Index: content/common/android/address_parser_internal.cc |
diff --git a/content/common/android/address_parser_internal.cc b/content/common/android/address_parser_internal.cc |
deleted file mode 100644 |
index 9892bd853fb393c2aa9ef8d1e201555e1c9ccff3..0000000000000000000000000000000000000000 |
--- a/content/common/android/address_parser_internal.cc |
+++ /dev/null |
@@ -1,640 +0,0 @@ |
-// 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 "content/common/android/address_parser_internal.h" |
- |
-#include <bitset> |
- |
-#include "base/logging.h" |
-#include "base/macros.h" |
-#include "base/strings/string_util.h" |
- |
-namespace { |
- |
-// Number of digits for a valid zip code. |
-const size_t kZipDigits = 5; |
- |
-// Number of digits for a valid zip code in the Zip Plus 4 format. |
-const size_t kZipPlus4Digits = 9; |
- |
-// Maximum number of digits of a house number, including possible hyphens. |
-const size_t kMaxHouseDigits = 5; |
- |
-base::char16 SafePreviousChar(const base::string16::const_iterator& it, |
- const base::string16::const_iterator& begin) { |
- if (it == begin) |
- return ' '; |
- return *(it - 1); |
-} |
- |
-base::char16 SafeNextChar(const base::string16::const_iterator& it, |
- const base::string16::const_iterator& end) { |
- if (it == end) |
- return ' '; |
- return *(it + 1); |
-} |
- |
-bool WordLowerCaseEqualsASCII(base::string16::const_iterator word_begin, |
- base::string16::const_iterator word_end, const char* ascii_to_match) { |
- for (base::string16::const_iterator it = word_begin; it != word_end; |
- ++it, ++ascii_to_match) { |
- if (!*ascii_to_match || base::ToLowerASCII(*it) != *ascii_to_match) |
- return false; |
- } |
- return *ascii_to_match == 0 || *ascii_to_match == ' '; |
-} |
- |
-bool LowerCaseEqualsASCIIWithPlural(base::string16::const_iterator word_begin, |
- base::string16::const_iterator word_end, const char* ascii_to_match, |
- bool allow_plural) { |
- for (base::string16::const_iterator it = word_begin; it != word_end; |
- ++it, ++ascii_to_match) { |
- if (!*ascii_to_match && allow_plural && *it == 's' && it + 1 == word_end) |
- return true; |
- |
- if (!*ascii_to_match || base::ToLowerASCII(*it) != *ascii_to_match) |
- return false; |
- } |
- return *ascii_to_match == 0; |
-} |
- |
-} // anonymous namespace |
- |
-namespace content { |
- |
-namespace address_parser { |
- |
-namespace internal { |
- |
-Word::Word() { |
-} |
- |
-Word::Word(const base::string16::const_iterator& begin, |
- const base::string16::const_iterator& end) |
- : begin(begin), |
- end(end) { |
- DCHECK(begin <= end); |
-} |
- |
-Word::Word(const Word& other) = default; |
- |
-HouseNumberParser::HouseNumberParser() { |
-} |
- |
-bool HouseNumberParser::IsPreDelimiter(base::char16 character) { |
- return character == ':' || IsPostDelimiter(character); |
-} |
- |
-bool HouseNumberParser::IsPostDelimiter(base::char16 character) { |
- return base::IsUnicodeWhitespace(character) || strchr(",\"'", character); |
-} |
- |
-void HouseNumberParser::RestartOnNextDelimiter() { |
- ResetState(); |
- for (; it_ != end_ && !IsPreDelimiter(*it_); ++it_) {} |
-} |
- |
-void HouseNumberParser::AcceptChars(size_t num_chars) { |
- size_t offset = std::min(static_cast<size_t>(std::distance(it_, end_)), |
- num_chars); |
- it_ += offset; |
- result_chars_ += offset; |
-} |
- |
-void HouseNumberParser::SkipChars(size_t num_chars) { |
- it_ += std::min(static_cast<size_t>(std::distance(it_, end_)), num_chars); |
-} |
- |
-void HouseNumberParser::ResetState() { |
- num_digits_ = 0; |
- result_chars_ = 0; |
-} |
- |
-bool HouseNumberParser::CheckFinished(Word* word) const { |
- // There should always be a number after a hyphen. |
- if (result_chars_ == 0 || SafePreviousChar(it_, begin_) == '-') |
- return false; |
- |
- if (word) { |
- word->begin = it_ - result_chars_; |
- word->end = it_; |
- } |
- return true; |
-} |
- |
-bool HouseNumberParser::Parse( |
- const base::string16::const_iterator& begin, |
- const base::string16::const_iterator& end, Word* word) { |
- it_ = begin_ = begin; |
- end_ = end; |
- ResetState(); |
- |
- // Iterations only used as a fail-safe against any buggy infinite loops. |
- size_t iterations = 0; |
- size_t max_iterations = end - begin + 1; |
- for (; it_ != end_ && iterations < max_iterations; ++iterations) { |
- |
- // Word finished case. |
- if (IsPostDelimiter(*it_)) { |
- if (CheckFinished(word)) |
- return true; |
- else if (result_chars_) |
- ResetState(); |
- |
- SkipChars(1); |
- continue; |
- } |
- |
- // More digits. There should be no more after a letter was found. |
- if (base::IsAsciiDigit(*it_)) { |
- if (num_digits_ >= kMaxHouseDigits) { |
- RestartOnNextDelimiter(); |
- } else { |
- AcceptChars(1); |
- ++num_digits_; |
- } |
- continue; |
- } |
- |
- if (base::IsAsciiAlpha(*it_)) { |
- // Handle special case 'one'. |
- if (result_chars_ == 0) { |
- if (it_ + 3 <= end_ && |
- base::LowerCaseEqualsASCII(base::StringPiece16(it_, it_ + 3), |
- "one")) |
- AcceptChars(3); |
- else |
- RestartOnNextDelimiter(); |
- continue; |
- } |
- |
- // There should be more than 1 character because of result_chars. |
- DCHECK_GT(result_chars_, 0U); |
- DCHECK(it_ != begin_); |
- base::char16 previous = SafePreviousChar(it_, begin_); |
- if (base::IsAsciiDigit(previous)) { |
- // Check cases like '12A'. |
- base::char16 next = SafeNextChar(it_, end_); |
- if (IsPostDelimiter(next)) { |
- AcceptChars(1); |
- continue; |
- } |
- |
- // Handle cases like 12a, 1st, 2nd, 3rd, 7th. |
- if (base::IsAsciiAlpha(next)) { |
- base::char16 last_digit = previous; |
- base::char16 first_letter = base::ToLowerASCII(*it_); |
- base::char16 second_letter = base::ToLowerASCII(next); |
- bool is_teen = SafePreviousChar(it_ - 1, begin_) == '1' && |
- num_digits_ == 2; |
- |
- switch (last_digit - '0') { |
- case 1: |
- if ((first_letter == 's' && second_letter == 't') || |
- (first_letter == 't' && second_letter == 'h' && is_teen)) { |
- AcceptChars(2); |
- continue; |
- } |
- break; |
- |
- case 2: |
- if ((first_letter == 'n' && second_letter == 'd') || |
- (first_letter == 't' && second_letter == 'h' && is_teen)) { |
- AcceptChars(2); |
- continue; |
- } |
- break; |
- |
- case 3: |
- if ((first_letter == 'r' && second_letter == 'd') || |
- (first_letter == 't' && second_letter == 'h' && is_teen)) { |
- AcceptChars(2); |
- continue; |
- } |
- break; |
- |
- case 0: |
- // Explicitly exclude '0th'. |
- if (num_digits_ == 1) |
- break; |
- |
- case 4: |
- case 5: |
- case 6: |
- case 7: |
- case 8: |
- case 9: |
- if (first_letter == 't' && second_letter == 'h') { |
- AcceptChars(2); |
- continue; |
- } |
- break; |
- |
- default: |
- NOTREACHED(); |
- } |
- } |
- } |
- |
- RestartOnNextDelimiter(); |
- continue; |
- } |
- |
- if (*it_ == '-' && num_digits_ > 0) { |
- AcceptChars(1); |
- ++num_digits_; |
- continue; |
- } |
- |
- RestartOnNextDelimiter(); |
- SkipChars(1); |
- } |
- |
- if (iterations >= max_iterations) |
- return false; |
- |
- return CheckFinished(word); |
-} |
- |
-bool FindStateStartingInWord(WordList* words, |
- size_t state_first_word, |
- size_t* state_last_word, |
- String16Tokenizer* tokenizer, |
- size_t* state_index) { |
- |
- // Bitmasks containing the allowed suffixes for 2-letter state codes. |
- static const int state_two_letter_suffix[23] = { |
- 0x02060c00, // A followed by: [KLRSZ]. |
- 0x00000000, // B. |
- 0x00084001, // C followed by: [AOT]. |
- 0x00000014, // D followed by: [CE]. |
- 0x00000000, // E. |
- 0x00001800, // F followed by: [LM]. |
- 0x00100001, // G followed by: [AU]. |
- 0x00000100, // H followed by: [I]. |
- 0x00002809, // I followed by: [ADLN]. |
- 0x00000000, // J. |
- 0x01040000, // K followed by: [SY]. |
- 0x00000001, // L followed by: [A]. |
- 0x000ce199, // M followed by: [ADEHINOPST]. |
- 0x0120129c, // N followed by: [CDEHJMVY]. |
- 0x00020480, // O followed by: [HKR]. |
- 0x00420001, // P followed by: [ARW]. |
- 0x00000000, // Q. |
- 0x00000100, // R followed by: [I]. |
- 0x0000000c, // S followed by: [CD]. |
- 0x00802000, // T followed by: [NX]. |
- 0x00080000, // U followed by: [T]. |
- 0x00080101, // V followed by: [AIT]. |
- 0x01200101 // W followed by: [AIVY]. |
- }; |
- |
- // Accumulative number of states for the 2-letter code indexed by the first. |
- static const int state_two_letter_accumulative[24] = { |
- 0, 5, 5, 8, 10, 10, 12, 14, |
- 15, 19, 19, 21, 22, 32, 40, 43, |
- 46, 46, 47, 49, 51, 52, 55, 59 |
- }; |
- |
- // State names sorted alphabetically with their lengths. |
- // There can be more than one possible name for a same state if desired. |
- static const struct StateNameInfo { |
- const char* string; |
- char first_word_length; |
- char length; |
- char state_index; // Relative to two-character code alphabetical order. |
- } state_names[59] = { |
- { "alabama", 7, 7, 1 }, { "alaska", 6, 6, 0 }, |
- { "american samoa", 8, 14, 3 }, { "arizona", 7, 7, 4 }, |
- { "arkansas", 8, 8, 2 }, |
- { "california", 10, 10, 5 }, { "colorado", 8, 8, 6 }, |
- { "connecticut", 11, 11, 7 }, { "delaware", 8, 8, 9 }, |
- { "district of columbia", 8, 20, 8 }, |
- { "federated states of micronesia", 9, 30, 11 }, { "florida", 7, 7, 10 }, |
- { "guam", 4, 4, 13 }, { "georgia", 7, 7, 12 }, |
- { "hawaii", 6, 6, 14 }, |
- { "idaho", 5, 5, 16 }, { "illinois", 8, 8, 17 }, { "indiana", 7, 7, 18 }, |
- { "iowa", 4, 4, 15 }, |
- { "kansas", 6, 6, 19 }, { "kentucky", 8, 8, 20 }, |
- { "louisiana", 9, 9, 21 }, |
- { "maine", 5, 5, 24 }, { "marshall islands", 8, 16, 25 }, |
- { "maryland", 8, 8, 23 }, { "massachusetts", 13, 13, 22 }, |
- { "michigan", 8, 8, 26 }, { "minnesota", 9, 9, 27 }, |
- { "mississippi", 11, 11, 30 }, { "missouri", 8, 8, 28 }, |
- { "montana", 7, 7, 31 }, |
- { "nebraska", 8, 8, 34 }, { "nevada", 6, 6, 38 }, |
- { "new hampshire", 3, 13, 35 }, { "new jersey", 3, 10, 36 }, |
- { "new mexico", 3, 10, 37 }, { "new york", 3, 8, 39 }, |
- { "north carolina", 5, 14, 32 }, { "north dakota", 5, 12, 33 }, |
- { "northern mariana islands", 8, 24, 29 }, |
- { "ohio", 4, 4, 40 }, { "oklahoma", 8, 8, 41 }, { "oregon", 6, 6, 42 }, |
- { "palau", 5, 5, 45 }, { "pennsylvania", 12, 12, 43 }, |
- { "puerto rico", 6, 11, 44 }, |
- { "rhode island", 5, 5, 46 }, |
- { "south carolina", 5, 14, 47 }, { "south dakota", 5, 12, 48 }, |
- { "tennessee", 9, 9, 49 }, { "texas", 5, 5, 50 }, |
- { "utah", 4, 4, 51 }, |
- { "vermont", 7, 7, 54 }, { "virgin islands", 6, 14, 53 }, |
- { "virginia", 8, 8, 52 }, |
- { "washington", 10, 10, 55 }, { "west virginia", 4, 13, 57 }, |
- { "wisconsin", 9, 9, 56 }, { "wyoming", 7, 7, 58 } |
- }; |
- |
- // Accumulative number of states for sorted names indexed by the first letter. |
- // Required a different one since there are codes that don't share their |
- // first letter with the name of their state (MP = Northern Mariana Islands). |
- static const int state_names_accumulative[24] = { |
- 0, 5, 5, 8, 10, 10, 12, 14, |
- 15, 19, 19, 21, 22, 31, 40, 43, |
- 46, 46, 47, 49, 51, 52, 55, 59 |
- }; |
- |
- DCHECK_EQ(state_names_accumulative[arraysize(state_names_accumulative) - 1], |
- static_cast<int>(arraysize(state_names))); |
- |
- const Word& first_word = words->at(state_first_word); |
- int length = first_word.end - first_word.begin; |
- if (length < 2 || !base::IsAsciiAlpha(*first_word.begin)) |
- return false; |
- |
- // No state names start with x, y, z. |
- base::char16 first_letter = base::ToLowerASCII(*first_word.begin); |
- if (first_letter > 'w') |
- return false; |
- |
- DCHECK(first_letter >= 'a'); |
- int first_index = first_letter - 'a'; |
- |
- // Look for two-letter state names. |
- if (length == 2 && base::IsAsciiAlpha(*(first_word.begin + 1))) { |
- base::char16 second_letter = base::ToLowerASCII(*(first_word.begin + 1)); |
- DCHECK(second_letter >= 'a'); |
- |
- int second_index = second_letter - 'a'; |
- if (!(state_two_letter_suffix[first_index] & (1 << second_index))) |
- return false; |
- |
- std::bitset<32> previous_suffixes = state_two_letter_suffix[first_index] & |
- ((1 << second_index) - 1); |
- *state_last_word = state_first_word; |
- *state_index = state_two_letter_accumulative[first_index] + |
- previous_suffixes.count(); |
- return true; |
- } |
- |
- // Look for full state names by their first letter. Discard by length. |
- for (int state = state_names_accumulative[first_index]; |
- state < state_names_accumulative[first_index + 1]; ++state) { |
- if (state_names[state].first_word_length != length) |
- continue; |
- |
- bool state_match = false; |
- size_t state_word = state_first_word; |
- for (int pos = 0; true; ) { |
- if (!WordLowerCaseEqualsASCII(words->at(state_word).begin, |
- words->at(state_word).end, &state_names[state].string[pos])) |
- break; |
- |
- pos += words->at(state_word).end - words->at(state_word).begin + 1; |
- if (pos >= state_names[state].length) { |
- state_match = true; |
- break; |
- } |
- |
- // Ran out of words, extract more from the tokenizer. |
- if (++state_word == words->size()) { |
- do { |
- if (!tokenizer->GetNext()) |
- break; |
- } while (tokenizer->token_is_delim()); |
- words->push_back( |
- Word(tokenizer->token_begin(), tokenizer->token_end())); |
- } |
- } |
- |
- if (state_match) { |
- *state_last_word = state_word; |
- *state_index = state_names[state].state_index; |
- return true; |
- } |
- } |
- |
- return false; |
-} |
- |
-bool IsZipValid(const Word& word, size_t state_index) { |
- size_t length = word.end - word.begin; |
- if (length != kZipDigits && length != kZipPlus4Digits + 1) |
- return false; |
- |
- for (base::string16::const_iterator it = word.begin; it != word.end; ++it) { |
- size_t pos = it - word.begin; |
- if (base::IsAsciiDigit(*it) || (*it == '-' && pos == kZipDigits)) |
- continue; |
- return false; |
- } |
- return IsZipValidForState(word, state_index); |
-} |
- |
-bool IsZipValidForState(const Word& word, size_t state_index) { |
- // List of valid zip code ranges. |
- static const struct { |
- signed char low; |
- signed char high; |
- signed char exception1; |
- signed char exception2; |
- } zip_range[] = { |
- { 99, 99, -1, -1 }, // AK Alaska. |
- { 35, 36, -1, -1 }, // AL Alabama. |
- { 71, 72, -1, -1 }, // AR Arkansas. |
- { 96, 96, -1, -1 }, // AS American Samoa. |
- { 85, 86, -1, -1 }, // AZ Arizona. |
- { 90, 96, -1, -1 }, // CA California. |
- { 80, 81, -1, -1 }, // CO Colorado. |
- { 6, 6, -1, -1 }, // CT Connecticut. |
- { 20, 20, -1, -1 }, // DC District of Columbia. |
- { 19, 19, -1, -1 }, // DE Delaware. |
- { 32, 34, -1, -1 }, // FL Florida. |
- { 96, 96, -1, -1 }, // FM Federated States of Micronesia. |
- { 30, 31, -1, -1 }, // GA Georgia. |
- { 96, 96, -1, -1 }, // GU Guam. |
- { 96, 96, -1, -1 }, // HI Hawaii. |
- { 50, 52, -1, -1 }, // IA Iowa. |
- { 83, 83, -1, -1 }, // ID Idaho. |
- { 60, 62, -1, -1 }, // IL Illinois. |
- { 46, 47, -1, -1 }, // IN Indiana. |
- { 66, 67, 73, -1 }, // KS Kansas. |
- { 40, 42, -1, -1 }, // KY Kentucky. |
- { 70, 71, -1, -1 }, // LA Louisiana. |
- { 1, 2, -1, -1 }, // MA Massachusetts. |
- { 20, 21, -1, -1 }, // MD Maryland. |
- { 3, 4, -1, -1 }, // ME Maine. |
- { 96, 96, -1, -1 }, // MH Marshall Islands. |
- { 48, 49, -1, -1 }, // MI Michigan. |
- { 55, 56, -1, -1 }, // MN Minnesota. |
- { 63, 65, -1, -1 }, // MO Missouri. |
- { 96, 96, -1, -1 }, // MP Northern Mariana Islands. |
- { 38, 39, -1, -1 }, // MS Mississippi. |
- { 55, 56, -1, -1 }, // MT Montana. |
- { 27, 28, -1, -1 }, // NC North Carolina. |
- { 58, 58, -1, -1 }, // ND North Dakota. |
- { 68, 69, -1, -1 }, // NE Nebraska. |
- { 3, 4, -1, -1 }, // NH New Hampshire. |
- { 7, 8, -1, -1 }, // NJ New Jersey. |
- { 87, 88, 86, -1 }, // NM New Mexico. |
- { 88, 89, 96, -1 }, // NV Nevada. |
- { 10, 14, 0, 6 }, // NY New York. |
- { 43, 45, -1, -1 }, // OH Ohio. |
- { 73, 74, -1, -1 }, // OK Oklahoma. |
- { 97, 97, -1, -1 }, // OR Oregon. |
- { 15, 19, -1, -1 }, // PA Pennsylvania. |
- { 6, 6, 0, 9 }, // PR Puerto Rico. |
- { 96, 96, -1, -1 }, // PW Palau. |
- { 2, 2, -1, -1 }, // RI Rhode Island. |
- { 29, 29, -1, -1 }, // SC South Carolina. |
- { 57, 57, -1, -1 }, // SD South Dakota. |
- { 37, 38, -1, -1 }, // TN Tennessee. |
- { 75, 79, 87, 88 }, // TX Texas. |
- { 84, 84, -1, -1 }, // UT Utah. |
- { 22, 24, 20, -1 }, // VA Virginia. |
- { 6, 9, -1, -1 }, // VI Virgin Islands. |
- { 5, 5, -1, -1 }, // VT Vermont. |
- { 98, 99, -1, -1 }, // WA Washington. |
- { 53, 54, -1, -1 }, // WI Wisconsin. |
- { 24, 26, -1, -1 }, // WV West Virginia. |
- { 82, 83, -1, -1 } // WY Wyoming. |
- }; |
- |
- // Zip numeric value for the first two characters. |
- DCHECK(word.begin != word.end); |
- DCHECK(base::IsAsciiDigit(*word.begin)); |
- DCHECK(base::IsAsciiDigit(*(word.begin + 1))); |
- int zip_prefix = (*word.begin - '0') * 10 + (*(word.begin + 1) - '0'); |
- |
- if ((zip_prefix >= zip_range[state_index].low && |
- zip_prefix <= zip_range[state_index].high) || |
- zip_prefix == zip_range[state_index].exception1 || |
- zip_prefix == zip_range[state_index].exception2) { |
- return true; |
- } |
- return false; |
-} |
- |
-bool IsValidLocationName(const Word& word) { |
- // Supported location names sorted alphabetically and grouped by first letter. |
- static const struct LocationNameInfo { |
- const char* string; |
- char length; |
- bool allow_plural; |
- } location_names[159] = { |
- { "alley", 5, false }, { "annex", 5, false }, { "arcade", 6, false }, |
- { "ave", 3, false }, { "ave.", 4, false }, { "avenue", 6, false }, |
- { "alameda", 7, false }, |
- { "bayou", 5, false }, { "beach", 5, false }, { "bend", 4, false }, |
- { "bluff", 5, true }, { "bottom", 6, false }, { "boulevard", 9, false }, |
- { "branch", 6, false }, { "bridge", 6, false }, { "brook", 5, true }, |
- { "burg", 4, true }, { "bypass", 6, false }, { "broadway", 8, false }, |
- { "camino", 6, false }, { "camp", 4, false }, { "canyon", 6, false }, |
- { "cape", 4, false }, { "causeway", 8, false }, { "center", 6, true }, |
- { "circle", 6, true }, { "cliff", 5, true }, { "club", 4, false }, |
- { "common", 6, false }, { "corner", 6, true }, { "course", 6, false }, |
- { "court", 5, true }, { "cove", 4, true }, { "creek", 5, false }, |
- { "crescent", 8, false }, { "crest", 5, false }, { "crossing", 8, false }, |
- { "crossroad", 9, false }, { "curve", 5, false }, { "circulo", 7, false }, |
- { "dale", 4, false }, { "dam", 3, false }, { "divide", 6, false }, |
- { "drive", 5, true }, |
- { "estate", 6, true }, { "expressway", 10, false }, |
- { "extension", 9, true }, |
- { "fall", 4, true }, { "ferry", 5, false }, { "field", 5, true }, |
- { "flat", 4, true }, { "ford", 4, true }, { "forest", 6, false }, |
- { "forge", 5, true }, { "fork", 4, true }, { "fort", 4, false }, |
- { "freeway", 7, false }, |
- { "garden", 6, true }, { "gateway", 7, false }, { "glen", 4, true }, |
- { "green", 5, true }, { "grove", 5, true }, |
- { "harbor", 6, true }, { "haven", 5, false }, { "heights", 7, false }, |
- { "highway", 7, false }, { "hill", 4, true }, { "hollow", 6, false }, |
- { "inlet", 5, false }, { "island", 6, true }, { "isle", 4, false }, |
- { "junction", 8, true }, |
- { "key", 3, true }, { "knoll", 5, true }, |
- { "lake", 4, true }, { "land", 4, false }, { "landing", 7, false }, |
- { "lane", 4, false }, { "light", 5, true }, { "loaf", 4, false }, |
- { "lock", 4, true }, { "lodge", 5, false }, { "loop", 4, false }, |
- { "mall", 4, false }, { "manor", 5, true }, { "meadow", 6, true }, |
- { "mews", 4, false }, { "mill", 4, true }, { "mission", 7, false }, |
- { "motorway", 8, false }, { "mount", 5, false }, { "mountain", 8, true }, |
- { "neck", 4, false }, |
- { "orchard", 7, false }, { "oval", 4, false }, { "overpass", 8, false }, |
- { "park", 4, true }, { "parkway", 7, true }, { "pass", 4, false }, |
- { "passage", 7, false }, { "path", 4, false }, { "pike", 4, false }, |
- { "pine", 4, true }, { "plain", 5, true }, { "plaza", 5, false }, |
- { "point", 5, true }, { "port", 4, true }, { "prairie", 7, false }, |
- { "privada", 7, false }, |
- { "radial", 6, false }, { "ramp", 4, false }, { "ranch", 5, false }, |
- { "rapid", 5, true }, { "rd", 2, false }, { "rd.", 3, false }, |
- { "rest", 4, false }, { "ridge", 5, true }, { "river", 5, false }, |
- { "road", 4, true }, { "route", 5, false }, { "row", 3, false }, |
- { "rue", 3, false }, { "run", 3, false }, |
- { "shoal", 5, true }, { "shore", 5, true }, { "skyway", 6, false }, |
- { "spring", 6, true }, { "spur", 4, true }, { "square", 6, true }, |
- { "station", 7, false }, { "stravenue", 9, false }, { "stream", 6, false }, |
- { "st", 2, false }, { "st.", 3, false }, { "street", 6, true }, |
- { "summit", 6, false }, { "speedway", 8, false }, |
- { "terrace", 7, false }, { "throughway", 10, false }, { "trace", 5, false }, |
- { "track", 5, false }, { "trafficway", 10, false }, { "trail", 5, false }, |
- { "tunnel", 6, false }, { "turnpike", 8, false }, |
- { "underpass", 9, false }, { "union", 5, true }, |
- { "valley", 6, true }, { "viaduct", 7, false }, { "view", 4, true }, |
- { "village", 7, true }, { "ville", 5, false }, { "vista", 5, false }, |
- { "walk", 4, true }, { "wall", 4, false }, { "way", 3, true }, |
- { "well", 4, true }, |
- { "xing", 4, false }, { "xrd", 3, false } |
- }; |
- |
- // Accumulative number of location names for each starting letter. |
- static const int location_names_accumulative[25] = { |
- 0, 7, 19, 40, 44, |
- 47, 57, 62, 68, 71, |
- 72, 74, 83, 92, 93, |
- 96, 109, 109, 123, 137, |
- 145, 147, 153, 157, 159 |
- }; |
- |
- DCHECK_EQ( |
- location_names_accumulative[arraysize(location_names_accumulative) - 1], |
- static_cast<int>(arraysize(location_names))); |
- |
- if (!base::IsAsciiAlpha(*word.begin)) |
- return false; |
- |
- // No location names start with y, z. |
- base::char16 first_letter = base::ToLowerASCII(*word.begin); |
- if (first_letter > 'x') |
- return false; |
- |
- DCHECK(first_letter >= 'a'); |
- int index = first_letter - 'a'; |
- int length = std::distance(word.begin, word.end); |
- for (int i = location_names_accumulative[index]; |
- i < location_names_accumulative[index + 1]; ++i) { |
- if (location_names[i].length != length && |
- (location_names[i].allow_plural && |
- location_names[i].length + 1 != length)) { |
- continue; |
- } |
- |
- if (LowerCaseEqualsASCIIWithPlural(word.begin, word.end, |
- location_names[i].string, |
- location_names[i].allow_plural)) { |
- return true; |
- } |
- } |
- |
- return false; |
-} |
- |
-} // namespace internal |
- |
-} // namespace address_parser |
- |
-} // namespace content |