Fix a load timing bug in the case of SPDY session reuse

net/base/load_timing_info.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 NET_BASE_LOAD_TIMING_INFO_H_
 #define NET_BASE_LOAD_TIMING_INFO_H_
 
 #include "base/basictypes.h"
 #include "base/time.h"
 #include "net/base/net_export.h"
 
 namespace net {
 
 // Structure containing timing information for a request.
 // It addresses the needs of
 // http://groups.google.com/group/http-archive-specification/web/har-1-1-spec,
 // http://dev.w3.org/2006/webapi/WebTiming/, and
 // http://www.w3.org/TR/resource-timing/.
 //
 // All events that do not apply to a request have null times.  For non-HTTP
 // requests, all times other than the request_start times are null.
 //
 // Requests with connection errors generally only have request start times as
 // well, since they never received an established socket.
 //
 // The general order for events is:
 // request_start
 // proxy_start
 // proxy_end
-// *dns_start
-// *dns_end
-// *connect_start
-// *ssl_start
-// *ssl_end
-// *connect_end
+// dns_start
+// dns_end
+// connect_start
+// ssl_start
+// ssl_end
+// connect_end
 // send_start
 // send_end
 // receive_headers_end
 //
-// Those times without an asterisk are computed by the URLRequest, or by objects
-// it directly creates and always owns.  Those with an asterisk are computed
-// by the connection attempt itself.  Since the connection attempt may be
-// started before a URLRequest, the starred times may occur before, during, or
-// after the request_start and proxy events.
+// Times represent the length of time that the request was blocked on an event,

[eroman, 2013/04/06 01:02:02]

This comment confused me on the first read (I thought it was implying the times
were durations rather than points in time).

Perhaps what threw me was "length of time"

[mmenke, 2013/04/08 15:31:24]

Updated it to say "times represent when a request starts/stops block on an
events"

+// not the time they actually occurred.  In particular, in the case of
+// preconnects and socket reuse, no time is spent blocked on establishing a
+// connection.  In the case of SPDY and HTTP pipelining, PAC scripts are only
+// run once for each shared session, so no time may be spent blocked on them.
 //
 // DNS and SSL times are both times for the host, not the proxy, so DNS times
 // when using proxies are null, and only requests to HTTPS hosts (Not proxies)
 // have SSL times.  One exception to this is when a proxy server itself returns
 // a redirect response.  In this case, the connect times treat the proxy as the
 // host.  The send and receive times will all be null, however.
 // See HttpNetworkTransaction::OnHttpsProxyTunnelResponse.
 // TODO(mmenke):  Is this worth fixing?
+//
+// Note that internal to the network stack, times are when events actually
+// occurred.  URLRequest converts them to time which the network stack was
+// blocked on each state.
 struct NET_EXPORT LoadTimingInfo {
   // Contains the LoadTimingInfo events related to establishing a connection.
   // These are all set by ConnectJobs.
   struct NET_EXPORT_PRIVATE ConnectTiming {
     ConnectTiming();
     ~ConnectTiming();
 
     // The time spent looking up the host's DNS address.  Null for requests that
     // used proxies to look up the DNS address.  Also null for SOCKS4 proxies,
     // since the DNS address is only looked up after the connection is
@@ skipping 65 matching lines @@
   base::TimeTicks send_start;
   base::TimeTicks send_end;
 
   // The time at which the end of the HTTP headers were received.
   base::TimeTicks receive_headers_end;
 };
 
 }  // namespace net
 
 #endif  // NET_BASE_LOAD_TIMING_INFO_H_