Add an AutofillProfileComparator class.

components/autofill/core/browser/autofill_profile_comparator.cc

+// Copyright 2016 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 "components/autofill/core/browser/autofill_profile_comparator.h"
+
+#include <algorithm>
+#include <vector>
+
+#include "base/i18n/char_iterator.h"
+#include "base/strings/string_split.h"
+#include "base/strings/string_util.h"
+#include "base/strings/utf_string_conversion_utils.h"
+#include "base/strings/utf_string_conversions.h"
+#include "components/autofill/core/browser/autofill_data_util.h"
+#include "third_party/libphonenumber/phonenumber_api.h"
+
+namespace autofill {
+namespace {
+
+const base::char16 kSpace[] = {L' ', L'\0'};
+
+}  // namespace
+
+AutofillProfileComparator::AutofillProfileComparator(
+    const base::StringPiece& app_locale)
+    : app_locale_(app_locale.data(), app_locale.size()) {
+  // Use ICU transliteration to remove diacritics and fold case.
+  // See http://userguide.icu-project.org/transforms/general
+  UErrorCode status = U_ZERO_ERROR;
+  std::unique_ptr<icu::Transliterator> transliterator(
+      icu::Transliterator::createInstance(
+          "NFD; [:Nonspacing Mark:] Remove; Lower; NFC", UTRANS_FORWARD,
+          status));
+  if (U_FAILURE(status) || transliterator == nullptr) {
+    // TODO(rogerm): Add a histogram to count how often this happens.
+    LOG(ERROR) << "Failed to create ICU Transliterator: "
+               << u_errorName(status);
+  }
+
+  transliterator_ = std::move(transliterator);
+}
+
+AutofillProfileComparator::~AutofillProfileComparator() {}
+
+base::string16 AutofillProfileComparator::NormalizeForComparison(
+    base::StringPiece16 text,
+    AutofillProfileComparator::WhitespaceSpec whitespace_spec) const {
+  // This algorithm is not designed to be perfect, we could get arbitrarily
+  // fancy here trying to canonicalize address lines. Instead, this is designed
+  // to handle common cases for all types of data (addresses and names) without
+  // the need of domain-specific logic.
+  //
+  // 1. Convert punctuation to spaces and normalize all whitespace to spaces.
+  //    This will convert "Mid-Island Plz." -> "Mid Island Plz " (the trailing
+  //    space will be trimmed off outside of the end of the loop).
+  //
+  // 2. Collapse consecutive punctuation/whitespace characters to a single
+  //    space. We pretend the string has already started with whitespace in
+  //    order to trim leading spaces.
+  //
+  // 3. Remove diacritics (accents and other non-spacing marks) and perform
+  //    case folding to lower-case.
+  base::string16 result;
+  result.reserve(text.length());
+  bool previous_was_whitespace = (whitespace_spec == RETAIN_WHITESPACE);
+  for (base::i18n::UTF16CharIterator iter(text.data(), text.length());
+       !iter.end(); iter.Advance()) {
+    switch (u_charType(iter.get())) {
+      // Punctuation
+      case U_DASH_PUNCTUATION:
+      case U_START_PUNCTUATION:
+      case U_END_PUNCTUATION:
+      case U_CONNECTOR_PUNCTUATION:
+      case U_OTHER_PUNCTUATION:
+      // Whitespace
+      case U_CONTROL_CHAR:  // To escape the '\n' character.
+      case U_SPACE_SEPARATOR:
+      case U_LINE_SEPARATOR:
+      case U_PARAGRAPH_SEPARATOR:
+        if (!previous_was_whitespace && whitespace_spec == RETAIN_WHITESPACE) {
+          result.push_back(' ');
+          previous_was_whitespace = true;
+        }
+        break;
+
+      default:
+        previous_was_whitespace = false;
+        base::WriteUnicodeCharacter(iter.get(), &result);
+        break;
+    }
+  }
+
+  // Trim off trailing whitespace if we left one.
+  if (previous_was_whitespace && !result.empty())
+    result.resize(result.size() - 1);
+
+  if (transliterator_ == nullptr)
+    return result;
+
+  icu::UnicodeString value = icu::UnicodeString(result.data(), result.length());
+  transliterator_->transliterate(value);
+  return base::string16(value.getBuffer(), value.length());
+}
+
+bool AutofillProfileComparator::AreMergeable(const AutofillProfile& p1,
+                                             const AutofillProfile& p2) const {
+  // Sorted in order to relative expense of the tests to fail early and cheaply
+  // if possible.
+  return HaveMergeableEmailAddresses(p1, p2) &&
+         HaveMergeableCompanyNames(p1, p2) &&
+         HaveMergeablePhoneNumbers(p1, p2) && HaveMergeableNames(p1, p2) &&
+         HaveMergeableAddresses(p1, p2);
+}
+
+// static
+std::set<base::StringPiece16> AutofillProfileComparator::UniqueTokens(
+    base::StringPiece16 s) {
+  std::vector<base::StringPiece16> tokens = base::SplitStringPiece(
+      s, kSpace, base::TRIM_WHITESPACE, base::SPLIT_WANT_NONEMPTY);
+  return std::set<base::StringPiece16>(tokens.begin(), tokens.end());
+}
+
+// static
+bool AutofillProfileComparator::HaveSameTokens(base::StringPiece16 s1,
+                                               base::StringPiece16 s2) {
+  std::set<base::StringPiece16> t1 = UniqueTokens(s1);
+  std::set<base::StringPiece16> t2 = UniqueTokens(s2);
+
+  // Note: std::include() expects the items in each range to be in sorted order,
+  // hence the use of std::set<> instead of std::unordered_set<>.
+  return std::includes(t1.begin(), t1.end(), t2.begin(), t2.end()) ||
+         std::includes(t2.begin(), t2.end(), t1.begin(), t1.end());
+}
+
+// static
+std::set<base::string16> AutofillProfileComparator::GetNamePartVariants(
+    const base::string16& name_part) {
+  const size_t kMaxSupportedSubNames = 8;
+
+  std::vector<base::string16> sub_names = base::SplitString(
+      name_part, kSpace, base::TRIM_WHITESPACE, base::SPLIT_WANT_NONEMPTY);
+
+  // Limit the number of sub-names we support (to constrain memory usage);
+  if (sub_names.size() > kMaxSupportedSubNames)
+    return {name_part};
+
+  // Start with the empty string as a variant.
+  std::set<base::string16> variants = {base::EmptyString16()};
+
+  // For each sub-name, add a variant of all the already existing variants that
+  // appends this sub-name and one that appends the initial of this sub-name.
+  // Duplicates will be discarded when they're added to the variants set.
+  for (const base::string16& sub_name : sub_names) {
+    if (sub_name.empty()) continue;
+    std::vector<base::string16> new_variants;
+    for (const base::string16& variant : variants) {
+      new_variants.push_back(base::CollapseWhitespace(
+          base::JoinString({variant, sub_name}, kSpace), true));
+      new_variants.push_back(base::CollapseWhitespace(
+          base::JoinString({variant, sub_name.substr(0, 1)}, kSpace), true));
+    }
+    variants.insert(new_variants.begin(), new_variants.end());
+  }
+
+  // As a common case, also add the variant that just concatenates all of the
+  // initials.
+  base::string16 initials;
+  for (const base::string16& sub_name : sub_names) {
+    if (sub_name.empty()) continue;
+    initials.push_back(sub_name[0]);
+  }
+  variants.insert(initials);
+
+  // And, we're done.
+  return variants;
+}
+
+bool AutofillProfileComparator::IsNameVariantOf(
+    const base::string16& full_name_1,
+    const base::string16& full_name_2) const {
+  data_util::NameParts name_1_parts = data_util::SplitName(full_name_1);
+
+  // Build the variants of full_name_1`s given, middle and family names.
+  //
+  // TODO(rogerm): Figure out whether or not we should break apart a compound
+  // family name into variants (crbug/619051)
+  const std::set<base::string16> given_name_variants =
+      GetNamePartVariants(name_1_parts.given);
+  const std::set<base::string16> middle_name_variants =
+      GetNamePartVariants(name_1_parts.middle);
+  const base::string16& family_name = name_1_parts.family;
+
+  // Iterate over all full name variants of profile 2 and see if any of them
+  // match the full name from profile 1.
+  for (const base::string16& given_name : given_name_variants) {
+    for (const base::string16& middle_name : middle_name_variants) {
+      base::string16 candidate = base::CollapseWhitespace(
+          base::JoinString({given_name, middle_name, family_name}, kSpace),
+          true);
+      if (candidate == full_name_2)
+        return true;
+    }
+  }
+
+  // Also check if the name is just composed of the user's initials. For
+  // example, "thomas jefferson miller" could be composed as "tj miller".
+  if (!name_1_parts.given.empty() && !name_1_parts.middle.empty()) {
+    base::string16 initials;
+    initials.push_back(name_1_parts.given[0]);
+    initials.push_back(name_1_parts.middle[0]);
+    base::string16 candidate = base::CollapseWhitespace(
+        base::JoinString({initials, family_name}, kSpace), true);
+    if (candidate == full_name_2)
+      return true;
+  }
+
+  // There was no match found.
+  return false;
+}
+
+bool AutofillProfileComparator::HaveMergeableNames(
+    const AutofillProfile& p1,
+    const AutofillProfile& p2) const {
+  base::string16 full_name_1 =
+      NormalizeForComparison(p1.GetInfo(AutofillType(NAME_FULL), app_locale_));
+  base::string16 full_name_2 =
+      NormalizeForComparison(p2.GetInfo(AutofillType(NAME_FULL), app_locale_));
+
+  // Is it reasonable to merge the names from p1 and p2.
+  return full_name_1.empty() || full_name_2.empty() ||
+         (full_name_1 == full_name_2) ||
+         IsNameVariantOf(full_name_1, full_name_2) ||
+         IsNameVariantOf(full_name_2, full_name_1);
+}
+
+bool AutofillProfileComparator::HaveMergeableEmailAddresses(
+    const AutofillProfile& p1,
+    const AutofillProfile& p2) const {
+  const base::string16& email_1 =
+      p1.GetInfo(AutofillType(EMAIL_ADDRESS), app_locale_);
+  const base::string16& email_2 =
+      p2.GetInfo(AutofillType(EMAIL_ADDRESS), app_locale_);
+  return email_1.empty() || email_2.empty() ||
+         case_insensitive_compare_.StringsEqual(email_1, email_2);
+}
+
+bool AutofillProfileComparator::HaveMergeableCompanyNames(
+    const AutofillProfile& p1,
+    const AutofillProfile& p2) const {
+  const base::string16& company_name_1 = NormalizeForComparison(
+      p1.GetInfo(AutofillType(COMPANY_NAME), app_locale_));
+  const base::string16& company_name_2 = NormalizeForComparison(
+      p2.GetInfo(AutofillType(COMPANY_NAME), app_locale_));
+  return company_name_1.empty() || company_name_2.empty() ||
+         HaveSameTokens(company_name_1, company_name_2);
+}
+
+bool AutofillProfileComparator::HaveMergeablePhoneNumbers(
+    const AutofillProfile& p1,
+    const AutofillProfile& p2) const {
+  // We work with the raw phone numbers to avoid losing any helpful information
+  // as we parse.
+  const base::string16& raw_phone_1 = p1.GetRawInfo(PHONE_HOME_WHOLE_NUMBER);
+  const base::string16& raw_phone_2 = p2.GetRawInfo(PHONE_HOME_WHOLE_NUMBER);
+
+  // Are the two phone numbers trivially mergeable?
+  if (raw_phone_1.empty() || raw_phone_2.empty() ||
+      raw_phone_1 == raw_phone_2) {
+    return true;
+  }
+
+  // TODO(rogerm): Modify ::autofill::i18n::PhoneNumbersMatch to support
+  // SHORT_NSN_MATCH and just call that instead of accessing the underlying
+  // utility library directly?
+
+  // The phone number util library needs the numbers in utf8.
+  const std::string phone_1 = base::UTF16ToUTF8(raw_phone_1);
+  const std::string phone_2 = base::UTF16ToUTF8(raw_phone_2);
+
+  // Parse and compare the phone numbers.
+  using ::i18n::phonenumbers::PhoneNumberUtil;
+  PhoneNumberUtil* phone_util = PhoneNumberUtil::GetInstance();
+  switch (phone_util->IsNumberMatchWithTwoStrings(phone_1, phone_2)) {
+    case PhoneNumberUtil::INVALID_NUMBER:
+    case PhoneNumberUtil::NO_MATCH:
+      return false;
+    case PhoneNumberUtil::SHORT_NSN_MATCH:
+    case PhoneNumberUtil::NSN_MATCH:
+    case PhoneNumberUtil::EXACT_MATCH:
+      return true;
+  }
+
+  NOTREACHED();
+  return false;
+}
+
+bool AutofillProfileComparator::HaveMergeableAddresses(
+    const AutofillProfile& p1,
+    const AutofillProfile& p2) const {
+  // If the address are not in the same country, then they're not the same. If
+  // one of the address countries is unknown/invalid the comparison continues.
+  const base::string16& country1 = p1.GetInfo(
+      AutofillType(HTML_TYPE_COUNTRY_CODE, HTML_MODE_NONE), app_locale_);
+  const base::string16& country2 = p2.GetInfo(
+      AutofillType(HTML_TYPE_COUNTRY_CODE, HTML_MODE_NONE), app_locale_);
+  if (!country1.empty() && !country2.empty() &&
+      !case_insensitive_compare_.StringsEqual(country1, country2)) {
+    return false;
+  }
+
+  // TODO(rogerm): Lookup the normalization rules for the (common) country of
+  // the address. The rules should be applied post NormalizeForComparison to
+  // the state, city, and address bag of words comparisons.
+
+  // Zip
+  // ----
+  // If the addresses are definitely not in the same zip/area code then we're
+  // done. Otherwise,the comparison continues.
+  const base::string16& zip1 = NormalizeForComparison(
+      p1.GetInfo(AutofillType(ADDRESS_HOME_ZIP), app_locale_),
+      DISCARD_WHITESPACE);
+  const base::string16& zip2 = NormalizeForComparison(
+      p2.GetInfo(AutofillType(ADDRESS_HOME_ZIP), app_locale_),
+      DISCARD_WHITESPACE);
+  if (!zip1.empty() && !zip2.empty() &&
+      zip1.find(zip2) == base::string16::npos &&
+      zip2.find(zip1) == base::string16::npos) {
+    return false;
+  }
+
+  // State
+  // ------
+  // Heuristic: If the match is between non-empty zip codes then we can infer
+  // that the two state strings are intended to have the same meaning. This
+  // handles the cases where we have invalid or poorly formed data in one of the
+  // state values (like "Select one", or "CA - California"). Otherwise, we
+  // actually have to check if the states map to the the same set of tokens.
+  const base::string16& state1 = NormalizeForComparison(
+      p1.GetInfo(AutofillType(ADDRESS_HOME_STATE), app_locale_));
+  const base::string16& state2 = NormalizeForComparison(
+      p2.GetInfo(AutofillType(ADDRESS_HOME_STATE), app_locale_));
+  if ((zip1.empty() || zip2.empty()) && !HaveSameTokens(state1, state2)) {
+    return false;
+  }
+
+  // City
+  // ------
+  // Heuristic: If the match is between non-empty zip codes then we can infer
+  // that the two city strings are intended to have the same meaning. This
+  // handles the cases where we have a city vs one of its suburbs. Otherwise, we
+  // actually have to check if the cities map to the the same set of tokens.
+  const base::string16& city1 = NormalizeForComparison(
+      p1.GetInfo(AutofillType(ADDRESS_HOME_CITY), app_locale_));
+  const base::string16& city2 = NormalizeForComparison(
+      p2.GetInfo(AutofillType(ADDRESS_HOME_CITY), app_locale_));
+  if ((zip1.empty() || zip2.empty()) && !HaveSameTokens(city1, city2)) {
+    return false;
+  }
+
+  // Address
+  // --------
+  // Heuristic: Use bag of words comparison on the post-normalized addresses.
+  const base::string16& address1 = NormalizeForComparison(
+      p1.GetInfo(AutofillType(ADDRESS_HOME_STREET_ADDRESS), app_locale_));
+  const base::string16& address2 = NormalizeForComparison(
+      p2.GetInfo(AutofillType(ADDRESS_HOME_STREET_ADDRESS), app_locale_));
+  if (!HaveSameTokens(address1, address2)) {
+    return false;
+  }
+
+  return true;
+}
+
+}  // namespace autofill