perf(html5): optimize gumbo code to add a libxml2 property

Introduce code that is an optimized version of libxml2's xmlNewNsProp
to avoid traversing the properties linked list to append an attribute
every time we add one.

ext/nokogiri/extconf.rb

@@ -1133,6 +1133,7 @@ def compile
 have_func("xmlCtxtSetOptions") # introduced in libxml2 2.13.0
 have_func("xmlCtxtGetOptions") # introduced in libxml2 2.14.0
 have_func("xmlSwitchEncodingName") # introduced in libxml2 2.13.0
+have_func("xmlAddIDSafe") # introduced in libxml2 2.13.0
 have_func("rb_category_warning") # introduced in Ruby 3.0 but had trouble resolving this symbol in truffleruby
 
 other_library_versions_string = OTHER_LIBRARY_VERSIONS.map { |k, v| [k, v].join(":") }.join(",")

ext/nokogiri/gumbo.c

@@ -113,6 +113,56 @@ set_line(xmlNodePtr node, size_t line)
   }
 }
 
+// This function is essentially xmlNewNsProp, but we skip the full list traversal to append by
+// providing the last property in the linked list as a parameter.
+static xmlAttrPtr
+append_property(xmlNodePtr node, xmlNsPtr ns, const xmlChar *name, const xmlChar *value, xmlAttrPtr last_prop)
+{
+  xmlAttrPtr cur = (xmlAttrPtr) xmlMalloc(sizeof(xmlAttr));
+  memset(cur, 0, sizeof(xmlAttr));
+  cur->type = XML_ATTRIBUTE_NODE;
+  cur->parent = node;
+  xmlDocPtr doc = node->doc;
+  cur->doc = doc;
+  cur->ns = ns;
+
+  if ((doc != NULL) && (doc->dict != NULL)) {
+    cur->name = (xmlChar *) xmlDictLookup(doc->dict, name, -1);
+  } else {
+    cur->name = xmlStrdup(name);
+  }
+
+  if (value != NULL) {
+    cur->children = xmlNewDocText(doc, value);
+    cur->last = NULL;
+    xmlNodePtr tmp = cur->children;
+    while (tmp != NULL) {
+      tmp->parent = (xmlNodePtr) cur;
+      if (tmp->next == NULL) {
+        cur->last = tmp;
+      }
+      tmp = tmp->next;
+    }
+
+    if (doc != NULL) {
+      int res = xmlIsID(doc, node, cur);
+
+      if (res == 1) {
+        xmlAddIDSafe(cur, value);
+      }
+    }
+  }
+
+  if (node->properties == NULL) {
+    node->properties = cur;
+  } else {
+    last_prop->next = cur;
+    cur->prev = last_prop;
+  }
+
+  return cur;
+}
+
 // Construct an XML tree rooted at xml_output_node from the Gumbo tree rooted
 // at gumbo_node.
 static void
@@ -200,6 +250,7 @@ build_tree(
         xmlAddChild(xml_node, xml_child);
 
         // Add the attributes.
+        xmlAttrPtr last_prop = NULL;
         const GumboVector *attrs = &gumbo_child->v.element.attributes;
         for (size_t i = 0; i < attrs->length; i++) {
           const GumboAttribute *attr = attrs->data[i];
@@ -220,7 +271,9 @@ build_tree(
             default:
               ns = NULL;
           }
-          xmlNewNsProp(xml_child, ns, (const xmlChar *)attr->name, (const xmlChar *)attr->value);
+
+          // We micromanage the attribute list for performance reasons.
+          last_prop = append_property(xml_child, ns, (const xmlChar *)attr->name, (const xmlChar *)attr->value, last_prop);
         }
 
         // Add children for this element.

ext/nokogiri/libxml2_polyfill.c

@@ -112,3 +112,14 @@ xmlSwitchEncodingName(xmlParserCtxtPtr ctxt, const char *encoding)
   return (xmlSwitchToEncoding(ctxt, handler));
 }
 #endif
+
+#ifndef HAVE_XMLADDIDSAFE
+int
+xmlAddIDSafe(xmlAttrPtr attr, const xmlChar *value) {
+  xmlIDPtr id = xmlAddID(NULL, attr->doc, value, attr);
+  if (id) {
+    return 1;
+  }
+  return 0;
+}
+#endif

ext/nokogiri/nokogiri.h

@@ -66,6 +66,9 @@ int xmlCtxtGetOptions(xmlParserCtxtPtr ctxt);
 #ifndef HAVE_XMLSWITCHENCODINGNAME
 int xmlSwitchEncodingName(xmlParserCtxtPtr ctxt, const char *encoding);
 #endif
+#ifndef HAVE_XMLADDIDSAFE
+int xmlAddIDSafe(xmlAttrPtr attr, const xmlChar *value);
+#endif
 
 #define XMLNS_PREFIX "xmlns"
 #define XMLNS_PREFIX_LEN 6 /* including either colon or \0 */