From 638801374b45cf1185e299b5206f7001a22d475d Mon Sep 17 00:00:00 2001
From: =?utf8?q?Bartosz=20Dziewo=C5=84ski?= <matma.rex@gmail.com>
Date: Sun, 20 Jul 2014 20:34:34 +0200
Subject: [PATCH] Collation: Workaround for incorrect collation of Estonian
MIME-Version: 1.0
Content-Type: text/plain; charset=utf8
Content-Transfer-Encoding: 8bit

'W' and 'V' should not be considered the same letter for the purposes
of collation in modern Estonian. We work around this by replacing 'W'
and 'w' with 'á´¡' U+1D21 'LATIN LETTER SMALL CAPITAL W' for sortkey
generation, which is collated like 'W' and is not tailored to have the
same primary weight as 'V' in Estonian.

Bug: 54168
Change-Id: I3e8031b9d1dc18fdb7595fb9fd23e5f930106fa4
---
 includes/AutoLoader.php |  1 +
 includes/Collation.php  | 43 ++++++++++++++++++++++++++++++++++++++++-
 2 files changed, 43 insertions(+), 1 deletion(-)

diff --git a/includes/AutoLoader.php b/includes/AutoLoader.php
index 127f2cd4d3..9b1dfc67ef 100644
--- a/includes/AutoLoader.php
+++ b/includes/AutoLoader.php
@@ -45,6 +45,7 @@ $wgAutoloadLocalClasses = array(
 	'ChannelFeed' => 'includes/Feed.php',
 	'Collation' => 'includes/Collation.php',
 	'CollationCkb' => 'includes/Collation.php',
+	'CollationEt' => 'includes/Collation.php',
 	'ConcatenatedGzipHistoryBlob' => 'includes/HistoryBlob.php',
 	'Cookie' => 'includes/Cookie.php',
 	'CookieJar' => 'includes/Cookie.php',
diff --git a/includes/Collation.php b/includes/Collation.php
index 71adb096cb..bcb0a5682a 100644
--- a/includes/Collation.php
+++ b/includes/Collation.php
@@ -49,6 +49,8 @@ abstract class Collation {
 				return new IcuCollation( 'root' );
 			case 'xx-uca-ckb':
 				return new CollationCkb;
+			case 'xx-uca-et':
+				return new CollationEt;
 			default:
 				$match = array();
 				if ( preg_match( '/^uca-([a-z@=-]+)$/', $collationName, $match ) ) {
@@ -253,7 +255,7 @@ class IcuCollation extends Collation {
 		'el' => array(),
 		'eo' => array( "Ĉ", "Ĝ", "Ĥ", "Ĵ", "Ŝ", "Ŭ" ),
 		'es' => array( "Ñ" ),
-		'et' => array( "Š", "Ž", "Õ", "Ä", "Ö", "Ü" ),
+		'et' => array( "Š", "Ž", "Õ", "Ä", "Ö", "Ü", "W" ), // added W for CollationEt (xx-uca-et)
 		'eu' => array( "Ñ" ),
 		'fo' => array( "Á", "Ð", "Í", "Ó", "Ú", "Ý", "Æ", "Ø", "Å" ),
 		'fur' => array( "À", "Á", "Â", "È", "Ì", "Ò", "Ù" ),
@@ -597,3 +599,42 @@ class CollationCkb extends IcuCollation {
 		$this->digitTransformLanguage = Language::factory( 'ckb' );
 	}
 }
+
+/**
+ * Workaround for incorrect collation of Estonian language ('et') in ICU (bug 54168).
+ *
+ * 'W' and 'V' should not be considered the same letter for the purposes of collation in modern
+ * Estonian. We work around this by replacing 'W' and 'w' with 'á´¡' U+1D21 'LATIN LETTER SMALL
+ * CAPITAL W' for sortkey generation, which is collated like 'W' and is not tailored to have the
+ * same primary weight as 'V' in Estonian.
+ */
+class CollationEt extends IcuCollation {
+	function __construct() {
+		parent::__construct( 'et' );
+	}
+
+	private static function mangle( $string ) {
+		return str_replace(
+			array( 'w', 'W' ),
+			'á´¡', // U+1D21 'LATIN LETTER SMALL CAPITAL W'
+			$string
+		);
+	}
+
+	private static function unmangle( $string ) {
+		// Casing data is lost…
+		return str_replace(
+			'á´¡', // U+1D21 'LATIN LETTER SMALL CAPITAL W'
+			'W',
+			$string
+		);
+	}
+
+	function getSortKey( $string ) {
+		return parent::getSortKey( self::mangle( $string ) );
+	}
+
+	function getFirstLetter( $string ) {
+		return self::unmangle( parent::getFirstLetter( self::mangle( $string ) ) );
+	}
+}
-- 
2.20.1