util::htmlEscape: Optimize a bit

The cause of slowness of the first implementation is the memory
allocation overhead and appending character by character.  In this
implementation, the output buffer is reserved the same size of input.
This is reasonable because most likely no replacement happens in
practice. And the unmodified region is copied using iterator range to
speed up a bit.

src/util.cc

@@ -1542,32 +1542,31 @@ getNumericNameInfo(const struct sockaddr* sockaddr, socklen_t len)
 
 std::string htmlEscape(const std::string& src)
 {
-  std::string rv(src);
-  std::string::size_type pos = 0;
-  while ((pos = rv.find_first_of("<>&\"'", pos)) != std::string::npos) {
-    auto ch = rv[pos];
-    if (ch == '<') {
-      rv.replace(pos, 1, "&lt;");
-      pos += 4;
-    }
-    else if (ch == '>') {
-      rv.replace(pos, 1, "&gt;");
-      pos += 4;
-    }
-    else if (ch == '&') {
-      rv.replace(pos, 1, "&amp;");
-      pos += 5;
-    }
-    else if (ch == '"') {
-      rv.replace(pos, 1, "&quot;");
-      pos += 6;
-    }
-    else { // '\''
-      rv.replace(pos, 1, "&#39;");
-      pos += 5;
+  std::string dest;
+  dest.reserve(src.size());
+  auto j = std::begin(src);
+  for(auto i = std::begin(src); i != std::end(src); ++i) {
+    char ch = *i;
+    const char *repl;
+    if(ch == '<') {
+      repl = "&lt;";
+    } else if(ch == '>') {
+      repl = "&gt;";
+    } else if(ch == '&') {
+      repl = "&amp;";
+    } else if(ch == '\'') {
+      repl = "&#39;";
+    } else if(ch == '"') {
+      repl = "&quot;";
+    } else {
+      continue;
     }
+    dest.append(j, i);
+    j = i + 1;
+    dest += repl;
   }
-  return rv;
+  dest.append(j, std::end(src));
+  return dest;
 }
 
 std::pair<size_t, std::string>