From: Bartosz Dziewoński <matma.rex@gmail.com>
Date: Wed, 28 Dec 2016 15:10:24 +0000 (+0100)
Subject: CollationFa: Third time's the charm
X-Git-Tag: 1.31.0-rc.0~4377^2
X-Git-Url: http://git.cyclocoop.org/%24self?a=commitdiff_plain;h=afc6e7cd157f33ec592ae7292d20569707af4397;p=lhc%2Fweb%2Fwiklou.git

CollationFa: Third time's the charm

We have to use a tertiary sortkey for everything with the primary
sortkey of 2627. Otherwise, the "Remove duplicate prefixes" logic
in IcuCollation would remove them.

The following characters will now be considered separate letters in
the 'xx-uca-fa' collation for the purpose of displaying the headings
on category pages: ء ئ ا و ٲ ٳ

Bug: T139110
Change-Id: Ibbea5d76348e4cdc38b74cba44286910b2ed592f
---

diff --git a/includes/collation/CollationFa.php b/includes/collation/CollationFa.php
index b7e45cc1d6..9cce087d3a 100644
--- a/includes/collation/CollationFa.php
+++ b/includes/collation/CollationFa.php
@@ -19,9 +19,12 @@
  */
 
 /**
- * Temporary workaround for incorrect collation of Persian language ('fa') in ICU (bug T139110).
+ * Temporary workaround for incorrect collation of Persian language ('fa') in ICU 52 (bug T139110).
  *
- * 'ا' and 'و' should not be considered the same letter for the purposes of collation in Persian.
+ * All of the following will be considered separate letters for category headings in Persian:
+ *  - Characters 'و' 'ا' (often appear at the beginning of words)
+ *  - Characters 'Ù²' 'Ù³' (may appear at the beginning of words in loanwords)
+ *  - Characters 'ء' 'ئ' (don't appear at the beginning of words, but it's easier to implement)
  *
  * @since 1.29
  */
@@ -34,11 +37,14 @@ class CollationFa extends IcuCollation {
 	}
 
 	public function getPrimarySortKey( $string ) {
-		$firstLetter = mb_substr( $string, 0, 1 );
-		if ( $firstLetter === 'و' || $firstLetter === 'ا' ) {
+		$primary = parent::getPrimarySortKey( $string );
+		// We have to use a tertiary sortkey for everything with the primary sortkey of 2627.
+		// Otherwise, the "Remove duplicate prefixes" logic in IcuCollation would remove them.
+		// This matches sortkeys for the following characters: ء ئ ا و ٲ ٳ
+		if ( substr( $primary, 0, 2 ) === "\x26\x27" ) {
+			wfDebug( "Using tertiary sortkey for '$string'\n" );
 			return $this->tertiaryCollator->getSortKey( $string );
 		}
-
-		return parent::getPrimarySortKey( $string );
+		return $primary;
 	}
 }