Optimize mb_str{,im}width for performance

Rather than doing a linear search of a table of fullwidth codepoint
ranges for every input character,

1) Short-cut the search if the codepoint is below the first such range
2) Otherwise, do a binary (rather than linear) search

Author: alexdowad

ext/mbstring/libmbfl/mbfl/eaw_table.h

@@ -14,6 +14,8 @@
  * which should be displayed as double-width.
  */
 
+#define FIRST_DOUBLEWIDTH_CODEPOINT 0x1100
+
 static const struct {
 	int begin;
 	int end;

ext/mbstring/libmbfl/mbfl/mbfilter.c

@@ -1203,31 +1203,33 @@ mbfl_strcut(
 	return result;
 }
 
-
-/*
- *  strwidth
- */
-static size_t is_fullwidth(int c)
+/* Some East Asian characters, when printed at a terminal (or the like), require double
+ * the usual amount of horizontal space. We call these "fullwidth" characters. */
+static size_t character_width(int c)
 {
-	int i;
-
-	if (c < mbfl_eaw_table[0].begin) {
-		return 0;
-	}
-
-	for (i = 0; i < sizeof(mbfl_eaw_table) / sizeof(mbfl_eaw_table[0]); i++) {
-		if (mbfl_eaw_table[i].begin <= c && c <= mbfl_eaw_table[i].end) {
-			return 1;
+	if (c < FIRST_DOUBLEWIDTH_CODEPOINT) {
+		return 1;
+	}
+
+	/* Do a binary search to see if we fall in any of the fullwidth ranges */
+	int lo = 0, hi = sizeof(mbfl_eaw_table) / sizeof(mbfl_eaw_table[0]);
+	while (lo < hi) {
+		int probe = (lo + hi) / 2;
+		if (c < mbfl_eaw_table[probe].begin) {
+			hi = probe;
+		} else if (c > mbfl_eaw_table[probe].end) {
+			lo = probe + 1;
+		} else {
+			return 2;
 		}
 	}
 
-	return 0;
+	return 1;
 }
 
-static int
-filter_count_width(int c, void* data)
+static int filter_count_width(int c, void* data)
 {
-	(*(size_t *)data) += (is_fullwidth(c) ? 2: 1);
+	(*(size_t *)data) += character_width(c);
 	return 0;
 }
 
@@ -1289,7 +1291,7 @@ collector_strimwidth(int c, void* data)
 		break;
 	default:
 		if (pc->outchar >= pc->from) {
-			pc->outwidth += (is_fullwidth(c) ? 2: 1);
+			pc->outwidth += character_width(c);
 
 			if (pc->outwidth > pc->width) {
 				if (pc->status == 0) {

ext/mbstring/ucgendat/ucgendat.php

@@ -700,7 +700,7 @@ function generateMPH(array $map, bool $fast) {
 }
 
 function generateEastAsianWidthData(array $wideRanges) {
-      $result = <<<'HEADER'
+    $result = <<<'HEADER'
 /* This file was generated by ext/mbstring/ucgendat/ucgendat.php.
  *
  *                     DO NOT EDIT THIS FILE!
@@ -717,12 +717,17 @@ function generateEastAsianWidthData(array $wideRanges) {
  * which should be displayed as double-width.
  */
 
+HEADER;
+
+    $result .= "\n#define FIRST_DOUBLEWIDTH_CODEPOINT 0x" . dechex($wideRanges[0]->start) . "\n\n";
+
+    $result .= <<<'TABLESTART'
 static const struct {
 	int begin;
 	int end;
 } mbfl_eaw_table[] = {
 
-HEADER;
+TABLESTART;
 
     foreach ($wideRanges as $range) {
         $startCode = dechex($range->start);