X-Git-Url: https://git.openstreetmap.org./nominatim.git/blobdiff_plain/2613ebfa0143cdd1e3220577d48d9aa8341dbf38..d68996127d70855f29035490e7e4eeb75f8dae8c:/lib/Geocode.php diff --git a/lib/Geocode.php b/lib/Geocode.php index 612c1a0f..b9e26f03 100644 --- a/lib/Geocode.php +++ b/lib/Geocode.php @@ -7,6 +7,7 @@ require_once(CONST_BasePath.'/lib/Phrase.php'); require_once(CONST_BasePath.'/lib/ReverseGeocode.php'); require_once(CONST_BasePath.'/lib/SearchDescription.php'); require_once(CONST_BasePath.'/lib/SearchContext.php'); +require_once(CONST_BasePath.'/lib/TokenList.php'); class Geocode { @@ -16,8 +17,6 @@ class Geocode protected $aLangPrefOrder = array(); - protected $bIncludeAddressDetails = false; - protected $aExcludePlaceIDs = array(); protected $bReverseInPlan = false; @@ -86,7 +85,6 @@ class Geocode $aParams['exclude_place_ids'] = implode(',', $this->aExcludePlaceIDs); } - if ($this->bIncludeAddressDetails) $aParams['addressdetails'] = '1'; if ($this->bBoundedSearch) $aParams['bounded'] = '1'; if ($this->aCountryCodes) { @@ -182,9 +180,6 @@ class Geocode public function loadParamArray($oParams, $sForceGeometryType = null) { - $this->bIncludeAddressDetails - = $oParams->getBool('addressdetails', $this->bIncludeAddressDetails); - $this->bBoundedSearch = $oParams->getBool('bounded', $this->bBoundedSearch); $this->setLimit($oParams->getInt('limit', $this->iFinalLimit)); @@ -246,8 +241,8 @@ class Geocode } $this->oPlaceLookup->loadParamArray($oParams, $sForceGeometryType); - $this->oPlaceLookup->setIncludeAddressDetails(false); $this->oPlaceLookup->setIncludePolygonAsPoints($oParams->getBool('polygon')); + $this->oPlaceLookup->setIncludeAddressDetails($oParams->getBool('addressdetails', false)); } public function setQueryFromParams($oParams) @@ -332,10 +327,10 @@ class Geocode return false; } - public function getGroupedSearches($aSearches, $aPhrases, $aValidTokens, $bIsStructured) + public function getGroupedSearches($aSearches, $aPhrases, $oValidTokens, $bIsStructured) { /* - Calculate all searches using aValidTokens i.e. + Calculate all searches using oValidTokens i.e. 'Wodsworth Road, Sheffield' => Phrase Wordset @@ -365,38 +360,37 @@ class Geocode //var_dump($oCurrentSearch); //echo ""; - // If the token is valid - if (isset($aValidTokens[' '.$sToken])) { - foreach ($aValidTokens[' '.$sToken] as $aSearchTerm) { - $aNewSearches = $oCurrentSearch->extendWithFullTerm( - $aSearchTerm, - isset($aValidTokens[$sToken]) - && strpos($sToken, ' ') === false, - $sPhraseType, - $iToken == 0 && $iPhrase == 0, - $iPhrase == 0, - $iToken + 1 == count($aWordset) - && $iPhrase + 1 == count($aPhrases) - ); - - foreach ($aNewSearches as $oSearch) { - if ($oSearch->getRank() < $this->iMaxRank) { - $aNewWordsetSearches[] = $oSearch; - } + // Tokens with full name matches. + foreach ($oValidTokens->get(' '.$sToken) as $oSearchTerm) { + $aNewSearches = $oCurrentSearch->extendWithFullTerm( + $oSearchTerm, + $oValidTokens->contains($sToken) + && strpos($sToken, ' ') === false, + $sPhraseType, + $iToken == 0 && $iPhrase == 0, + $iPhrase == 0, + $iToken + 1 == count($aWordset) + && $iPhrase + 1 == count($aPhrases) + ); + + foreach ($aNewSearches as $oSearch) { + if ($oSearch->getRank() < $this->iMaxRank) { + $aNewWordsetSearches[] = $oSearch; } } } // Look for partial matches. // Note that there is no point in adding country terms here // because country is omitted in the address. - if (isset($aValidTokens[$sToken]) && $sPhraseType != 'country') { + if ($sPhraseType != 'country') { // Allow searching for a word - but at extra cost - foreach ($aValidTokens[$sToken] as $aSearchTerm) { + foreach ($oValidTokens->get($sToken) as $oSearchTerm) { $aNewSearches = $oCurrentSearch->extendWithPartialTerm( - $aSearchTerm, + $sToken, + $oSearchTerm, $bIsStructured, $iPhrase, - isset($aValidTokens[' '.$sToken]) ? $aValidTokens[' '.$sToken] : array() + $oValidTokens->get(' '.$sToken) ); foreach ($aNewSearches as $oSearch) { @@ -645,75 +639,51 @@ class Geocode Debug::printDebugTable('Phrases', $aPhrases); Debug::printVar('Tokens', $aTokens); + $oValidTokens = new TokenList(); + if (!empty($aTokens)) { - // Check which tokens we have, get the ID numbers $sSQL = 'SELECT word_id, word_token, word, class, type, country_code, operator, search_name_count'; $sSQL .= ' FROM word '; $sSQL .= ' WHERE word_token in ('.join(',', array_map('getDBQuoted', $aTokens)).')'; Debug::printSQL($sSQL); - $aValidTokens = array(); - $aDatabaseWords = chksql( - $this->oDB->getAll($sSQL), - 'Could not get word tokens.' + $oValidTokens->addTokensFromDB( + $this->oDB, + $aTokens, + $this->aCountryCodes, + $sNormQuery, + $this->oNormalizer ); - $aWordFrequencyScores = array(); - foreach ($aDatabaseWords as $aToken) { - // Filter country tokens that do not match restricted countries. - if ($this->aCountryCodes - && $aToken['country_code'] - && !in_array($aToken['country_code'], $this->aCountryCodes) - ) { - continue; - } - // Special terms need to appear in their normalized form. - if ($aToken['word'] && $aToken['class']) { - $sNormWord = $this->normTerm($aToken['word']); - if (strpos($sNormQuery, $sNormWord) === false) { - continue; - } - } - - if (isset($aValidTokens[$aToken['word_token']])) { - $aValidTokens[$aToken['word_token']][] = $aToken; - } else { - $aValidTokens[$aToken['word_token']] = array($aToken); - } - $aWordFrequencyScores[$aToken['word_id']] = $aToken['search_name_count'] + 1; - } - - // US ZIP+4 codes - if there is no token, merge in the 5-digit ZIP code + // Try more interpretations for Tokens that could not be matched. foreach ($aTokens as $sToken) { - if (!isset($aValidTokens[$sToken]) && preg_match('/^([0-9]{5}) [0-9]{4}$/', $sToken, $aData)) { - if (isset($aValidTokens[$aData[1]])) { - foreach ($aValidTokens[$aData[1]] as $aToken) { - if (!$aToken['class']) { - if (isset($aValidTokens[$sToken])) { - $aValidTokens[$sToken][] = $aToken; - } else { - $aValidTokens[$sToken] = array($aToken); - } - } - } + if ($sToken[0] == ' ' && !$oValidTokens->contains($sToken)) { + if (preg_match('/^ ([0-9]{5}) [0-9]{4}$/', $sToken, $aData)) { + // US ZIP+4 codes - merge in the 5-digit ZIP code + $oValidTokens->addToken( + $sToken, + new Token\Postcode(null, $aData[1], 'us') + ); + } elseif (preg_match('/^ [0-9]+$/', $sToken)) { + // Unknown single word token with a number. + // Assume it is a house number. + $oValidTokens->addToken( + $sToken, + new Token\HouseNumber(null, trim($sToken)) + ); } } } - foreach ($aTokens as $sToken) { - // Unknown single word token with a number - assume it is a house number - if (!isset($aValidTokens[' '.$sToken]) && strpos($sToken, ' ') === false && preg_match('/^[0-9]+$/', $sToken)) { - $aValidTokens[' '.$sToken] = array(array('class' => 'place', 'type' => 'house', 'word_token' => ' '.$sToken)); - } - } - Debug::printGroupTable('Valid Tokens', $aValidTokens); - // Any words that have failed completely? // TODO: suggestions + + Debug::printGroupTable('Valid Tokens', $oValidTokens->debugInfo()); + Debug::newSection('Search candidates'); - $aGroupedSearches = $this->getGroupedSearches($aSearches, $aPhrases, $aValidTokens, $bStructuredPhrases); + $aGroupedSearches = $this->getGroupedSearches($aSearches, $aPhrases, $oValidTokens, $bStructuredPhrases); if ($this->bReverseInPlan) { // Reverse phrase array and also reverse the order of the wordsets in @@ -724,7 +694,7 @@ class Geocode if (count($aPhrases) > 1) { $aPhrases[count($aPhrases)-1]->invertWordSets(); } - $aReverseGroupedSearches = $this->getGroupedSearches($aSearches, $aPhrases, $aValidTokens, false); + $aReverseGroupedSearches = $this->getGroupedSearches($aSearches, $aPhrases, $oValidTokens, false); foreach ($aGroupedSearches as $aSearches) { foreach ($aSearches as $aSearch) { @@ -764,7 +734,10 @@ class Geocode } } - if (CONST_Debug) _debugDumpGroupedSearches($aGroupedSearches, $aValidTokens); + Debug::printGroupedSearch( + $aGroupedSearches, + $oValidTokens->debugTokenByWordIdList() + ); // Start the search process $iGroupLoop = 0; @@ -774,14 +747,14 @@ class Geocode foreach ($aSearches as $oSearch) { $iQueryLoop++; - if (CONST_Debug) { - echo "