Issue 1727 - Fix WebSocket constructor without second argument in Chrome 47

include.preload.js

Index: include.preload.js
===================================================================
--- a/include.preload.js
+++ b/include.preload.js
@@ -373,6 +373,7 @@
     // As far as possible we must track everything we use that could be
     // sabotaged by the website later in order to circumvent us.
     var RealWebSocket = WebSocket;
+    var bindWebSocket = Function.prototype.bind.apply.bind(Function.prototype.bind, RealWebSocket);
     var closeWebSocket = Function.prototype.call.bind(RealWebSocket.prototype.close);
     var addEventListener = document.addEventListener.bind(document);
     var removeEventListener = document.removeEventListener.bind(document);
@@ -394,13 +395,14 @@
       }));
     }
 
-    WebSocket = function WrappedWebSocket(url, protocols)
+    WebSocket = function WrappedWebSocket()
     {
-      // Throw correct exceptions if the constructor is used improperly.
-      if (!(this instanceof WrappedWebSocket)) return RealWebSocket();
-      if (arguments.length < 1) return new RealWebSocket();
+      var args = [null];

[kzar, 2016/08/11 14:25:05]

I think this approach is extremely hard to follow, when something way simpler
would do:

```
if (!(this instanceof WrappedWebSocket)) return RealWebSocket();
if (arguments.length == 0) return new RealWebSocket();

var websocket;
if (arguments.length == 1) websocket = new RealWebSocket(url);
else if (arguments.length == 2) websocket = new RealWebSocket(url, protocols);
```

Sure it has a little duplication and isn't as clever, but is that such a bad
thing? I mean bindWebSocket for example is getting nearly impenetrable to
follow, good luck trying to figure out what
`Function.prototype.bind.apply.bind(Function.prototype.bind, RealWebSocket);`
does! The logic below with the creator is hardly easy to follow either.

[Sebastian Noack, 2016/08/11 14:38:46]

Well, if you'd write your logic compliant to our coding styles (add newlines
where appropriate) it wouldn't be any shorter, however it has quite a few
assumptions, and it's virtually impossible to make sure those hold true for all
current and future supported browser versions. The pure fact that we'd have to
extend that logic already, is a clear indication that this was a bad idea in the
first place.

My code on the other hand makes sure that the original WebSocket function is
called in the exact same way our wrapper is called, delegating the whole
behavior to the browser.

[kzar, 2016/08/11 14:54:16]

I didn't say anything about the code being shorter, but simpler. As far as I see
it we can fix the problem you spotted by simply adding a check to see if the
arguments' length is 1, I just don't agree about instead doing something so
complicated and hard to follow.

The WebSocket constructor takes two arguments url and protocol, we know this.
Maybe one day in the future that will change, but maybe one day in the future
Chrome will support interception of WebSocket connections anyway. Since when do
we do things in case it's one day useful?

[Sebastian Noack, 2016/08/11 15:11:35]

I don't think that I agree. But there you go.

+      for (var i = 0; i < arguments.length; i++)
+        args.push(arguments[i]);

[kzar, 2016/08/11 14:25:05]

This would allow for circumvention since we didn't keep track of
Array.prototype.push, but like I said above I think we should rather keep this
simple.

[Sebastian Noack, 2016/08/11 14:38:46]

Well spotted, fixed.

 
-      var websocket = new RealWebSocket(url, protocols);
+      var ctor = bindWebSocket(args);
+      var websocket = this instanceof WrappedWebSocket ? new ctor : ctor();
 
       checkRequest(websocket.url, function(blocked)
       {