More informative error messages for non-Transferables.

Source/bindings/v8/V8Binding.h

 /*
 * Copyright (C) 2009 Google Inc. All rights reserved.
 * Copyright (C) 2012 Ericsson AB. All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions are
 * met:
 *
 *     * Redistributions of source code must retain the above copyright
 * notice, this list of conditions and the following disclaimer.
@@ skipping 57 matching lines @@
 
     // A helper for throwing JavaScript TypeError.
     v8::Handle<v8::Value> throwTypeError(v8::Isolate*);
     v8::Handle<v8::Value> throwTypeError(const String&, v8::Isolate*);
 
     // A helper for throwing JavaScript TypeError for not enough arguments.
     v8::Handle<v8::Value> throwNotEnoughArgumentsError(v8::Isolate*);
 
     v8::ArrayBuffer::Allocator* v8ArrayBufferAllocator();
 
-    v8::Handle<v8::Value> toV8Sequence(v8::Handle<v8::Value>, uint32_t& length, v8::Isolate*);
+    v8::Handle<v8::Value> toV8Sequence(v8::Handle<v8::Value>, uint32_t& length, v8::Isolate*, bool* notASequence = 0);

[Mike West, 2013/10/22 06:50:40]

I think I'd prefer to keep the Isolate as the last argument by creating another
version of this function. I don't know of anywhere else in the bindings code
where the Isolate isn't at the end, and I'd like to maintain that unless there's
a good reason not to.

That is, I'd prefer to see:

 v8::Handle<v8::Value> toV8Sequence(v8::Handle<v8::Value>, uint32_t& length,
v8::Isolate*);
 v8::Handle<v8::Value> toV8Sequence(v8::Handle<v8::Value>, uint32_t& length,
bool* notASequence, v8::Isolate*);

where the first just wraps the second with a temporary bool. If that overhead
causes perf regressions, that'd be a good reason to break convention, but I
don't think it will.

[sof, 2013/10/22 16:06:56]

Done, and dropped the "nicety" of having the extra argument be optional.
However, the argument ordering convention of having 'out' arguments appear last
isn't followed, but I assume that doesn't trump Isolate consistency in
this particular case.

(The previous version was patterned on the ordering that toRefPtrNativeArray()
uses.)

 
     inline v8::Handle<v8::Value> argumentOrNull(const v8::FunctionCallbackInfo<v8::Value>& args, int index)
     {
         return index >= args.Length() ? v8::Local<v8::Value>() : args[index];
     }
 
     // Since v8::Null(isolate) crashes if we pass a null isolate,
     // we need to use v8NullWithCheck(isolate) if an isolate can be null.
     //
     // FIXME: Remove all null isolates from V8 bindings, and remove v8NullWithCheck(isolate).
@@ skipping 414 matching lines @@
         result.reserveInitialCapacity(length);
         for (int i = startIndex; i < length; ++i)
             result.uncheckedAppend(TraitsType::nativeValue(args[i]));
         return result;
     }
 
     Vector<v8::Handle<v8::Value> > toVectorOfArguments(const v8::FunctionCallbackInfo<v8::Value>& args);
 
     // Validates that the passed object is a sequence type per WebIDL spec
     // http://www.w3.org/TR/2012/CR-WebIDL-20120419/#es-sequence
-    inline v8::Handle<v8::Value> toV8Sequence(v8::Handle<v8::Value> value, uint32_t& length, v8::Isolate* isolate)
+    inline v8::Handle<v8::Value> toV8Sequence(v8::Handle<v8::Value> value, uint32_t& length, v8::Isolate* isolate, bool* notASequence)
     {
         // Attempt converting to a sequence if the value is not already an array but is
         // any kind of object except for a native Date object or a native RegExp object.
         ASSERT(!value->IsArray());
         // FIXME: Do we really need to special case Date and RegExp object?
         // https://www.w3.org/Bugs/Public/show_bug.cgi?id=22806
         if (!value->IsObject() || value->IsDate() || value->IsRegExp()) {
-            throwTypeError(isolate);
+            // Signal that the caller must handle the type error.
+            if (notASequence)
+                *notASequence = true;
+            else
+                throwTypeError(isolate);

[Mike West, 2013/10/22 06:50:40]

I'd be happy to review a subsequent patch that added a nicer error message here,
and on line 542. :)

[sof, 2013/10/22 16:06:56]

Definitely, I can have a look at improving that.

             return v8Undefined();
         }
 
         v8::Local<v8::Value> v8Value(v8::Local<v8::Value>::New(isolate, value));
         v8::Local<v8::Object> object = v8::Local<v8::Object>::Cast(v8Value);
 
         // FIXME: The specification states that the length property should be used as fallback, if value
         // is not a platform object that supports indexed properties. If it supports indexed properties,
         // length should actually be one greater than value’s maximum indexed property index.
         V8TRYCATCH(v8::Local<v8::Value>, lengthValue, object->Get(v8::String::NewSymbol("length")));
 
         if (lengthValue->IsUndefined() || lengthValue->IsNull()) {
-            throwTypeError(isolate);
+            // Signal that the caller must handle the type error.
+            if (notASequence)
+                *notASequence = true;
+            else
+                throwTypeError(isolate);
             return v8Undefined();
         }
 
         V8TRYCATCH(uint32_t, sequenceLength, lengthValue->Int32Value());
         length = sequenceLength;
 
         return v8Value;
     }
 
     PassRefPtr<NodeFilter> toNodeFilter(v8::Handle<v8::Value>, v8::Isolate*);
@@ skipping 111 matching lines @@
 
     v8::Isolate* mainThreadIsolate();
     v8::Isolate* toIsolate(ExecutionContext*);
     v8::Isolate* toIsolate(Frame*);
 
     // Can only be called by WebKit::initialize
     void setMainThreadIsolate(v8::Isolate*);
 } // namespace WebCore
 
 #endif // V8Binding_h