LanguageConverter performance: Reuse the same string object for regexp

The regular expression used by LanguageConverter::autoConvert() is a
constant, but it is being created on-the-fly by every invocation.
This causes an expensive full-string comparison when the compiled
regular expression is fetched from the cache -- since the regex is 332
bytes long, the time taken for this comparison can add up quickly: on
page with a lot of tags, the regexp cache may spend more time looking
up the regexp than it takes to execute it.

Bug: T223969
Change-Id: I53c3e631e47a791cf3f0844dd79d4357605c59e3

diff --git a/languages/LanguageConverter.php b/languages/LanguageConverter.php
index 7fd3631..9fc7d73 100644 (file)
--- a/languages/LanguageConverter.php
+++ b/languages/LanguageConverter.php
@@ -391,27 +391,30 @@ class LanguageConverter {
                   IMPORTANT: Beware of failure from pcre.backtrack_limit (T124404).
                   Minimize use of backtracking where possible.
                */
-               $marker = '|' . Parser::MARKER_PREFIX . '[^\x7f]++\x7f';
-
-               // this one is needed when the text is inside an HTML markup
-               $htmlfix = '|<[^>\004]++(?=\004$)|^[^<>]*+>';
-
-               // Optimize for the common case where these tags have
-               // few or no children. Thus try and possesively get as much as
-               // possible, and only engage in backtracking when we hit a '<'.
-
-               // disable convert to variants between <code> tags
-               $codefix = '<code>[^<]*+(?:(?:(?!<\/code>).)[^<]*+)*+<\/code>|';
-               // disable conversion of <script> tags
-               $scriptfix = '<script[^>]*+>[^<]*+(?:(?:(?!<\/script>).)[^<]*+)*+<\/script>|';
-               // disable conversion of <pre> tags
-               $prefix = '<pre[^>]*+>[^<]*+(?:(?:(?!<\/pre>).)[^<]*+)*+<\/pre>|';
-               // The "|.*+)" at the end, is in case we missed some part of html syntax,
-               // we will fail securely (hopefully) by matching the rest of the string.
-               $htmlFullTag = '<(?:[^>=]*+(?>[^>=]*+=\s*+(?:"[^"]*"|\'[^\']*\'|[^\'">\s]*+))*+[^>=]*+>|.*+)|';
-
-               $reg = '/' . $codefix . $scriptfix . $prefix . $htmlFullTag .
-                       '&[a-zA-Z#][a-z0-9]++;' . $marker . $htmlfix . '|\004$/s';
+               static $reg;
+               if ( $reg === null ) {
+                       $marker = '|' . Parser::MARKER_PREFIX . '[^\x7f]++\x7f';
+
+                       // this one is needed when the text is inside an HTML markup
+                       $htmlfix = '|<[^>\004]++(?=\004$)|^[^<>]*+>';
+
+                       // Optimize for the common case where these tags have
+                       // few or no children. Thus try and possesively get as much as
+                       // possible, and only engage in backtracking when we hit a '<'.
+
+                       // disable convert to variants between <code> tags
+                       $codefix = '<code>[^<]*+(?:(?:(?!<\/code>).)[^<]*+)*+<\/code>|';
+                       // disable conversion of <script> tags
+                       $scriptfix = '<script[^>]*+>[^<]*+(?:(?:(?!<\/script>).)[^<]*+)*+<\/script>|';
+                       // disable conversion of <pre> tags
+                       $prefix = '<pre[^>]*+>[^<]*+(?:(?:(?!<\/pre>).)[^<]*+)*+<\/pre>|';
+                       // The "|.*+)" at the end, is in case we missed some part of html syntax,
+                       // we will fail securely (hopefully) by matching the rest of the string.
+                       $htmlFullTag = '<(?:[^>=]*+(?>[^>=]*+=\s*+(?:"[^"]*"|\'[^\']*\'|[^\'">\s]*+))*+[^>=]*+>|.*+)|';
+
+                       $reg = '/' . $codefix . $scriptfix . $prefix . $htmlFullTag .
+                                '&[a-zA-Z#][a-z0-9]++;' . $marker . $htmlfix . '|\004$/s';
+               }
                $startPos = 0;
                $sourceBlob = '';
                $literalBlob = '';