From: C. Scott Ananian <cscott@cscott.net>
Date: Tue, 2 Jul 2019 18:04:39 +0000 (-0400)
Subject: LanguageConverter performance: Reuse the same string object for regexp
X-Git-Tag: 1.34.0-rc.0~1201^2
X-Git-Url: https://git.heureux-cyclage.org/?p=lhc%2Fweb%2Fwiklou.git;a=commitdiff_plain;h=25512652d289b02fa90268f9ff06ba094af6781d

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 7fd3631e75..9fc7d73f0e 100644
--- 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 = '';