Make V8 string to NSString conversions UTF-16 faithful

NativeScript/runtime/ArgConverter.mm

@@ -294,7 +294,8 @@
   } else if (value->IsString()) {
     if (type == BinaryTypeEncodingType::IdEncoding ||
         type == BinaryTypeEncodingType::InterfaceDeclarationReference) {
-      id data = tns::ToNSString(isolate, value);
+      std::u16string strValue = tns::ToUtf16String(isolate, value);
+      id data = [NSString stringWithCharacters:(const unichar*)strValue.data() length:strValue.size()];
       // this feels wrong but follows the other CFBridgingRetain calls
       // and also solves a leak
       auto ref = CFBridgingRetain(data);
@@ -929,8 +930,8 @@
   }
 
   if ([obj isKindOfClass:[NSString class]]) {
-    const char* str = [obj UTF8String];
-    args.GetReturnValue().Set(tns::ToV8String(isolate, str));
+    NSString* nativeStr = (NSString*)obj;
+    args.GetReturnValue().Set(tns::ToV8String(isolate, nativeStr));
     return;
   }
 

NativeScript/runtime/DictionaryAdapter.mm

@@ -48,7 +48,8 @@ - (id)nextObject {
         bool success = array->Get(context, self->index_).ToLocal(&key);
         tns::Assert(success, isolate);
         self->index_ += 2;
-        NSString* result = tns::ToNSString(isolate, key);
+        std::u16string keyStr = tns::ToUtf16String(isolate, key);
+        NSString* result = [NSString stringWithCharacters:(const unichar*)keyStr.data() length:keyStr.length()];
         return result;
     }
 
@@ -116,8 +117,8 @@ - (id)nextObject {
         bool success = properties->Get(context, (uint)self->index_).ToLocal(&value);
         tns::Assert(success, isolate);
         self->index_++;
-        std::string result = tns::ToString(isolate, value);
-        return [NSString stringWithUTF8String:result.c_str()];
+        std::u16string result = tns::ToUtf16String(isolate, value);
+        return [NSString stringWithCharacters:(const unichar*)result.data() length:result.size()];
     }
 
     return nil;
@@ -139,8 +140,8 @@ - (NSArray*)allObjects {
         Local<Value> value;
         bool success = properties->Get(context, i).ToLocal(&value);
         tns::Assert(success, isolate);
-        std::string result = tns::ToString(isolate, value);
-        [array addObject:[NSString stringWithUTF8String:result.c_str()]];
+        std::u16string result = tns::ToUtf16String(isolate, value);
+        [array addObject:[NSString stringWithCharacters:(const unichar*)result.data() length:result.size()]];
     }
 
     return array;
@@ -214,7 +215,7 @@ - (id)objectForKey:(id)aKey {
         bool success = obj->Get(context, key).ToLocal(&value);
         tns::Assert(success, isolate);
     } else if ([aKey isKindOfClass:[NSString class]]) {
-        const char* key = [aKey UTF8String];
+        NSString* key = (NSString*)aKey;
         Local<v8::String> keyV8Str = tns::ToV8String(isolate, key);
 
         if (obj->IsMap()) {

NativeScript/runtime/Helpers.mm

@@ -24,13 +24,26 @@
 }  // namespace
 
 std::u16string tns::ToUtf16String(Isolate* isolate, const Local<Value>& value) {
-  std::string valueStr = tns::ToString(isolate, value);
-#pragma GCC diagnostic ignored "-Wdeprecated-declarations"
-  // FIXME: std::codecvt_utf8_utf16 is deprecated
-  std::wstring_convert<std::codecvt_utf8_utf16<char16_t>, char16_t> convert;
-  std::u16string value16 = convert.from_bytes(valueStr);
+  // Read the V8 string's native UTF-16 buffer directly instead of round-tripping
+  // through UTF-8, which corrupts lone surrogates (replaced with U+FFFD) and is
+  // slower. This also drops the deprecated std::codecvt_utf8_utf16.
+  if (value.IsEmpty()) {
+    return std::u16string();
+  }
+
+  if (value->IsStringObject()) {
+    Local<v8::String> obj = value.As<StringObject>()->ValueOf();
+    return tns::ToUtf16String(isolate, obj);
+  }
+
+  v8::String::Value result(isolate, value);
+
+  uint16_t* val = *result;
+  if (val == nullptr) {
+    return std::u16string();
+  }
 
-  return value16;
+  return std::u16string((char16_t*)val, result.length());
 }
 
 std::vector<uint16_t> tns::ToVector(const std::string& value) {

NativeScript/runtime/Interop.mm

@@ -324,7 +324,8 @@ inline bool isBool() {
   } else if (argHelper.isString() &&
              (typeEncoding->type == BinaryTypeEncodingType::InterfaceDeclarationReference ||
               typeEncoding->type == BinaryTypeEncodingType::IdEncoding)) {
-    NSString* result = tns::ToNSString(isolate, arg);
+    std::u16string str = tns::ToUtf16String(isolate, arg);
+    NSString* result = [NSString stringWithCharacters:(const unichar*)str.data() length:str.size()];
     Interop::SetValue(dest, result);
   } else if (Interop::IsNumbericType(typeEncoding->type) || tns::IsNumber(arg)) {
     double value = tns::ToNumber(isolate, arg);
@@ -686,7 +687,8 @@ inline bool isBool() {
   if (arg.IsEmpty() || arg->IsNullOrUndefined()) {
     return nil;
   } else if (tns::IsString(arg)) {
-    NSString* result = tns::ToNSString(isolate, arg);
+    std::u16string value = tns::ToUtf16String(isolate, arg);
+    NSString* result = [NSString stringWithCharacters:(const unichar*)value.data() length:value.size()];
     return result;
   } else if (tns::IsNumber(arg)) {
     double value = tns::ToNumber(isolate, arg);

TestRunner/app/tests/ApiTests.js

@@ -12,6 +12,39 @@ describe(module.id, function () {
         expect(object.hash).toBe(3);
     });
 
+    it("preserves a lone high surrogate when bridging a JS string to NSString", function () {
+        // A lone high surrogate (U+D834, range U+D800-U+DBFF) is a valid JS string
+        // code unit but has no UTF-8 encoding. The old UTF-8 round-trip replaced it
+        // with U+FFFD; faithful UTF-16 bridging keeps it. Read the code unit straight
+        // out of the bridged string's UTF-16 buffer as a number: reading it back as a
+        // JS string would re-corrupt a lone surrogate, and converting it to UTF-8 to
+        // measure it is not reliable across OS versions.
+        var ns = NSString.stringWithString("\uD834");
+        expect(ns.length).toBe(1);
+
+        var buffer = interop.alloc(interop.sizeof(interop.types.uint16));
+        ns.getCharactersRange(buffer, NSMakeRange(0, 1));
+        var codeUnit = new interop.Reference(interop.types.uint16, buffer).value;
+        interop.free(buffer);
+
+        expect(codeUnit).toBe(0xD834); // 0xFFFD (65533) after a lossy UTF-8 round-trip
+    });
+
+    it("preserves a lone low surrogate when bridging a JS string to NSString", function () {
+        // The low surrogate range (U+DC00-U+DFFF) is a different bit pattern that also
+        // has no UTF-8 encoding and must survive the bridge intact; observed the same
+        // way as the high-surrogate case above.
+        var ns = NSString.stringWithString("\uDC00");
+        expect(ns.length).toBe(1);
+
+        var buffer = interop.alloc(interop.sizeof(interop.types.uint16));
+        ns.getCharactersRange(buffer, NSMakeRange(0, 1));
+        var codeUnit = new interop.Reference(interop.types.uint16, buffer).value;
+        interop.free(buffer);
+
+        expect(codeUnit).toBe(0xDC00); // 0xFFFD (65533) after a lossy UTF-8 round-trip
+    });
+
     it("NSArray from native (uncached) array access", function () {
         const res = TNSObjCTypes.new().getNSArrayOfNSURLs();
         console.log(res);