Don't create separate SiteInstances for pages from the same domain and scheme...

chrome/browser/site_instance.cc

 // Copyright (c) 2006-2008 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.
 
 #include "chrome/browser/site_instance.h"
 
 #include "net/base/registry_controlled_domain.h"
 
 SiteInstance::~SiteInstance() {
   // Now that no one is referencing us, we can safely remove ourselves from
@@ skipping 67 matching lines @@
 /*static*/
 GURL SiteInstance::GetSiteForURL(const GURL& url) {
   // URLs with no host should have an empty site.
   GURL site;
 
   // TODO(creis): For many protocols, we should just treat the scheme as the
   // site, since there is no host.  e.g., file:, about:, chrome-resource:
 
   // If the url has a host, then determine the site.
   if (url.has_host()) {
-    // Only keep the scheme, registered domain, and port as given by GetOrigin.
+    // Only keep the scheme and registered domain as given by GetOrigin.  This
+    // may also include a port, which we need to drop.
     site = url.GetOrigin();
 
+    // Remove port, if any.
+    if (site.has_port()) {
+      GURL::Replacements rep;
+      rep.ClearPort();
+      site = site.ReplaceComponents(rep);
+    }
+
     // If this URL has a registered domain, we only want to remember that part.
     std::string domain =
         net::RegistryControlledDomainService::GetDomainAndRegistry(url);
     if (!domain.empty()) {
       GURL::Replacements rep;
       rep.SetHostStr(domain);
       site = site.ReplaceComponents(rep);
     }
   }
   return site;
 }
 
 /*static*/
 bool SiteInstance::IsSameWebSite(const GURL& url1, const GURL& url2) {
   // We infer web site boundaries based on the registered domain name of the
-  // top-level page, as well as the scheme and the port.
+  // top-level page and the scheme.  We do not pay attention to the port if
+  // one is present, because pages served from different ports can still
+  // access each other if they change their document.domain variable.
 
   // We must treat javascript: URLs as part of the same site, regardless of
   // the site.
   if (url1.SchemeIs("javascript") || url2.SchemeIs("javascript"))
     return true;
 
   // We treat about:crash, about:hang, and about:shorthang as the same site as
   // any URL, since they are used as demos for crashing/hanging a process.
   GURL about_crash = GURL("about:crash");
   GURL about_hang = GURL("about:hang");
   GURL about_shorthang = GURL("about:shorthang");
   if (url1 == about_crash || url2 == about_crash ||
           url1 == about_hang || url2 == about_hang ||
           url1 == about_shorthang || url2 == about_shorthang)
     return true;
 
   // If either URL is invalid, they aren't part of the same site.
   if (!url1.is_valid() || !url2.is_valid()) {
     return false;
   }
 
-  // If the scheme or port differ, they aren't part of the same site.
-  if (url1.scheme() != url2.scheme() || url1.port() != url2.port()) {
+  // If the schemes differ, they aren't part of the same site.
+  if (url1.scheme() != url2.scheme()) {
     return false;
   }
 
   return net::RegistryControlledDomainService::SameDomainOrHost(url1, url2);
 }