EarlGrey tests for Payment Request (base CL)

ios/chrome/browser/payments/payment_request_manager.mm

Index: ios/chrome/browser/payments/payment_request_manager.mm
diff --git a/ios/chrome/browser/payments/payment_request_manager.mm b/ios/chrome/browser/payments/payment_request_manager.mm
index 657183145f2098c5333df7c8a8daa4518e4d0ad9..5cec27fa7e84b558a408cacdc86dff00d22d7992 100644
--- a/ios/chrome/browser/payments/payment_request_manager.mm
+++ b/ios/chrome/browser/payments/payment_request_manager.mm
@@ -15,6 +15,7 @@
 #include "components/autofill/core/browser/personal_data_manager.h"
 #include "ios/chrome/browser/autofill/personal_data_manager_factory.h"
 #include "ios/chrome/browser/browser_state/chrome_browser_state.h"
+#include "ios/chrome/browser/experimental_flags.h"
 #import "ios/chrome/browser/payments/js_payment_request_manager.h"
 #import "ios/chrome/browser/payments/payment_request_coordinator.h"
 #import "ios/chrome/browser/payments/payment_request_web_state_observer.h"
@@ -123,7 +124,6 @@ const std::string kCommandPrefix = "paymentRequest";
                 instanceOfClass:[JSPaymentRequestManager class]]));
     _webState = webState;
     _webStateObserver->ObserveWebState(webState);
-    [self enableCurrentWebState];

[marq (ping after 24h), 2017/01/11 15:09:57]

Removing this call means that the payment request JS won't be injected into
|webState| until the next time it calls the -webPageLoadedWithStatus: observer
method. 

[Moe, 2017/01/11 16:45:19]

does the JS really need to be injected when switching tabs or inserting a new
one, before a webpage is actually loaded? pageLoadedWithStatus: observer method
seems to work reliably with different reload/navigation scenarios.

   } else {
     _webState = nullptr;
   }
@@ -146,7 +146,6 @@ const std::string kCommandPrefix = "paymentRequest";
       [self dismissUI];
     }
     _enabled = enabled;
-    [self enableCurrentWebState];

[marq (ping after 24h), 2017/01/11 15:09:57]

Removing this call means that if -enablePaymentRequest: is called, changing the
current value of _enabled, the change won't be reflected in a currently-loaded
web page.

[Moe, 2017/01/11 16:45:19]

enablePaymentRequest: gets called in BrowserViewController's setActive:
What are the scenario's where bvc is toggled between active and inactive?

   }
 }
 
@@ -167,7 +166,7 @@ const std::string kCommandPrefix = "paymentRequest";
 }
 
 - (void)enableCurrentWebState {
-  if (![self webState]) {
+  if (![self webState] || !experimental_flags::IsPaymentRequestEnabled()) {

[marq (ping after 24h), 2017/01/11 15:09:57]

I would prefer that the experimental flag gates creating an instance of this
class (at all) rather than just stopping the JS injection.

[Moe, 2017/01/11 16:45:19]

I Agree. But I found no way of adding a command line switch to enable this
feature in the EG test before an instance of the BrowserViewController is
created.

     return;
   }