diff options
author | Phil Sturgeon <email@philsturgeon.co.uk> | 2011-08-13 18:26:04 +0200 |
---|---|---|
committer | Phil Sturgeon <email@philsturgeon.co.uk> | 2011-08-13 18:26:04 +0200 |
commit | 50e57bcfe659f9992760a1972c6c20c1298537f5 (patch) | |
tree | 01388a821afc765be69e603abb5951970b5c49b5 /system | |
parent | b0eae5f81a4cb92911bb215ad814ae5caef4f61d (diff) |
Improved logic for plural() and singular() as words like "tax" would become "taxs" and it was tough to code in new rules.
Diffstat (limited to 'system')
-rw-r--r-- | system/helpers/inflector_helper.php | 115 |
1 files changed, 67 insertions, 48 deletions
diff --git a/system/helpers/inflector_helper.php b/system/helpers/inflector_helper.php index c7c113b8a..7b99bc5b4 100644 --- a/system/helpers/inflector_helper.php +++ b/system/helpers/inflector_helper.php @@ -41,30 +41,48 @@ if ( ! function_exists('singular')) { function singular($str) { - $str = trim($str); - $end = substr($str, -3); - - $str = preg_replace('/(.*)?([s|c]h)es/i','$1$2',$str); - - if (strtolower($end) == 'ies') + $result = strval($str); + + $singular_rules = array( + '/(matr)ices$/' => '\1ix', + '/(vert|ind)ices$/' => '\1ex', + '/^(ox)en/' => '\1', + '/(alias)es$/' => '\1', + '/([octop|vir])i$/' => '\1us', + '/(cris|ax|test)es$/' => '\1is', + '/(shoe)s$/' => '\1', + '/(o)es$/' => '\1', + '/(bus|campus)es$/' => '\1', + '/([m|l])ice$/' => '\1ouse', + '/(x|ch|ss|sh)es$/' => '\1', + '/(m)ovies$/' => '\1\2ovie', + '/(s)eries$/' => '\1\2eries', + '/([^aeiouy]|qu)ies$/' => '\1y', + '/([lr])ves$/' => '\1f', + '/(tive)s$/' => '\1', + '/(hive)s$/' => '\1', + '/([^f])ves$/' => '\1fe', + '/(^analy)ses$/' => '\1sis', + '/((a)naly|(b)a|(d)iagno|(p)arenthe|(p)rogno|(s)ynop|(t)he)ses$/' => '\1\2sis', + '/([ti])a$/' => '\1um', + '/(p)eople$/' => '\1\2erson', + '/(m)en$/' => '\1an', + '/(s)tatuses$/' => '\1\2tatus', + '/(c)hildren$/' => '\1\2hild', + '/(n)ews$/' => '\1\2ews', + '/([^u])s$/' => '\1', + ); + + foreach ($singular_rules as $rule => $replacement) { - $str = substr($str, 0, strlen($str)-3).(preg_match('/[a-z]/',$end) ? 'y' : 'Y'); - } - elseif (strtolower($end) == 'ses') - { - $str = substr($str, 0, strlen($str)-2); - } - else - { - $end = strtolower(substr($str, -1)); - - if ($end == 's') + if (preg_match($rule, $result)) { - $str = substr($str, 0, strlen($str)-1); + $result = preg_replace($rule, $replacement, $result); + break; } } - return $str; + return $result; } } @@ -83,40 +101,41 @@ if ( ! function_exists('singular')) if ( ! function_exists('plural')) { function plural($str, $force = FALSE) - { - $str = trim($str); - $end = substr($str, -1); - - if (preg_match('/y/i',$end)) - { - // Y preceded by vowel => regular plural - $vowels = array('a', 'e', 'i', 'o', 'u', 'A', 'E', 'I', 'O', 'U'); - $str = in_array(substr($str, -2, 1), $vowels) ? $str.'s' : substr($str, 0, -1).'ies'; - } - elseif (preg_match('/h/i',$end)) - { - if(preg_match('/^[c|s]h$/i',substr($str, -2))) - { - $str .= 'es'; - } - else - { - $str .= 's'; - } - } - elseif (preg_match('/s/i',$end)) + { + $result = strval($str); + + $plural_rules = array( + '/^(ox)$/' => '\1\2en', // ox + '/([m|l])ouse$/' => '\1ice', // mouse, louse + '/(matr|vert|ind)ix|ex$/' => '\1ices', // matrix, vertex, index + '/(x|ch|ss|sh)$/' => '\1es', // search, switch, fix, box, process, address + '/([^aeiouy]|qu)y$/' => '\1ies', // query, ability, agency + '/(hive)$/' => '\1s', // archive, hive + '/(?:([^f])fe|([lr])f)$/' => '\1\2ves', // half, safe, wife + '/sis$/' => 'ses', // basis, diagnosis + '/([ti])um$/' => '\1a', // datum, medium + '/(p)erson$/' => '\1eople', // person, salesperson + '/(m)an$/' => '\1en', // man, woman, spokesman + '/(c)hild$/' => '\1hildren', // child + '/(buffal|tomat)o$/' => '\1\2oes', // buffalo, tomato + '/(bu|campu)s$/' => '\1\2ses', // bus, campus + '/(alias|status|virus)/' => '\1es', // alias + '/(octop)us$/' => '\1i', // octopus + '/(ax|cris|test)is$/' => '\1es', // axis, crisis + '/s$/' => 's', // no change (compatibility) + '/$/' => 's', + ); + + foreach ($plural_rules as $rule => $replacement) { - if ($force == TRUE) + if (preg_match($rule, $result)) { - $str .= 'es'; + $result = preg_replace($rule, $replacement, $result); + break; } } - else - { - $str .= 's'; - } - return $str; + return $result; } } |