Add reprioritization to socket pools.

net/socket/websocket_transport_client_socket_pool.cc

Index: net/socket/websocket_transport_client_socket_pool.cc
diff --git a/net/socket/websocket_transport_client_socket_pool.cc b/net/socket/websocket_transport_client_socket_pool.cc
index 0239226da9910a94aea263729e8da7bbcad538f9..05ee439d36fa21e4c68a15db5e7e9f4971ee62e9 100644
--- a/net/socket/websocket_transport_client_socket_pool.cc
+++ b/net/socket/websocket_transport_client_socket_pool.cc
@@ -392,6 +392,16 @@ void WebSocketTransportClientSocketPool::RequestSockets(
   NOTIMPLEMENTED();
 }
 
+void WebSocketTransportClientSocketPool::SetPriority(
+    const std::string& group_name,
+    ClientSocketHandle* handle,
+    RequestPriority priority) {
+  // Since sockets requested by RequestSocket are bound early, there's
+  // nothing to do within the pool to change priority or the request.
+  // TODO(rdsmith): Investigate plumbing the reprioritization request to the
+  // connect job.

[mmenke, 2017/01/18 20:37:27]

What if it's in stalled_request_map_?

[Randy Smith (Not in Mondays), 2017/01/22 21:37:00]

Hmmm.  Good point, I missed that.

However, stalled_request_map_ is blissfully priority unaware, even of the
priority that request carries when it comes in, AFAICT.  So making SetPriority()
have any effect would almost force stalled_request_map_ over to being a priority
queue, which would be a functional change to WebSockets, which I'm a bit
reluctant to do?

Adam, was this intentional?  Are you positive, neutral, or negative about
changing stalled_request_map_ to know about priorities?  (Not promising to do it
if you're positive, just trying to understand the lay of the land.)

[Adam Rice, 2017/01/23 02:26:34]

All WebSocket handshakes happen at DEFAULT_PRIORITY, so supporting SetPriority
correctly for WebSocket would be pointless. Unless it makes the code simpler,
ignoring priorities is the best option.

+}
+
 void WebSocketTransportClientSocketPool::CancelRequest(
     const std::string& group_name,
     ClientSocketHandle* handle) {