Logo Search packages:      
Sourcecode: chromium-browser version File versions  Download package


// 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/history/visit_tracker.h"

#include "base/stl_util-inl.h"

namespace history {

// When the list gets longer than 'MaxItems', CleanupTransitionList will resize
// the list down to 'ResizeTo' size. This is so we only do few block moves of
// the data rather than constantly shuffle stuff around in the vector.
static const size_t kMaxItemsInTransitionList = 96;
static const size_t kResizeBigTransitionListTo = 64;
COMPILE_ASSERT(kResizeBigTransitionListTo < kMaxItemsInTransitionList,

VisitTracker::VisitTracker() {

VisitTracker::~VisitTracker() {
  STLDeleteContainerPairSecondPointers(hosts_.begin(), hosts_.end());

// This function is potentially slow because it may do up to two brute-force
// searches of the transitions list. This transitions list is kept to a
// relatively small number by CleanupTransitionList so it shouldn't be a big
// deal. However, if this ends up being noticable for performance, we may want
// to optimize lookup.
VisitID VisitTracker::GetLastVisit(const void* host,
                                   int32 page_id,
                                   const GURL& referrer) {
  if (referrer.is_empty() || !host)
    return 0;

  HostList::iterator i = hosts_.find(host);
  if (i == hosts_.end())
    return 0;  // We don't have any entries for this host.
  TransitionList& transitions = *i->second;

  // Recall that a page ID is associated with a single session history entry.
  // In the case of automatically loaded iframes, many visits/URLs can have the
  // same page ID.
  // We search backwards, starting at the current page ID, for the referring
  // URL. This won't always be correct. For example, if a render process has
  // the same page open in two different tabs, or even in two different frames,
  // we can get confused about which was which. We can have the renderer
  // report more precise referrer information in the future, but this is a
  // hard problem and doesn't affect much in terms of real-world issues.
  // We assume that the page IDs are increasing over time, so larger IDs than
  // the current input ID happened in the future (this will occur if the user
  // goes back). We can ignore future transitions because if you navigate, go
  // back, and navigate some more, we'd like to have one node with two out
  // edges in our visit graph.
  for (int i = static_cast<int>(transitions.size()) - 1; i >= 0; i--) {
    if (transitions[i].page_id <= page_id && transitions[i].url == referrer) {
      // Found it.
      return transitions[i].visit_id;

  // We can't find the referrer.
  return 0;

void VisitTracker::AddVisit(const void* host,
                            int32 page_id,
                            const GURL& url,
                            VisitID visit_id) {
  TransitionList* transitions = hosts_[host];
  if (!transitions) {
    transitions = new TransitionList;
    hosts_[host] = transitions;

  Transition t;
  t.url = url;
  t.page_id = page_id;
  t.visit_id = visit_id;


void VisitTracker::NotifyRenderProcessHostDestruction(const void* host) {
  HostList::iterator i = hosts_.find(host);
  if (i == hosts_.end())
    return;  // We don't have any entries for this host.

  delete i->second;

void VisitTracker::CleanupTransitionList(TransitionList* transitions) {
  if (transitions->size() <= kMaxItemsInTransitionList)
    return;  // Nothing to do.

                     transitions->begin() + kResizeBigTransitionListTo);

}  // namespace history

Generated by  Doxygen 1.6.0   Back to index