Get rid of websocket_over_spdy_enabled global.

chrome/browser/io_thread.h

 // Copyright (c) 2012 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef CHROME_BROWSER_IO_THREAD_H_
 #define CHROME_BROWSER_IO_THREAD_H_
 
 #include <string>
 #include <vector>
 
@@ skipping 163 matching lines @@
     Optional<bool> enable_quic_persist_server_info;
     Optional<bool> enable_quic_port_selection;
     Optional<size_t> quic_max_packet_length;
     Optional<net::QuicVersionVector> quic_supported_versions;
     Optional<net::HostPortPair> origin_to_force_quic_on;
     bool enable_user_alternate_protocol_ports;
     // NetErrorTabHelper uses |dns_probe_service| to send DNS probes when a
     // main frame load fails with a DNS error in order to provide more useful
     // information to the renderer so it can show a more specific error page.
     scoped_ptr<chrome_browser_net::DnsProbeService> dns_probe_service;
+
+    Optional<bool> enable_websocket_over_spdy;

[Adam Rice, 2014/05/30 01:06:53]

I assume the argument about the usefulness of Optional<bool> has long since
blown over, so I won't restart it. Unless you want to.

[mmenke, 2014/05/30 01:46:46]

I have no real opinion, just keeping the existing style.

Actually, my preference would be to get rid of all these we could, and just read
command line flags directly whenever initializing a context.  Reading the flags
and copying values seems weird (Except for globals), and putting them here
encourages use externally as well, and any code that manually copies these
values is bound to get out of date.

[Ryan Hamilton, 2014/05/30 02:37:02]

I'm not sure what the objection to Optional<> is here. The idea is that
HttpNetworkSession::Params gets to define the defaults for various options. As a
result, IOThread only sets the params that it needs to override. For example, if
neither --enable-quic nor --disable-quic are present (and none of the QUIC field
trials are active) then IOThread leaves the QUIC state alone. I'm happy to
consider an alternative approach than Optional<>, so long as it retains these
semantics.

[Adam Rice, 2014/05/30 03:25:24]

Well, in this case there's no "disable" flag, so we're essentially just using
Optional for a rainy day.

An alternative would be to define the defaults here instead of in
HttpNetworkSession::Params. If you wanted to be fancy you could do something
like

Defaulted<bool, false> enable_quic;

[Ryan Hamilton, 2014/05/30 03:43:27]

*nod*
We definitely do not want the defaults to be defined here :> The defaults should
live in net/ so that our unit tests can run with the right config by default.

   };
 
   // |net_log| must either outlive the IOThread or be NULL.
   IOThread(PrefService* local_state,
            policy::PolicyService* policy_service,
            ChromeNetLog* net_log,
            extensions::EventRouterForwarder* extension_event_router_forwarder);
 
   virtual ~IOThread();
 
@@ skipping 175 matching lines @@
   bool is_spdy_disabled_by_policy_;
 
   base::WeakPtrFactory<IOThread> weak_factory_;
 
   const base::TimeTicks creation_time_;
 
   DISALLOW_COPY_AND_ASSIGN(IOThread);
 };
 
 #endif  // CHROME_BROWSER_IO_THREAD_H_