Reject PaymentResponse.complete() promise on error.

third_party/WebKit/Source/modules/payments/PaymentRequest.cpp

Index: third_party/WebKit/Source/modules/payments/PaymentRequest.cpp
diff --git a/third_party/WebKit/Source/modules/payments/PaymentRequest.cpp b/third_party/WebKit/Source/modules/payments/PaymentRequest.cpp
index 8e6e3691072eb920671ba1936f791eba2e4ad105..5660628384c046f4e1bb6780fb3a0699765bc16a 100644
--- a/third_party/WebKit/Source/modules/payments/PaymentRequest.cpp
+++ b/third_party/WebKit/Source/modules/payments/PaymentRequest.cpp
@@ -325,6 +325,10 @@ ScriptPromise PaymentRequest::complete(ScriptState* scriptState, PaymentComplete
     if (m_completeResolver)
         return ScriptPromise::rejectWithDOMException(scriptState, DOMException::create(InvalidStateError, "Already called complete() once"));
 
+    // User has cancelled the transaction while the website was processing it.
+    if (!m_paymentProvider)
+        return ScriptPromise::rejectWithDOMException(scriptState, DOMException::create(InvalidStateError, "Request cancelled"));

[Marijn Kruisselbrink, 2016/06/21 20:45:50]

It seems a bit odd that this throws a different exception than what would be
thrown in the (only different by race conditions) situation where the mojo
connection just happens to have closed right after complete() was called (in
that case OnError throws a SyntaxError). But maybe it's that SyntaxError that
doesn't really make sense?

Also is there a corresponding spec bug for this case? Although I'm not quite
clear on how much of the UI behavior is actually part of the spec and how much
is just up to user agents.

[please use gerrit instead, 2016/06/21 21:03:01]

User cancelling the payment request does not appear anywhere in the spec. I will
bring it up with the editors. UI behavior is indeed up to user agents to
determine.

+
     // The payment provider should respond in PaymentRequest::OnComplete().
     m_paymentProvider->Complete(mojom::blink::PaymentComplete(result));