blob: ab481b406b8a21ce9a9c9603cae23d291b0e2a21 [file] [log] [blame]
// 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.
#include "chrome/browser/autocomplete/history_url_provider.h"
#include <algorithm>
#include "base/basictypes.h"
#include "base/bind.h"
#include "base/command_line.h"
#include "base/message_loop/message_loop.h"
#include "base/metrics/histogram.h"
#include "base/prefs/pref_service.h"
#include "base/strings/string_util.h"
#include "base/strings/utf_string_conversions.h"
#include "base/time/time.h"
#include "chrome/browser/autocomplete/autocomplete_match.h"
#include "chrome/browser/autocomplete/autocomplete_provider_listener.h"
#include "chrome/browser/autocomplete/autocomplete_result.h"
#include "chrome/browser/history/history_backend.h"
#include "chrome/browser/history/history_database.h"
#include "chrome/browser/history/history_service.h"
#include "chrome/browser/history/history_service_factory.h"
#include "chrome/browser/history/history_types.h"
#include "chrome/browser/history/in_memory_url_index_types.h"
#include "chrome/browser/history/scored_history_match.h"
#include "chrome/browser/omnibox/omnibox_field_trial.h"
#include "chrome/browser/profiles/profile.h"
#include "chrome/browser/search_engines/template_url_service.h"
#include "chrome/browser/search_engines/template_url_service_factory.h"
#include "chrome/browser/search_engines/ui_thread_search_terms_data.h"
#include "chrome/common/chrome_switches.h"
#include "chrome/common/pref_names.h"
#include "chrome/common/url_constants.h"
#include "components/autocomplete/url_prefix.h"
#include "components/bookmarks/browser/bookmark_utils.h"
#include "components/metrics/proto/omnibox_input_type.pb.h"
#include "components/url_fixer/url_fixer.h"
#include "content/public/browser/browser_thread.h"
#include "net/base/net_util.h"
#include "net/base/registry_controlled_domains/registry_controlled_domain.h"
#include "url/gurl.h"
#include "url/url_parse.h"
#include "url/url_util.h"
namespace {
// Acts like the > operator for URLInfo classes.
bool CompareHistoryMatch(const history::HistoryMatch& a,
const history::HistoryMatch& b) {
// A URL that has been typed at all is better than one that has never been
// typed. (Note "!"s on each side)
if (!a.url_info.typed_count() != !b.url_info.typed_count())
return a.url_info.typed_count() > b.url_info.typed_count();
// Innermost matches (matches after any scheme or "www.") are better than
// non-innermost matches.
if (a.innermost_match != b.innermost_match)
return a.innermost_match;
// URLs that have been typed more often are better.
if (a.url_info.typed_count() != b.url_info.typed_count())
return a.url_info.typed_count() > b.url_info.typed_count();
// For URLs that have each been typed once, a host (alone) is better than a
// page inside.
if ((a.url_info.typed_count() == 1) && (a.IsHostOnly() != b.IsHostOnly()))
return a.IsHostOnly();
// URLs that have been visited more often are better.
if (a.url_info.visit_count() != b.url_info.visit_count())
return a.url_info.visit_count() > b.url_info.visit_count();
// URLs that have been visited more recently are better.
return a.url_info.last_visit() > b.url_info.last_visit();
// Sorts and dedups the given list of matches.
void SortAndDedupMatches(history::HistoryMatches* matches) {
// Sort by quality, best first.
std::sort(matches->begin(), matches->end(), &CompareHistoryMatch);
// Remove duplicate matches (caused by the search string appearing in one of
// the prefixes as well as after it). Consider the following scenario:
// User has visited "" once and "" twice.
// User types "http". The autocomplete search with prefix "http://" returns
// the first host, while the search with prefix "" returns both hosts. Now
// we sort them into rank order:
// (innermost_match)
// (!innermost_match, url_info.visit_count == 2)
// (!innermost_match, url_info.visit_count == 1)
// The above scenario tells us we can't use std::unique(), since our
// duplicates are not always sequential. It also tells us we should remove
// the lower-quality duplicate(s), since otherwise the returned results won't
// be ordered correctly. This is easy to do: we just always remove the later
// element of a duplicate pair.
// Be careful! Because the vector contents may change as we remove elements,
// we use an index instead of an iterator in the outer loop, and don't
// precalculate the ending position.
for (size_t i = 0; i < matches->size(); ++i) {
for (history::HistoryMatches::iterator j(matches->begin() + i + 1);
j != matches->end(); ) {
if ((*matches)[i].url_info.url() == j->url_info.url())
j = matches->erase(j);
// Calculates a new relevance score applying half-life time decaying to |count|
// using |time_since_last_visit| and |score_buckets|. This function will never
// return a score higher than |undecayed_relevance|; in other words, it can only
// demote the old score.
double CalculateRelevanceUsingScoreBuckets(
const HUPScoringParams::ScoreBuckets& score_buckets,
const base::TimeDelta& time_since_last_visit,
int undecayed_relevance,
int count) {
// Back off if above relevance cap.
if ((score_buckets.relevance_cap() != -1) &&
(undecayed_relevance >= score_buckets.relevance_cap()))
return undecayed_relevance;
// Time based decay using half-life time.
double decayed_count = count;
if (decayed_count > 0)
decayed_count *= score_buckets.HalfLifeTimeDecay(time_since_last_visit);
// Find a threshold where decayed_count >= bucket.
const HUPScoringParams::ScoreBuckets::CountMaxRelevance* score_bucket = NULL;
for (size_t i = 0; i < score_buckets.buckets().size(); ++i) {
score_bucket = &score_buckets.buckets()[i];
if (decayed_count >= score_bucket->first)
break; // Buckets are in descending order, so we can ignore the rest.
return (score_bucket && (undecayed_relevance > score_bucket->second)) ?
score_bucket->second : undecayed_relevance;
// Returns a new relevance score for the given |match| based on the
// |old_relevance| score and |scoring_params|. The new relevance score is
// guaranteed to be less than or equal to |old_relevance|. In other words, this
// function can only demote a score, never boost it. Returns |old_relevance| if
// experimental scoring is disabled.
int CalculateRelevanceScoreUsingScoringParams(
const history::HistoryMatch& match,
int old_relevance,
const HUPScoringParams& scoring_params) {
if (!scoring_params.experimental_scoring_enabled)
return old_relevance;
const base::TimeDelta time_since_last_visit =
base::Time::Now() - match.url_info.last_visit();
int relevance = CalculateRelevanceUsingScoreBuckets(
scoring_params.typed_count_buckets, time_since_last_visit, old_relevance,
// Additional demotion (on top of typed_count demotion) of URLs that were
// never typed.
if (match.url_info.typed_count() == 0) {
relevance = CalculateRelevanceUsingScoreBuckets(
scoring_params.visited_count_buckets, time_since_last_visit, relevance,
DCHECK_LE(relevance, old_relevance);
return relevance;
// Extracts typed_count, visit_count, and last_visited time from the URLRow and
// puts them in the additional info field of the |match| for display in
// about:omnibox.
void RecordAdditionalInfoFromUrlRow(const history::URLRow& info,
AutocompleteMatch* match) {
match->RecordAdditionalInfo("typed count", info.typed_count());
match->RecordAdditionalInfo("visit count", info.visit_count());
match->RecordAdditionalInfo("last visit", info.last_visit());
// If |create_if_necessary| is true, ensures that |matches| contains an entry
// for |info|, creating a new such entry if necessary (using |input_location|
// and |match_in_scheme|).
// If |promote| is true, this also ensures the entry is the first element in
// |matches|, moving or adding it to the front as appropriate. When |promote|
// is false, existing matches are left in place, and newly added matches are
// placed at the back.
// It's OK to call this function with both |create_if_necessary| and |promote|
// false, in which case we'll do nothing.
// Returns whether the match exists regardless if it was promoted/created.
bool CreateOrPromoteMatch(const history::URLRow& info,
size_t input_location,
bool match_in_scheme,
history::HistoryMatches* matches,
bool create_if_necessary,
bool promote) {
// |matches| may already have an entry for this.
for (history::HistoryMatches::iterator i(matches->begin());
i != matches->end(); ++i) {
if (i->url_info.url() == info.url()) {
// Rotate it to the front if the caller wishes.
if (promote)
std::rotate(matches->begin(), i, i + 1);
return true;
if (!create_if_necessary)
return false;
// No entry, so create one.
history::HistoryMatch match(info, input_location, match_in_scheme, true);
if (promote)
return true;
// Returns whether |match| is suitable for inline autocompletion.
bool CanPromoteMatchForInlineAutocomplete(const history::HistoryMatch& match) {
// We can promote this match if it's been typed at least n times, where n == 1
// for "simple" (host-only) URLs and n == 2 for others. We set a higher bar
// for these long URLs because it's less likely that users will want to visit
// them again. Even though we don't increment the typed_count for pasted-in
// URLs, if the user manually edits the URL or types some long thing in by
// hand, we wouldn't want to immediately start autocompleting it.
return match.url_info.typed_count() &&
((match.url_info.typed_count() > 1) || match.IsHostOnly());
// Given the user's |input| and a |match| created from it, reduce the match's
// URL to just a host. If this host still matches the user input, return it.
// Returns the empty string on failure.
GURL ConvertToHostOnly(const history::HistoryMatch& match,
const base::string16& input) {
// See if we should try to do host-only suggestions for this URL. Nonstandard
// schemes means there's no authority section, so suggesting the host name
// is useless. File URLs are standard, but host suggestion is not useful for
// them either.
const GURL& url = match.url_info.url();
if (!url.is_valid() || !url.IsStandard() || url.SchemeIsFile())
return GURL();
// Transform to a host-only match. Bail if the host no longer matches the
// user input (e.g. because the user typed more than just a host).
GURL host = url.GetWithEmptyPath();
if ((host.spec().length() < (match.input_location + input.length())))
return GURL(); // User typing is longer than this host suggestion.
const base::string16 spec = base::UTF8ToUTF16(host.spec());
if (, input.length(), input))
return GURL(); // User typing is no longer a prefix.
return host;
} // namespace
// -----------------------------------------------------------------
// SearchTermsDataSnapshot
// Implementation of SearchTermsData that takes a snapshot of another
// SearchTermsData by copying all the responses to the different getters into
// member strings, then returning those strings when its own getters are called.
// This will typically be constructed on the UI thread from
// UIThreadSearchTermsData but is subsequently safe to use on any thread.
class SearchTermsDataSnapshot : public SearchTermsData {
explicit SearchTermsDataSnapshot(const SearchTermsData& search_terms_data);
virtual ~SearchTermsDataSnapshot();
virtual std::string GoogleBaseURLValue() const OVERRIDE;
virtual std::string GetApplicationLocale() const OVERRIDE;
virtual base::string16 GetRlzParameterValue(
bool from_app_list) const OVERRIDE;
virtual std::string GetSearchClient() const OVERRIDE;
virtual std::string NTPIsThemedParam() const OVERRIDE;
std::string google_base_url_value_;
std::string application_locale_;
base::string16 rlz_parameter_value_;
std::string search_client_;
std::string ntp_is_themed_param_;
const SearchTermsData& search_terms_data)
: google_base_url_value_(search_terms_data.GoogleBaseURLValue()),
ntp_is_themed_param_(search_terms_data.NTPIsThemedParam()) {}
SearchTermsDataSnapshot::~SearchTermsDataSnapshot() {
std::string SearchTermsDataSnapshot::GoogleBaseURLValue() const {
return google_base_url_value_;
std::string SearchTermsDataSnapshot::GetApplicationLocale() const {
return application_locale_;
base::string16 SearchTermsDataSnapshot::GetRlzParameterValue(
bool from_app_list) const {
return rlz_parameter_value_;
std::string SearchTermsDataSnapshot::GetSearchClient() const {
return search_client_;
std::string SearchTermsDataSnapshot::NTPIsThemedParam() const {
return ntp_is_themed_param_;
// -----------------------------------------------------------------
// HistoryURLProvider
// These ugly magic numbers will go away once we switch all scoring
// behavior (including URL-what-you-typed) to HistoryQuick provider.
const int HistoryURLProvider::kScoreForBestInlineableResult = 1413;
const int HistoryURLProvider::kScoreForUnvisitedIntranetResult = 1403;
const int HistoryURLProvider::kScoreForWhatYouTypedResult = 1203;
const int HistoryURLProvider::kBaseScoreForNonInlineableResult = 900;
// VisitClassifier is used to classify the type of visit to a particular url.
class HistoryURLProvider::VisitClassifier {
enum Type {
INVALID, // Navigations to the URL are not allowed.
UNVISITED_INTRANET, // A navigable URL for which we have no visit data but
// which is known to refer to a visited intranet host.
VISITED, // The site has been previously visited.
VisitClassifier(HistoryURLProvider* provider,
const AutocompleteInput& input,
history::URLDatabase* db);
// Returns the type of visit for the specified input.
Type type() const { return type_; }
// Returns the URLRow for the visit.
const history::URLRow& url_row() const { return url_row_; }
HistoryURLProvider* provider_;
history::URLDatabase* db_;
Type type_;
history::URLRow url_row_;
HistoryURLProvider* provider,
const AutocompleteInput& input,
history::URLDatabase* db)
: provider_(provider),
type_(INVALID) {
const GURL& url = input.canonicalized_url();
// Detect email addresses. These cases will look like "http://user@site/",
// and because the history backend strips auth creds, we'll get a bogus exact
// match below if the user has visited "site".
if (!url.is_valid() ||
((input.type() == metrics::OmniboxInputType::UNKNOWN) && &&
! &&
if (db_->GetRowForURL(url, &url_row_)) {
type_ = VISITED;
if (provider_->CanFindIntranetURL(db_, input)) {
// The user typed an intranet hostname that they've visited (albeit with a
// different port and/or path) before.
url_row_ = history::URLRow(url);
const AutocompleteInput& input,
bool trim_http,
const AutocompleteMatch& what_you_typed_match,
const std::string& languages,
TemplateURL* default_search_provider,
const SearchTermsData& search_terms_data)
: message_loop(base::MessageLoop::current()),
default_search_provider(default_search_provider ?
new TemplateURL(default_search_provider->data()) : NULL),
search_terms_data(new SearchTermsDataSnapshot(search_terms_data)) {
HistoryURLProviderParams::~HistoryURLProviderParams() {
HistoryURLProvider::HistoryURLProvider(AutocompleteProviderListener* listener,
Profile* profile)
: HistoryProvider(listener, profile,
!OmniboxFieldTrial::InHUPCullRedirectsFieldTrial() ||
!OmniboxFieldTrial::InHUPCreateShorterMatchFieldTrial() ||
InHUPCreateShorterMatchFieldTrialExperimentGroup()) {
// Initialize HUP scoring params based on the current experiment.
void HistoryURLProvider::Start(const AutocompleteInput& input,
bool minimal_changes) {
// NOTE: We could try hard to do less work in the |minimal_changes| case
// here; some clever caching would let us reuse the raw matches from the
// history DB without re-querying. However, we'd still have to go back to
// the history thread to mark these up properly, and if pass 2 is currently
// running, we'd need to wait for it to return to the main thread before
// doing this (we can't just write new data for it to read due to thread
// safety issues). At that point it's just as fast, and easier, to simply
// re-run the query from scratch and ignore |minimal_changes|.
// Cancel any in-progress query.
if ((input.type() == metrics::OmniboxInputType::INVALID) ||
(input.type() == metrics::OmniboxInputType::FORCED_QUERY))
// Do some fixup on the user input before matching against it, so we provide
// good results for local file paths, input with spaces, etc.
const FixupReturn fixup_return(FixupUserInput(input));
if (!fixup_return.first)
url::Parsed parts;
url_fixer::SegmentURL(fixup_return.second, &parts);
AutocompleteInput fixed_up_input(input);
fixed_up_input.UpdateText(fixup_return.second, base::string16::npos, parts);
// Create a match for what the user typed.
const bool trim_http = !AutocompleteInput::HasHTTPScheme(input.text());
AutocompleteMatch what_you_typed_match(SuggestExactInput(
fixed_up_input.text(), fixed_up_input.canonicalized_url(), trim_http));
what_you_typed_match.relevance = CalculateRelevance(WHAT_YOU_TYPED, 0);
// Add the WYT match as a fallback in case we can't get the history service or
// URL DB; otherwise, we'll replace this match lower down. Don't do this for
// queries, though -- while we can sometimes mark up a match for them, it's
// not what the user wants, and just adds noise.
if (fixed_up_input.type() != metrics::OmniboxInputType::QUERY)
// We'll need the history service to run both passes, so try to obtain it.
if (!profile_)
HistoryService* const history_service =
HistoryServiceFactory::GetForProfile(profile_, Profile::EXPLICIT_ACCESS);
if (!history_service)
// Get the default search provider and search terms data now since we have to
// retrieve these on the UI thread, and the second pass runs on the history
// thread. |template_url_service| can be NULL when testing.
TemplateURLService* template_url_service =
TemplateURL* default_search_provider = template_url_service ?
template_url_service->GetDefaultSearchProvider() : NULL;
UIThreadSearchTermsData data(profile_);
// Create the data structure for the autocomplete passes. We'll save this off
// onto the |params_| member for later deletion below if we need to run pass
// 2.
scoped_ptr<HistoryURLProviderParams> params(
new HistoryURLProviderParams(
fixed_up_input, trim_http, what_you_typed_match,
default_search_provider, data));
// Note that we use the non-fixed-up input here, since fixup may strip
// trailing whitespace.
params->prevent_inline_autocomplete = PreventInlineAutocomplete(input);
// Pass 1: Get the in-memory URL database, and use it to find and promote
// the inline autocomplete match, if any.
history::URLDatabase* url_db = history_service->InMemoryDatabase();
// url_db can be NULL if it hasn't finished initializing (or failed to
// initialize). In this case all we can do is fall back on the second
// pass.
// TODO(pkasting): We should just block here until this loads. Any time
// someone unloads the history backend, we'll get inconsistent inline
// autocomplete behavior here.
if (url_db) {
DoAutocomplete(NULL, url_db, params.get());
// NOTE: We don't reset |params| here since at least the |promote_type|
// field on it will be read by the second pass -- see comments in
// DoAutocomplete().
// Pass 2: Ask the history service to call us back on the history thread,
// where we can read the full on-disk DB.
if (input.want_asynchronous_matches()) {
done_ = false;
params_ = params.release(); // This object will be destroyed in
// QueryComplete() once we're done with it.
history_service->ScheduleAutocomplete(this, params_);
void HistoryURLProvider::Stop(bool clear_cached_results) {
done_ = true;
if (params_)
AutocompleteMatch HistoryURLProvider::SuggestExactInput(
const base::string16& text,
const GURL& destination_url,
bool trim_http) {
// The FormattedStringWithEquivalentMeaning() call below requires callers to
// be on the UI thread.
DCHECK(content::BrowserThread::CurrentlyOn(content::BrowserThread::UI) ||
AutocompleteMatch match(this, 0, false,
if (destination_url.is_valid()) {
match.destination_url = destination_url;
// Trim off "http://" if the user didn't type it.
// NOTE: We use TrimHttpPrefix() here rather than StringForURLDisplay() to
// strip the scheme as we need to know the offset so we can adjust the
// |match_location| below. StringForURLDisplay() and TrimHttpPrefix() have
// slightly different behavior as well (the latter will strip even without
// two slashes after the scheme).
DCHECK(!trim_http || !AutocompleteInput::HasHTTPScheme(text));
base::string16 display_string(
StringForURLDisplay(destination_url, false, false));
const size_t offset = trim_http ? TrimHttpPrefix(&display_string) : 0;
match.fill_into_edit =
match.allowed_to_be_default_match = true;
// NOTE: Don't set match.inline_autocompletion to something non-empty here;
// it's surprising and annoying.
// Try to highlight "innermost" match location. If we fix up "w" into
// "", we want to highlight the fifth character, not the first.
// This relies on match.destination_url being the non-prefix-trimmed version
// of match.contents.
match.contents = display_string;
const URLPrefix* best_prefix = URLPrefix::BestURLPrefix(
base::UTF8ToUTF16(destination_url.spec()), text);
// It's possible for match.destination_url to not contain the user's input
// at all (so |best_prefix| is NULL), for example if the input is
// "view-source:x" and |destination_url| has an inserted "http://" in the
// middle.
if (best_prefix == NULL) {
AutocompleteMatch::ClassifyMatchInString(text, match.contents,
} else {
best_prefix->prefix.length() - offset, text.length(),
match.contents.length(), ACMatchClassification::URL,
match.is_history_what_you_typed_match = true;
return match;
void HistoryURLProvider::ExecuteWithDB(history::HistoryBackend* backend,
history::URLDatabase* db,
HistoryURLProviderParams* params) {
// We may get called with a NULL database if it couldn't be properly
// initialized.
if (!db) {
params->failed = true;
} else if (!params->cancel_flag.IsSet()) {
base::TimeTicks beginning_time = base::TimeTicks::Now();
DoAutocomplete(backend, db, params);
base::TimeTicks::Now() - beginning_time);
// Return the results (if any) to the main thread.
params->message_loop->PostTask(FROM_HERE, base::Bind(
&HistoryURLProvider::QueryComplete, this, params));
HistoryURLProvider::~HistoryURLProvider() {
// Note: This object can get leaked on shutdown if there are pending
// requests on the database (which hold a reference to us). Normally, these
// messages get flushed for each thread. We do a round trip from main, to
// history, back to main while holding a reference. If the main thread
// completes before the history thread, the message to delegate back to the
// main thread will not run and the reference will leak. Therefore, don't do
// anything on destruction.
// static
int HistoryURLProvider::CalculateRelevance(MatchType match_type,
int match_number) {
switch (match_type) {
return kScoreForBestInlineableResult;
return kScoreForUnvisitedIntranetResult;
return kScoreForWhatYouTypedResult;
default: // NORMAL
return kBaseScoreForNonInlineableResult + match_number;
// static
ACMatchClassifications HistoryURLProvider::ClassifyDescription(
const base::string16& input_text,
const base::string16& description) {
base::string16 clean_description = bookmark_utils::CleanUpTitleForMatching(
history::TermMatches description_matches(SortAndDeoverlapMatches(
history::MatchTermInString(input_text, clean_description, 0)));
history::WordStarts description_word_starts;
clean_description, false, &description_word_starts);
// If HistoryURL retrieves any matches (and hence we reach this code), we
// are guaranteed that the beginning of input_text must be a word break.
history::WordStarts offsets(1, 0u);
description_matches =
description_matches, offsets, description_word_starts, 0,
return SpansFromTermMatch(
description_matches, clean_description.length(), false);
void HistoryURLProvider::DoAutocomplete(history::HistoryBackend* backend,
history::URLDatabase* db,
HistoryURLProviderParams* params) {
// Get the matching URLs from the DB.
history::URLRows url_matches;
const URLPrefixes& prefixes = URLPrefix::GetURLPrefixes();
for (URLPrefixes::const_iterator i(prefixes.begin()); i != prefixes.end();
++i) {
if (params->cancel_flag.IsSet())
return; // Canceled in the middle of a query, give up.
// We only need kMaxMatches results in the end, but before we get there we
// need to promote lower-quality matches that are prefixes of higher-quality
// matches, and remove lower-quality redirects. So we ask for more results
// than we need, of every prefix type, in hopes this will give us far more
// than enough to work with. CullRedirects() will then reduce the list to
// the best kMaxMatches results.
base::UTF16ToUTF8(i->prefix + params->input.text()), kMaxMatches * 2,
!backend, &url_matches);
for (history::URLRows::const_iterator j(url_matches.begin());
j != url_matches.end(); ++j) {
const URLPrefix* best_prefix = URLPrefix::BestURLPrefix(
base::UTF8ToUTF16(j->url().spec()), base::string16());
*j, i->prefix.length(), !i->num_components,
i->num_components >= best_prefix->num_components));
// Create sorted list of suggestions.
// Try to create a shorter suggestion from the best match.
// We consider the what you typed match to be eligible to be displayed when
// there's a reasonable chance the user actually cares:
// * Their input can be opened as a URL, and
// * We parsed the input as a URL, or it starts with an explicit "http:" or
// "https:".
// Otherwise, this is just low-quality noise. In the cases where we've parsed
// as UNKNOWN, we'll still show an accidental search infobar if need be.
VisitClassifier classifier(this, params->input, db);
params->have_what_you_typed_match =
(params->input.type() != metrics::OmniboxInputType::QUERY) &&
((params->input.type() != metrics::OmniboxInputType::UNKNOWN) ||
(classifier.type() == VisitClassifier::UNVISITED_INTRANET) ||
!params->trim_http ||
(AutocompleteInput::NumNonHostComponents(params-> > 0));
const bool have_shorter_suggestion_suitable_for_inline_autocomplete =
db, params->have_what_you_typed_match, params);
// Check whether what the user typed appears in history.
const bool can_check_history_for_exact_match =
// Checking what_you_typed_match.is_history_what_you_typed_match tells us
// whether SuggestExactInput() succeeded in constructing a valid match.
params->what_you_typed_match.is_history_what_you_typed_match &&
// Additionally, in the case where the user has typed "" and
// visited (but not typed) "foo/", and the input is "foo", the first pass
// will fall into the FRONT_HISTORY_MATCH case for "" but the
// second pass can suggest the exact input as a better URL. Since we need
// both passes to agree, and since during the first pass there's no way to
// know about "foo/", ensure that if the promote type was set to
// FRONT_HISTORY_MATCH during the first pass, the second pass will not
// consider the exact suggestion to be in history and therefore will not
// suggest the exact input as a better match. (Note that during the first
// pass, this conditional will always succeed since |promote_type| is
// initialized to NEITHER.)
(params->promote_type != HistoryURLProviderParams::FRONT_HISTORY_MATCH);
params->exact_suggestion_is_in_history = can_check_history_for_exact_match &&
FixupExactSuggestion(db, classifier, params);
// If we succeeded in fixing up the exact match based on the user's history,
// we should treat it as the best match regardless of input type. If not,
// then we check whether there's an inline autocompletion we can create from
// this input, so we can promote that as the best match.
if (params->exact_suggestion_is_in_history) {
params->promote_type = HistoryURLProviderParams::WHAT_YOU_TYPED_MATCH;
} else if (!params->prevent_inline_autocomplete && !params->matches.empty() &&
(have_shorter_suggestion_suitable_for_inline_autocomplete ||
CanPromoteMatchForInlineAutocomplete(params->matches[0]))) {
params->promote_type = HistoryURLProviderParams::FRONT_HISTORY_MATCH;
} else {
// Failed to promote any URLs. Use the What You Typed match, if we have it.
params->promote_type = params->have_what_you_typed_match ?
HistoryURLProviderParams::WHAT_YOU_TYPED_MATCH :
const size_t max_results =
kMaxMatches + (params->exact_suggestion_is_in_history ? 1 : 0);
if (backend && cull_redirects_) {
// Remove redirects and trim list to size. We want to provide up to
// kMaxMatches results plus the What You Typed result, if it was added to
// params->matches above.
CullRedirects(backend, &params->matches, max_results);
} else if (params->matches.size() > max_results) {
// Simply trim the list to size.
void HistoryURLProvider::PromoteMatchesIfNecessary(
const HistoryURLProviderParams& params) {
if (params.promote_type == HistoryURLProviderParams::FRONT_HISTORY_MATCH) {
matches_.push_back(HistoryMatchToACMatch(params, 0, INLINE_AUTOCOMPLETE,
CalculateRelevance(INLINE_AUTOCOMPLETE, 0)));
if (OmniboxFieldTrial::AddUWYTMatchEvenIfPromotedURLs() &&
params.have_what_you_typed_match) {
} else if (params.promote_type ==
HistoryURLProviderParams::WHAT_YOU_TYPED_MATCH) {
void HistoryURLProvider::QueryComplete(
HistoryURLProviderParams* params_gets_deleted) {
// Ensure |params_gets_deleted| gets deleted on exit.
scoped_ptr<HistoryURLProviderParams> params(params_gets_deleted);
// If the user hasn't already started another query, clear our member pointer
// so we can't write into deleted memory.
if (params_ == params_gets_deleted)
params_ = NULL;
// Don't send responses for queries that have been canceled.
if (params->cancel_flag.IsSet())
return; // Already set done_ when we canceled, no need to set it again.
// Don't modify |matches_| if the query failed, since it might have a default
// match in it, whereas |params->matches| will be empty.
if (!params->failed) {
// Determine relevance of highest scoring match, if any.
int relevance = matches_.empty() ?
static_cast<int>(params->matches.size() - 1)) :
// Convert the history matches to autocomplete matches. If we promoted the
// first match, skip over it.
const size_t first_match =
(params->exact_suggestion_is_in_history ||
(params->promote_type ==
HistoryURLProviderParams::FRONT_HISTORY_MATCH)) ? 1 : 0;
for (size_t i = first_match; i < params->matches.size(); ++i) {
// All matches score one less than the previous match.
// The experimental scoring must not change the top result's score.
if (!matches_.empty()) {
relevance = CalculateRelevanceScoreUsingScoringParams(
params->matches[i], relevance, scoring_params_);
matches_.push_back(HistoryMatchToACMatch(*params, i, NORMAL, relevance));
done_ = true;
bool HistoryURLProvider::FixupExactSuggestion(
history::URLDatabase* db,
const VisitClassifier& classifier,
HistoryURLProviderParams* params) const {
switch (classifier.type()) {
case VisitClassifier::INVALID:
return false;
case VisitClassifier::UNVISITED_INTRANET:
DCHECK_EQ(VisitClassifier::VISITED, classifier.type());
// We have data for this match, use it.
params->what_you_typed_match.deletable = true;
params->what_you_typed_match.description = classifier.url_row().title();
params->what_you_typed_match.description_class = ClassifyDescription(
params->input.text(), params->what_you_typed_match.description);
if (!classifier.url_row().typed_count()) {
// If we reach here, we must be in the second pass, and we must not have
// this row's data available during the first pass. That means we
// either scored it as WHAT_YOU_TYPED or UNVISITED_INTRANET, and to
// maintain the ordering between passes consistent, we need to score it
// the same way here.
type = CanFindIntranetURL(db, params->input) ?
const GURL& url = params->what_you_typed_match.destination_url;
const url::Parsed& parsed = url.parsed_for_possibly_invalid_spec();
// If the what-you-typed result looks like a single word (which can be
// interpreted as an intranet address) followed by a pound sign ("#"),
// leave the score for the url-what-you-typed result as is. It will be
// outscored by a search query from the SearchProvider. This test fixes
// cases such as "c#" and "c# foo" where the user has visited an intranet
// site "c". We want the search-what-you-typed score to beat the
// URL-what-you-typed score in this case. Most of the below test tries to
// make sure that this code does not trigger if the user did anything to
// indicate the desired match is a URL. For instance, "c/# foo" will not
// pass the test because that will be classified as input type URL. The
// parsed.CountCharactersBefore() in the test looks for the presence of a
// reference fragment in the URL by checking whether the position differs
// included the delimiter (pound sign) versus not including the delimiter.
// (One cannot simply check url.ref() because it will not distinguish
// between the input "c" and the input "c#", both of which will have empty
// reference fragments.)
if ((type == UNVISITED_INTRANET) &&
(params->input.type() != metrics::OmniboxInputType::URL) &&
url.username().empty() && url.password().empty() && url.port().empty() &&
(url.path() == "/") && url.query().empty() &&
(parsed.CountCharactersBefore(url::Parsed::REF, true) !=
parsed.CountCharactersBefore(url::Parsed::REF, false))) {
return false;
params->what_you_typed_match.relevance = CalculateRelevance(type, 0);
// If there are any other matches, then don't promote this match here, in
// hopes the caller will be able to inline autocomplete a better suggestion.
// DoAutocomplete() will fall back on this match if inline autocompletion
// fails. This matches how we react to never-visited URL inputs in the non-
// intranet case.
if (type == UNVISITED_INTRANET && !params->matches.empty())
return false;
// Put it on the front of the HistoryMatches for redirect culling.
CreateOrPromoteMatch(classifier.url_row(), base::string16::npos, false,
&params->matches, true, true);
return true;
bool HistoryURLProvider::CanFindIntranetURL(
history::URLDatabase* db,
const AutocompleteInput& input) const {
// Normally passing the first two conditions below ought to guarantee the
// third condition, but because FixupUserInput() can run and modify the
// input's text and parts between Parse() and here, it seems better to be
// paranoid and check.
if ((input.type() != metrics::OmniboxInputType::UNKNOWN) ||
!LowerCaseEqualsASCII(input.scheme(), url::kHttpScheme) ||
return false;
const std::string host(base::UTF16ToUTF8(
const size_t registry_length =
return registry_length == 0 && db->IsTypedHost(host);
bool HistoryURLProvider::PromoteOrCreateShorterSuggestion(
history::URLDatabase* db,
bool have_what_you_typed_match,
HistoryURLProviderParams* params) {
if (params->matches.empty())
return false; // No matches, nothing to do.
// Determine the base URL from which to search, and whether that URL could
// itself be added as a match. We can add the base iff it's not "effectively
// the same" as any "what you typed" match.
const history::HistoryMatch& match = params->matches[0];
GURL search_base = ConvertToHostOnly(match, params->input.text());
bool can_add_search_base_to_matches = !have_what_you_typed_match;
if (search_base.is_empty()) {
// Search from what the user typed when we couldn't reduce the best match
// to a host. Careful: use a substring of |match| here, rather than the
// first match in |params|, because they might have different prefixes. If
// the user typed "", params->what_you_typed_match will hold
// "", but |match| might begin with
// "".
// TODO: this should be cleaned up, and is probably incorrect for IDN.
std::string new_match = match.url_info.url().possibly_invalid_spec().
substr(0, match.input_location + params->input.text().length());
search_base = GURL(new_match);
if (search_base.is_empty())
return false; // Can't construct a URL from which to start a search.
} else if (!can_add_search_base_to_matches) {
can_add_search_base_to_matches =
(search_base != params->what_you_typed_match.destination_url);
if (search_base == match.url_info.url())
return false; // Couldn't shorten |match|, so no URLs to search over.
// Search the DB for short URLs between our base and |match|.
history::URLRow info(search_base);
bool promote = true;
// A short URL is only worth suggesting if it's been visited at least a third
// as often as the longer URL.
const int min_visit_count = ((match.url_info.visit_count() - 1) / 3) + 1;
// For stability between the in-memory and on-disk autocomplete passes, when
// the long URL has been typed before, only suggest shorter URLs that have
// also been typed. Otherwise, the on-disk pass could suggest a shorter URL
// (which hasn't been typed) that the in-memory pass doesn't know about,
// thereby making the top match, and thus the behavior of inline
// autocomplete, unstable.
const int min_typed_count = match.url_info.typed_count() ? 1 : 0;
if (!db->FindShortestURLFromBase(search_base.possibly_invalid_spec(),
match.url_info.url().possibly_invalid_spec(), min_visit_count,
min_typed_count, can_add_search_base_to_matches, &info)) {
if (!can_add_search_base_to_matches)
return false; // Couldn't find anything and can't add the search base.
// Try to get info on the search base itself. Promote it to the top if the
// original best match isn't good enough to autocomplete.
db->GetRowForURL(search_base, &info);
promote = match.url_info.typed_count() <= 1;
// Promote or add the desired URL to the list of matches.
const bool ensure_can_inline =
promote && CanPromoteMatchForInlineAutocomplete(match);
return CreateOrPromoteMatch(info, match.input_location, match.match_in_scheme,
&params->matches, create_shorter_match_,
promote) &&
void HistoryURLProvider::CullPoorMatches(
HistoryURLProviderParams* params) const {
const base::Time& threshold(history::AutocompleteAgeThreshold());
for (history::HistoryMatches::iterator i(params->matches.begin());
i != params->matches.end(); ) {
if (RowQualifiesAsSignificant(i->url_info, threshold) &&
(!params->default_search_provider ||
i->url_info.url(), *params->search_terms_data))) {
} else {
i = params->matches.erase(i);
void HistoryURLProvider::CullRedirects(history::HistoryBackend* backend,
history::HistoryMatches* matches,
size_t max_results) const {
for (size_t source = 0;
(source < matches->size()) && (source < max_results); ) {
const GURL& url = (*matches)[source].url_info.url();
// TODO(brettw) this should go away when everything uses GURL.
history::RedirectList redirects;
backend->GetMostRecentRedirectsFrom(url, &redirects);
if (!redirects.empty()) {
// Remove all but the first occurrence of any of these redirects in the
// search results. We also must add the URL we queried for, since it may
// not be the first match and we'd want to remove it.
// For example, when A redirects to B and our matches are [A, X, B],
// we'll get B as the redirects from, and we want to remove the second
// item of that pair, removing B. If A redirects to B and our matches are
// [B, X, A], we'll want to remove A instead.
source = RemoveSubsequentMatchesOf(matches, source, redirects);
} else {
// Advance to next item.
if (matches->size() > max_results)
size_t HistoryURLProvider::RemoveSubsequentMatchesOf(
history::HistoryMatches* matches,
size_t source_index,
const std::vector<GURL>& remove) const {
size_t next_index = source_index + 1; // return value = item after source
// Find the first occurrence of any URL in the redirect chain. We want to
// keep this one since it is rated the highest.
history::HistoryMatches::iterator first(std::find_first_of(
matches->begin(), matches->end(), remove.begin(), remove.end(),
DCHECK(first != matches->end()) << "We should have always found at least the "
"original URL.";
// Find any following occurrences of any URL in the redirect chain, these
// should be deleted.
for (history::HistoryMatches::iterator next(std::find_first_of(first + 1,
matches->end(), remove.begin(), remove.end(),
next != matches->end(); next = std::find_first_of(next, matches->end(),
remove.begin(), remove.end(), history::HistoryMatch::EqualsGURL)) {
// Remove this item. When we remove an item before the source index, we
// need to shift it to the right and remember that so we can return it.
next = matches->erase(next);
if (static_cast<size_t>(next - matches->begin()) < next_index)
return next_index;
AutocompleteMatch HistoryURLProvider::HistoryMatchToACMatch(
const HistoryURLProviderParams& params,
size_t match_number,
MatchType match_type,
int relevance) {
// The FormattedStringWithEquivalentMeaning() call below requires callers to
// be on the UI thread.
DCHECK(content::BrowserThread::CurrentlyOn(content::BrowserThread::UI) ||
const history::HistoryMatch& history_match = params.matches[match_number];
const history::URLRow& info = history_match.url_info;
AutocompleteMatch match(this, relevance,
!!info.visit_count(), AutocompleteMatchType::HISTORY_URL);
match.typed_count = info.typed_count();
match.destination_url = info.url();
size_t inline_autocomplete_offset =
history_match.input_location + params.input.text().length();
std::string languages = (match_type == WHAT_YOU_TYPED) ?
std::string() : params.languages;
const net::FormatUrlTypes format_types = net::kFormatUrlOmitAll &
~((params.trim_http && !history_match.match_in_scheme) ?
0 : net::kFormatUrlOmitHTTP);
match.fill_into_edit =
net::FormatUrl(info.url(), languages, format_types,
net::UnescapeRule::SPACES, NULL, NULL,
if (!params.prevent_inline_autocomplete &&
(inline_autocomplete_offset != base::string16::npos)) {
DCHECK(inline_autocomplete_offset <= match.fill_into_edit.length());
match.inline_autocompletion =
// The latter part of the test effectively asks "is the inline completion
// empty?" (i.e., is this match effectively the what-you-typed match?).
match.allowed_to_be_default_match = !params.prevent_inline_autocomplete ||
((inline_autocomplete_offset != base::string16::npos) &&
(inline_autocomplete_offset >= match.fill_into_edit.length()));
size_t match_start = history_match.input_location;
match.contents = net::FormatUrl(info.url(), languages,
format_types, net::UnescapeRule::SPACES, NULL, NULL, &match_start);
if ((match_start != base::string16::npos) &&
(inline_autocomplete_offset != base::string16::npos) &&
(inline_autocomplete_offset != match_start)) {
DCHECK(inline_autocomplete_offset > match_start);
inline_autocomplete_offset - match_start, match.contents.length(),
ACMatchClassification::URL, &match.contents_class);
} else {
AutocompleteMatch::ClassifyLocationInString(base::string16::npos, 0,
match.contents.length(), ACMatchClassification::URL,
match.description = info.title();
match.description_class =
ClassifyDescription(params.input.text(), match.description);
RecordAdditionalInfoFromUrlRow(info, &match);
return match;