[emval] Support dynamically-typed conversion to/from val

stbergmann · stbergmann · commit 5c7a48a31d26 · 2025-07-29T22:08:21.000+02:00
...for cases where a dynamically given user-defined type is known to be
ABI-compatible with a statically given wire type (e.g., an enum type being
compatible with its underlying type, or a struct type being compatible with
`const void*`; for the latter, a returned `const void*` needs to be transformed
somehow into the desired dynamic type before destructors are run and the
pointed-to object is destroyed).
diff --git a/src/lib/libemval.js b/src/lib/libemval.js
@@ -248,23 +248,33 @@ var LibraryEmVal = {
 
   // Leave id 0 undefined.  It's not a big deal, but might be confusing
   // to have null be a valid method caller.
-  $emval_methodCallers: [undefined],
+  $emval_methodCallers: {list: [undefined], lookup: {}},
 
   $emval_addMethodCaller__deps: ['$emval_methodCallers'],
-  $emval_addMethodCaller: (caller) => {
-    var id = emval_methodCallers.length;
-    emval_methodCallers.push(caller);
+  $emval_addMethodCaller: (sig, caller) => {
+    var id = emval_methodCallers.list.length;
+    emval_methodCallers.list.push(caller);
+    emval_methodCallers.lookup[sig] = id;
     return id;
   },
 
   _emval_create_invoker__deps: [
-    '$emval_addMethodCaller', '$emval_lookupTypes',
+    '$emval_methodCallers', '$emval_addMethodCaller', '$emval_lookupTypes',
     '$createNamedFunction', '$emval_returnValue',
     '$Emval', '$getStringOrSymbol',
   ],
   _emval_create_invoker: (argCount, argTypesPtr, kind) => {
     var GenericWireTypeSize = {{{ 2 * POINTER_SIZE }}};
 
+    var sig = kind.toString();
+    for (var i = 0; i < argCount; ++i) {
+      sig += ',' + {{{ makeGetValue('argTypesPtr', `i*${POINTER_SIZE}`, '*') }}};
+    }
+    var id = emval_methodCallers.lookup[sig];
+    if (id) {
+      return id;
+    }
+
     var [retType, ...argTypes] = emval_lookupTypes(argCount, argTypesPtr);
     var toReturnWire = retType.toWireType.bind(retType);
     var argFromPtr = argTypes.map(type => type.readValueFromPointer.bind(type));
@@ -332,12 +342,12 @@ ${functionBody}
     var invokerFunction = new Function(Object.keys(captures), functionBody)(...Object.values(captures));
 #endif
     var functionName = `methodCaller<(${argTypes.map(t => t.name)}) => ${retType.name}>`;
-    return emval_addMethodCaller(createNamedFunction(functionName, invokerFunction));
+    return emval_addMethodCaller(sig, createNamedFunction(functionName, invokerFunction));
   },
 
   _emval_invoke__deps: ['$getStringOrSymbol', '$emval_methodCallers', '$Emval'],
   _emval_invoke: (caller, handle, methodName, destructorsRef, args) => {
-    return emval_methodCallers[caller](handle, methodName, destructorsRef, args);
+    return emval_methodCallers.list[caller](handle, methodName, destructorsRef, args);
   },
 
   // Same as `_emval_invoke`, just imported into Wasm under a different return type.
diff --git a/system/include/emscripten/val.h b/system/include/emscripten/val.h
@@ -18,10 +18,12 @@
 #include <cstdint> // uintptr_t
 #include <vector>
 #include <type_traits>
+#include <typeinfo>
 #include <pthread.h>
 #if __cplusplus >= 202002L
 #include <coroutine>
 #include <exception>
+#include <functional>
 #include <variant>
 #endif
 
@@ -256,6 +258,17 @@ struct WireTypePack {
   std::array<GenericWireType, PackSize<Args...>::value> elements;
 };
 
+#if __cplusplus >= 202002L
+using identity = std::identity;
+#else
+struct identity {
+  template<typename T>
+  constexpr T&& operator()(T&& t) const noexcept {
+    return std::forward<T>(t);
+  }
+};
+#endif
+
 } // end namespace internal
 
 #define EMSCRIPTEN_SYMBOL(name)                                         \
@@ -341,6 +354,14 @@ class EMBIND_VISIBILITY_DEFAULT val {
     return val(internal::_emval_get_module_property(name));
   }
 
+#if __has_feature(cxx_rtti)
+  // Create a `val` from a user-defined dynamic `type` that is ABI-compatible with static type `T`:
+  template<typename T>
+  static val dyn(const std::type_info& type, T&& value) {
+    return internalDynCast<val>(internal::TypeID<val>::get(), &type, nullptr, std::forward<T>(value), internal::identity());
+  }
+#endif
+
   template<typename T, typename... Policies>
   explicit val(T&& value, Policies...) {
     using namespace internal;
@@ -516,6 +537,14 @@ class EMBIND_VISIBILITY_DEFAULT val {
     return internalCall<EM_INVOKER_KIND::CAST, WithPolicies<Policies...>, T>(as_handle(), nullptr, *this);
   }
 
+#if __has_feature(cxx_rtti)
+  // Retrieve a user-defined dynamic `type` that is ABI-compatible with static type `T`:
+  template<typename T, typename TransformFn>
+  auto dyn_as(const std::type_info& type, TransformFn transform) const {
+    return internalDynCast<T>(&type, internal::TypeID<val>::get(), as_handle(), *this, transform);
+  }
+#endif
+
 // Prefer calling val::typeOf() over val::typeof(), since this form works in both C++11 and GNU++11 build modes. "typeof" is a reserved word in GNU++11 extensions.
   val typeOf() const {
     return val(internal::_emval_typeof(as_handle()));
@@ -573,42 +602,68 @@ class EMBIND_VISIBILITY_DEFAULT val {
   template<typename WrapperType>
   friend val internal::wrapped_extend(const std::string& , const val& );
 
-  template<internal::EM_INVOKER_KIND Kind, typename Policy, typename Ret, typename... Args>
-  static Ret internalCall(EM_VAL handle, const char *methodName, Args&&... args) {
-    static_assert(!std::is_lvalue_reference<Ret>::value,
-                  "Cannot create a lvalue reference out of a JS value.");
-
+  template<typename Ret, typename... Args>
+  static auto internalDoCall(internal::EM_INVOKER caller, EM_VAL handle, const char* methodName, internal::EM_DESTRUCTORS* destructors, Args&&... args) {
     using namespace internal;
 
     using RetWire = BindingType<Ret>::WireType;
 
-    static constexpr typename Policy::template ArgTypeList<Ret, Args...> argTypes;
-    thread_local EM_INVOKER mc = _emval_create_invoker(argTypes.getCount(), argTypes.getTypes(), Kind);
-
     WireTypePack<Args...> argv(std::forward<Args>(args)...);
-    EM_DESTRUCTORS destructors = nullptr;
 
     RetWire result;
     if constexpr (std::is_integral<RetWire>::value && sizeof(RetWire) == 8) {
       // 64-bit integers can't go through "generic wire type" because double and int64 have different ABI.
       result = static_cast<RetWire>(_emval_invoke_i64(
-        mc,
+        caller,
         handle,
         methodName,
-        &destructors,
+        destructors,
         argv));
     } else {
       result = GenericWireTypeConverter<RetWire>::from(_emval_invoke(
-        mc,
+        caller,
         handle,
         methodName,
-        &destructors,
+        destructors,
         argv));
     }
+    return result;
+  }
+
+  template<internal::EM_INVOKER_KIND Kind, typename Policy, typename Ret, typename... Args>
+  static Ret internalCall(EM_VAL handle, const char *methodName, Args&&... args) {
+    static_assert(!std::is_lvalue_reference<Ret>::value,
+                  "Cannot create a lvalue reference out of a JS value.");
+
+    using namespace internal;
+
+    static constexpr typename Policy::template ArgTypeList<Ret, Args...> argTypes;
+    thread_local EM_INVOKER mc = _emval_create_invoker(argTypes.getCount(), argTypes.getTypes(), Kind);
+
+    EM_DESTRUCTORS destructors = nullptr;
+
+    auto result = internalDoCall<Ret>(mc, handle, methodName, &destructors, std::forward<Args>(args)...);
     DestructorsRunner rd(destructors);
     return BindingType<Ret>::fromWireType(result);
   }
 
+  template<typename Ret, typename Arg, typename TransformFn>
+  static auto internalDynCast(internal::TYPEID retType, internal::TYPEID argType, EM_VAL handle, Arg&& arg, TransformFn transform) {
+    static_assert(!std::is_lvalue_reference<Ret>::value,
+                  "Cannot create a lvalue reference out of a JS value.");
+
+    using namespace internal;
+
+    TYPEID const argTypes[2] { retType, argType };
+    EM_INVOKER mc = _emval_create_invoker(2, argTypes, EM_INVOKER_KIND::CAST);
+
+    EM_DESTRUCTORS destructors = nullptr;
+
+    auto result = internalDoCall<Ret>(mc, handle, nullptr, &destructors, std::forward<Arg>(arg));
+    DestructorsRunner rd(destructors);
+    return transform(BindingType<Ret>::fromWireType(result));
+  }
+
   template<typename T>
   val val_ref(const T& v) const {
     return val(v);
diff --git a/test/embind/test_val_dyn.cpp b/test/embind/test_val_dyn.cpp
@@ -0,0 +1,189 @@
+#include <cassert>
+#include <cstddef>
+#include <cstring>
+#include <functional>
+#include <stdexcept>
+#include <string>
+#include <typeinfo>
+#include <utility>
+
+#include <emscripten.h>
+#include <emscripten/bind.h>
+
+namespace {
+
+#if __cplusplus >= 202002L
+using identity = std::identity;
+#else
+struct identity {
+  template<typename T>
+  constexpr T&& operator()(T&& t) const noexcept { return std::forward<T>(t); }
+};
+#endif
+
+enum class Enum1: int { E1 = 1 };
+
+enum class Enum2: int { E2 = 2 };
+
+struct Struct1 { int n; };
+
+struct Struct2 { double d; };
+
+class Any {
+public:
+  enum Kind { KIND_ENUM, KIND_STRUCT };
+
+  Any(std::type_info const& type, int enum_value):
+    kind_(KIND_ENUM), type_(&type), enum_value_(enum_value) {}
+
+  Any(std::type_info const& type, void const* struct_pointer):
+    kind_(KIND_STRUCT), type_(&type), struct_pointer_(copyStruct(type, struct_pointer)) {}
+
+  Any(Any const& other): kind_(other.kind_), type_(other.type_) {
+    switch (kind_) {
+    case KIND_ENUM:
+      enum_value_ = other.enum_value_;
+      break;
+    case KIND_STRUCT:
+      struct_pointer_ = copyStruct(*other.type_, other.struct_pointer_);
+      break;
+    }
+  }
+
+  ~Any() { release(); }
+
+  Any& operator=(Any const& other) {
+    if (&other != this) {
+      release();
+      kind_ = other.kind_;
+      type_ = other.type_;
+      switch (kind_) {
+      case KIND_ENUM:
+        enum_value_ = other.enum_value_;
+        break;
+      case KIND_STRUCT:
+        struct_pointer_ = copyStruct(*other.type_, other.struct_pointer_);
+        break;
+      }
+    }
+    return *this;
+  }
+
+  template<typename T> T get() const;
+
+  template<> Enum1 get() const {
+    verify(KIND_ENUM, typeid(Enum1));
+    return static_cast<Enum1>(enum_value_);
+  }
+
+  template<> Enum2 get() const {
+    verify(KIND_ENUM, typeid(Enum2));
+    return static_cast<Enum2>(enum_value_);
+  }
+
+  template<> Struct1 get() const {
+    verify(KIND_STRUCT, typeid(Struct1));
+    return *static_cast<Struct1 const*>(struct_pointer_);
+  }
+
+  template<> Struct2 get() const {
+    verify(KIND_STRUCT, typeid(Struct2));
+    return *static_cast<Struct2 const*>(struct_pointer_);
+  }
+
+  static Any jsNew(std::string id, emscripten::val const& value) {
+    Kind kind;
+    std::type_info const* type;
+    if (id == "Enum1") {
+      kind = KIND_ENUM;
+      type = &typeid(Enum1);
+    } else if (id == "Enum2") {
+      kind = KIND_ENUM;
+      type = &typeid(Enum2);
+    } else if (id == "Struct1") {
+      kind = KIND_STRUCT;
+      type = &typeid(Struct1);
+    } else if (id == "Struct2") {
+      kind = KIND_STRUCT;
+      type = &typeid(Struct2);
+    } else {
+      throw std::runtime_error("unknown type ID " + id);
+    }
+    switch (kind) {
+    case KIND_ENUM:
+      return Any(*type, value.dyn_as<int>(*type, identity()));
+    case KIND_STRUCT:
+      return value.dyn_as<void const*>(*type, [&type](void const* p) { return Any(*type, p); });
+    }
+    assert(false);
+  }
+
+  emscripten::val jsGet() {
+    switch (kind_) {
+    case Any::KIND_ENUM:
+      return emscripten::val::dyn(*type_, enum_value_);
+    case Any::KIND_STRUCT:
+      return emscripten::val::dyn(*type_, copyStruct(*type_, struct_pointer_));
+    }
+    assert(false);
+  }
+
+private:
+  static void const* copyStruct(std::type_info const& type, void const* struct_pointer) {
+    std::size_t n;
+    if (type == typeid(Struct1)) {
+      n = sizeof(Struct1);
+    } else if (type == typeid(Struct2)) {
+      n = sizeof(Struct2);
+    } else {
+      assert(false);
+    }
+    return std::memcpy(new unsigned char[n], struct_pointer, n);
+  }
+
+  void verify(Kind kind, std::type_info const& type) const {
+    assert(kind_ == kind);
+    assert(*type_ == type);
+  }
+
+  void release() {
+    if (kind_ == KIND_STRUCT) {
+      delete[] static_cast<unsigned char const*>(struct_pointer_);
+    }
+  }
+
+  Kind kind_;
+  std::type_info const* type_;
+  union {
+    int enum_value_;
+    void const* struct_pointer_;
+  };
+};
+
+}
+
+int main() {
+  emscripten::enum_<Enum1>("Enum1").value("E1", Enum1::E1);
+  emscripten::enum_<Enum2>("Enum2").value("E2", Enum2::E2);
+  emscripten::value_object<Struct1>("Struct1").field("n", &Struct1::n);
+  emscripten::value_object<Struct2>("Struct2").field("d", &Struct2::d);
+  emscripten::class_<Any>("Any").constructor(Any::jsNew).function("get", &Any::jsGet);
+  EM_ASM(
+    globalThis.AnyEnum1 = new Module.Any("Enum1", Module.Enum1.E1);
+    globalThis.AnyEnum2 = new Module.Any("Enum2", Module.Enum2.E2);
+    globalThis.AnyStruct1 = new Module.Any("Struct1", {n: 1});
+    globalThis.AnyStruct2 = new Module.Any("Struct2", {d: 1.5});
+    globalThis.testGetEnum1 = () => globalThis.AnyEnum1.get() === Module.Enum1.E1;
+    globalThis.testGetEnum2 = () => globalThis.AnyEnum2.get() === Module.Enum2.E2;
+    globalThis.testGetStruct1 = () => globalThis.AnyStruct1.get().n === 1;
+    globalThis.testGetStruct2 = () => globalThis.AnyStruct2.get().d === 1.5;
+  );
+  assert(emscripten::val::global("AnyEnum1").as<Any>().get<Enum1>() == Enum1::E1);
+  assert(emscripten::val::global("AnyEnum2").as<Any>().get<Enum2>() == Enum2::E2);
+  assert(emscripten::val::global("AnyStruct1").as<Any>().get<Struct1>().n == 1);
+  assert(emscripten::val::global("AnyStruct2").as<Any>().get<Struct2>().d == 1.5);
+  assert(emscripten::val::global().call<bool>("testGetEnum1"));
+  assert(emscripten::val::global().call<bool>("testGetEnum2"));
+  assert(emscripten::val::global().call<bool>("testGetStruct1"));
+  assert(emscripten::val::global().call<bool>("testGetStruct2"));
+}
diff --git a/test/test_core.py b/test/test_core.py
@@ -7636,6 +7636,9 @@ def test_embind_val_coro_propogate_js_error(self):
     self.cflags += ['-std=c++20', '--bind', '--pre-js=pre.js', '-fexceptions']
     self.do_runf('embind/test_val_coro.cpp', 'rejected with: bang from JS promise!\n')
 
+  def test_embind_val_dyn(self):
+    self.do_runf('embind/test_val_dyn.cpp', cflags=['-lembind'])
+
   def test_embind_dynamic_initialization(self):
     self.cflags += ['-lembind']
     self.do_run_in_out_file_test('embind/test_dynamic_initialization.cpp')
