PaymentRequest Mojo bindings.

mojo/public/cpp/bindings/wtf_array.h

Index: mojo/public/cpp/bindings/wtf_array.h
diff --git a/mojo/public/cpp/bindings/wtf_array.h b/mojo/public/cpp/bindings/wtf_array.h
index b5d1a30c36304ace290d60d6735919ec408083fd..d693ad9e1eed684bd35351640c713fef14c6f9eb 100644
--- a/mojo/public/cpp/bindings/wtf_array.h
+++ b/mojo/public/cpp/bindings/wtf_array.h
@@ -16,6 +16,10 @@
 #include "mojo/public/cpp/bindings/type_converter.h"
 #include "third_party/WebKit/Source/wtf/Vector.h"
 
+namespace blink {
+class HeapAllocator;
+}
+
 namespace mojo {
 
 // Represents an array backed by WTF::Vector. Comparing with WTF::Vector,
@@ -230,6 +234,21 @@ class WTFArray {
   bool is_null_;
 };
 
+// A |TypeConverter| that will create a |WTFArray<T>| containing a copy of the
+// contents of a |WTF::Vector<E, 0, blink::HeapAllocator>|, using
+// |TypeConverter<T, E>| to copy each element. The returned array will always be
+// non-null.
+template <typename T, typename E>
+struct TypeConverter<WTFArray<T>, WTF::Vector<E, 0, blink::HeapAllocator>> {

[yzshen1, 2016/03/29 22:49:31]

Is it a common operation to use a non-default allocator? It is unfortunate
because it requires a deep copy. :/

Can this converter be generalized to have the allocator as a template parameter
as well? That way we don't have to have blink::HeapAllocator here which seems a
little weird?

An alternative is to consider putting it somewhere inside blink, because mojo
bindings itself doesn't need this converter, putting it inside blink reduces
what the bindings has to know about blink.

WDYT? Thanks!

[please use gerrit instead, 2016/03/29 23:08:11]

The blink::HeapAllocator is used in blink::HeapVector<E>, which is a subclass of
WTF::Vector<E, 0, blink::HeapAllocator>. That's the type used by code generated
from WebIDL. These objects represent their JavaScript counterparts. See:

https://code.google.com/p/chromium/codesearch#chromium/src/out/Debug/gen/blin...

...for an example of a blink::HeapVector<E> generated from:

https://code.google.com/p/chromium/codesearch#chromium/src/third_party/WebKit...
Why does this require a deep copy?
Unfortunately, "struct TypeConverter" needs to be a template of exactly two
typenames. See:

https://code.google.com/p/chromium/codesearch#chromium/src/mojo/public/cpp/bi...
I'm OK with that.

esprehn@: where would be a good place for such a common code?

[yzshen1, 2016/03/29 23:20:15]

I mean, this TypeConverter is a deep copy.
The TypeConverter takes exactly two template parameters, but I think you can do:

template <typename T, typename E, typename Allocator>
struct TypeConverter<WTFArray<T>, WTF::Vector<E, 0, Allocator>> {
};

+  static WTFArray<T> Convert(
+      const WTF::Vector<E, 0, blink::HeapAllocator>& input) {
+    WTFArray<T> result(input.size());
+    for (size_t i = 0; i < input.size(); ++i)
+      result[i] = TypeConverter<T, E>::Convert(input[i]);
+    return std::move(result);
+  }
+};
+
 }  // namespace mojo
 
 #endif  // MOJO_PUBLIC_CPP_BINDINGS_WTF_ARRAY_H_