Initial support for built-in DNS resolver/cache.

net/base/host_resolver_impl.cc

Index: net/base/host_resolver_impl.cc
diff --git a/net/base/host_resolver_impl.cc b/net/base/host_resolver_impl.cc
index 246f50a4a199dc51c0ae2d2a3532ba9dd144f8e4..13dce85f31f0e1aae7a696118c993f7151f0e394 100644
--- a/net/base/host_resolver_impl.cc
+++ b/net/base/host_resolver_impl.cc
@@ -72,6 +72,7 @@ HostCache* CreateDefaultCache() {
 }  // anonymous namespace
 
 HostResolver* CreateSystemHostResolver(size_t max_concurrent_resolves,
+                                       HostResolverProc* resolver_proc,

[eroman, 2011/01/10 20:13:45]

This change doesn't make sense to me.

The "CreateSystemHostResolver()" is supposed to be creating a host resolver that
uses the underlying system's getaddrinfo().

However if we pass in a custom HostResolverProc then that is not the case.

I don't think we should be adding an extra parameter to this function which will
always be NULL. Rather the one location(s) which want to create their custom
non-system host resolver should just call:

   new HostResolverImpl(resolver_proc, ....)

In fact for the feature you ware working on I think it should be implemented as
a separate HostResolver subclass alltogether, and not use a custom
HostResolverProc.

(The HostResolverProc were really just a feature for unit-tests, in order to
mock out the system dependency).

                                        NetLog* net_log) {
   // Maximum of 50 concurrent threads.
   // TODO(eroman): Adjust this, do some A/B experiments.
@@ -81,7 +82,7 @@ HostResolver* CreateSystemHostResolver(size_t max_concurrent_resolves,
     max_concurrent_resolves = kDefaultMaxJobs;
 
   HostResolverImpl* resolver =
-      new HostResolverImpl(NULL, CreateDefaultCache(),
+      new HostResolverImpl(resolver_proc, CreateDefaultCache(),
                            max_concurrent_resolves, net_log);
 
   return resolver;