Change default $wgShellLocale to C.UTF-8, and use it to set LC_ALL

It's less likely to cause surprises than language-specific defaults.

Bug: T107128
Change-Id: Ife7673255798f3a3d72028a26607c56b9b7fb224

diff --git a/RELEASE-NOTES-1.30 b/RELEASE-NOTES-1.30
index 8b6a932..cdf8ba4 100644 (file)
--- a/RELEASE-NOTES-1.30
+++ b/RELEASE-NOTES-1.30
@@ -6,7 +6,12 @@ MediaWiki 1.30 is an alpha-quality branch and is not recommended for use in
 production.
 
 === Configuration changes in 1.30 ===
-* …
+* The C.UTF-8 locale should be used for $wgShellLocale, if available, to avoid
+  unexpected behavior when things use local-sensitive string comparisons. For
+  example, Scribunto considers "bar" < "Foo" in most locales since it ignores
+  case.
+* $wgShellLocale now affects LC_ALL rather than only LC_CTYPE. See
+  documentation of $wgShellLocale for details.
 
 === New features in 1.30 ===
 * …

diff --git a/includes/DefaultSettings.php b/includes/DefaultSettings.php
index ac2261c..7c18fcc 100644 (file)
--- a/includes/DefaultSettings.php
+++ b/includes/DefaultSettings.php
@@ -8162,11 +8162,41 @@ $wgShellCgroup = false;
 $wgPhpCli = '/usr/bin/php';
 
 /**
- * Locale for LC_CTYPE, to work around https://bugs.php.net/bug.php?id=45132
- * For Unix-like operating systems, set this to to a locale that has a UTF-8
- * character set. Only the character set is relevant.
- */
-$wgShellLocale = 'en_US.utf8';
+ * Locale for LC_ALL, to provide a known environment for locale-sensitive operations
+ *
+ * For Unix-like operating systems, this should be set to C.UTF-8 or an
+ * equivalent to provide the most consistent behavior for locale-sensitive
+ * C library operations across different-language wikis. If that locale is not
+ * available, use another locale that has a UTF-8 character set.
+ *
+ * This setting mainly affects the behavior of C library functions, including:
+ *  - String collation (order when sorting using locale-sensitive comparison)
+ *    - For example, whether "Å" and "A" are considered to be the same letter or
+ *      different letters and if different whether it comes after "A" or after
+ *      "Z", and whether sorting is case sensitive.
+ *  - String character set (how characters beyond basic ASCII are represented)
+ *    - We need this to be a UTF-8 character set to work around
+ *      https://bugs.php.net/bug.php?id=45132
+ *  - Language used for low-level error messages.
+ *  - Formatting of date/time and numeric values (e.g. '.' versus ',' as the
+ *    decimal separator)
+ *
+ * MediaWiki provides its own methods and classes to perform many
+ * locale-sensitive operations, which are designed to be able to vary locale
+ * based on wiki language or user preference:
+ *  - MediaWiki's Collation class should generally be used instead of the C
+ *    library collation functions when locale-sensitive sorting is needed.
+ *  - MediaWiki's Message class should be used for localization of messages
+ *    displayed to the user.
+ *  - MediaWiki's Language class should be used for formatting numeric and
+ *    date/time values.
+ *
+ * @note If multiple wikis are being served from the same process (e.g. the
+ *  same fastCGI or Apache server), this setting must be the same on all those
+ *  wikis.
+ * @see wfInitShellLocale()
+ */
+$wgShellLocale = 'C.UTF-8';
 
 /** @} */ # End shell }
 

diff --git a/includes/GlobalFunctions.php b/includes/GlobalFunctions.php
index 4325328..c7f132a 100644 (file)
--- a/includes/GlobalFunctions.php
+++ b/includes/GlobalFunctions.php
@@ -2535,8 +2535,15 @@ function wfShellExecWithStderr( $cmd, &$retval = null, $environ = [], $limits =
 }
 
 /**
- * Workaround for https://bugs.php.net/bug.php?id=45132
- * escapeshellarg() destroys non-ASCII characters if LANG is not a UTF-8 locale
+ * Set the locale for locale-sensitive operations
+ *
+ * Sets LC_ALL to a known value to work around issues like the following:
+ * - https://bugs.php.net/bug.php?id=45132 escapeshellarg() destroys non-ASCII
+ *   characters if LANG is not a UTF-8 locale
+ * - T107128 Scribunto string comparison works case insensitive while the
+ *   standard Lua case sensitive
+ *
+ * @see $wgShellLocale
  */
 function wfInitShellLocale() {
        static $done = false;
@@ -2545,8 +2552,8 @@ function wfInitShellLocale() {
        }
        $done = true;
        global $wgShellLocale;
-       putenv( "LC_CTYPE=$wgShellLocale" );
-       setlocale( LC_CTYPE, $wgShellLocale );
+       putenv( "LC_ALL=$wgShellLocale" );
+       setlocale( LC_ALL, $wgShellLocale );
 }
 
 /**

diff --git a/includes/installer/Installer.php b/includes/installer/Installer.php
index 12e8dd1..7028224 100644 (file)
--- a/includes/installer/Installer.php
+++ b/includes/installer/Installer.php
@@ -1016,7 +1016,7 @@ abstract class Installer {
                }
 
                # Try the most common ones.
-               $commonLocales = [ 'en_US.UTF-8', 'en_US.utf8', 'de_DE.UTF-8', 'de_DE.utf8' ];
+               $commonLocales = [ 'C.UTF-8', 'en_US.UTF-8', 'en_US.utf8', 'de_DE.UTF-8', 'de_DE.utf8' ];
                foreach ( $commonLocales as $commonLocale ) {
                        if ( isset( $candidatesByLocale[$commonLocale] ) ) {
                                $this->setVar( 'wgShellLocale', $commonLocale );

diff --git a/includes/installer/LocalSettingsGenerator.php b/includes/installer/LocalSettingsGenerator.php
index 697188e..7df1009 100644 (file)
--- a/includes/installer/LocalSettingsGenerator.php
+++ b/includes/installer/LocalSettingsGenerator.php
@@ -241,7 +241,7 @@ class LocalSettingsGenerator {
                }
 
                if ( !$this->values['wgShellLocale'] ) {
-                       $this->values['wgShellLocale'] = 'en_US.UTF-8';
+                       $this->values['wgShellLocale'] = 'C.UTF-8';
                        $locale = '#';
                } else {
                        $locale = '';