_mode_it =
+      std::find(args.begin(), args.end(), "--embedder-snapshot-create");
   auto snapshot_arg_it =
       std::find(args.begin(), args.end(), "--embedder-snapshot-blob");
-  if (snapshot_arg_it < args.end() - 1) {
+  if (snapshot_arg_it < args.end() - 1 &&
+      snapshot_build_mode_it == args.end()) {
     FILE* fp = fopen((snapshot_arg_it + 1)->c_str(), "r");
     assert(fp != nullptr);
     snapshot = node::EmbedderSnapshotData::FromFile(fp);
@@ -69,9 +75,11 @@ int RunNodeInstance(MultiIsolatePlatform* platform,
 
   std::vector<std::string> errors;
   std::unique_ptr<CommonEnvironmentSetup> setup =
-      snapshot
+      snapshot ? CommonEnvironmentSetup::CreateFromSnapshot(
-          ? CommonEnvironmentSetup::CreateWithSnapshot(
                      platform, &errors, snapshot.get(), args, exec_args)
+      : snapshot_build_mode_it != args.end()
+          ? CommonEnvironmentSetup::CreateForSnapshotting(
+                platform, &errors, args, exec_args)
           : CommonEnvironmentSetup::Create(platform, &errors, args, exec_args);
   if (!setup) {
     for (const std::string& err : errors)
@@ -94,9 +102,12 @@ int RunNodeInstance(MultiIsolatePlatform* platform,
     } else {
       loadenv_ret = node::LoadEnvironment(
           env,
-          "const publicRequire ="
+          // Snapshots do not support userland require()s (yet)
+          "if (!require('v8').startupSnapshot.isBuildingSnapshot()) {"
+          "  const publicRequire ="
           "    require('module').createRequire(process.cwd() + '/');"
-          "globalThis.require = publicRequire;"
+          "  globalThis.require = publicRequire;"
+          "} else globalThis.require = require;"
           "globalThis.embedVars = { nön_ascıı: '🏳️‍🌈' };"
           "require('vm').runInThisContext(process.argv[1]);");
     }
@@ -105,9 +116,20 @@ int RunNodeInstance(MultiIsolatePlatform* platform,
       return 1;
 
     exit_code = node::SpinEventLoop(env).FromMaybe(1);
+  }
+
+  if (snapshot_arg_it < args.end() - 1 &&
+      snapshot_build_mode_it != args.end()) {
+    snapshot = setup->CreateSnapshot();
+    assert(snapshot);
+
+    FILE* fp = fopen((snapshot_arg_it + 1)->c_str(), "w");
+    assert(fp != nullptr);
+    snapshot->ToFile(fp);
+    fclose(fp);
+  }
 
   node::Stop(env);
-  }
 
   return exit_code;
 }

test/embedding/test-embedding.js

@@ -20,7 +20,6 @@ function resolveBuiltBinary(bin) {
 }
 
 const binary = resolveBuiltBinary('embedtest');
-const standaloneNodeBinary = resolveBuiltBinary('node');
 
 assert.strictEqual(
   child_process.spawnSync(binary, ['console.log(42)'])
@@ -51,22 +50,30 @@ assert.strictEqual(
   child_process.spawnSync(binary, [`require(${fixturePath})`, 92]).status,
   92);
 
+function getReadFileCodeForPath(path) {
+  return `(require("fs").readFileSync(${JSON.stringify(path)}, "utf8"))`;
+}
+
 // Basic snapshot support
 {
+  // readSync + eval since snapshots don't support userland require() (yet)
   const snapshotFixture = fixtures.path('snapshot', 'echo-args.js');
   const blobPath = path.join(tmpdir.path, 'embedder-snapshot.blob');
-  const buildSnapshotArgs = [snapshotFixture, 'arg1', 'arg2'];
+  const buildSnapshotArgs = [
+    `eval(${getReadFileCodeForPath(snapshotFixture)})`, 'arg1', 'arg2',
+    '--embedder-snapshot-blob', blobPath, '--embedder-snapshot-create',
+  ];
   const runEmbeddedArgs = ['--embedder-snapshot-blob', blobPath, 'arg3', 'arg4'];
 
   fs.rmSync(blobPath, { force: true });
-  assert.strictEqual(child_process.spawnSync(standaloneNodeBinary, [
+  assert.strictEqual(child_process.spawnSync(binary, [
-    '--snapshot-blob', blobPath, '--build-snapshot', ...buildSnapshotArgs,
+    '--', ...buildSnapshotArgs,
   ], {
     cwd: tmpdir.path,
   }).status, 0);
   const spawnResult = child_process.spawnSync(binary, ['--', ...runEmbeddedArgs]);
   assert.deepStrictEqual(JSON.parse(spawnResult.stdout), {
-    originalArgv: [standaloneNodeBinary, ...buildSnapshotArgs],
+    originalArgv: [binary, ...buildSnapshotArgs],
     currentArgv: [binary, ...runEmbeddedArgs],
   });
 }
@@ -75,10 +82,14 @@ assert.strictEqual(
 {
   const snapshotFixture = fixtures.path('snapshot', 'create-worker-and-vm.js');
   const blobPath = path.join(tmpdir.path, 'embedder-snapshot.blob');
+  const buildSnapshotArgs = [
+    `eval(${getReadFileCodeForPath(snapshotFixture)})`,
+    '--embedder-snapshot-blob', blobPath, '--embedder-snapshot-create',
+  ];
 
   fs.rmSync(blobPath, { force: true });
-  assert.strictEqual(child_process.spawnSync(standaloneNodeBinary, [
+  assert.strictEqual(child_process.spawnSync(binary, [
-    '--snapshot-blob', blobPath, '--build-snapshot', snapshotFixture,
+    '--', ...buildSnapshotArgs,
   ], {
     cwd: tmpdir.path,
   }).status, 0);