#include #include /* * call-seq: * length * * Get the length of the node set */ static VALUE length(VALUE self) { xmlNodeSetPtr node_set; Data_Get_Struct(self, xmlNodeSet, node_set); if(node_set) return INT2NUM(node_set->nodeNr); return INT2NUM(0); } /* * call-seq: * push(node) * * Append +node+ to the NodeSet. */ static VALUE push(VALUE self, VALUE rb_node) { xmlNodeSetPtr node_set; xmlNodePtr node; Data_Get_Struct(self, xmlNodeSet, node_set); Data_Get_Struct(rb_node, xmlNode, node); xmlXPathNodeSetAdd(node_set, node); return self; } /* * call-seq: * [](i) * * Get the node at index +i+ */ static VALUE index_at(VALUE self, VALUE number) { int i = NUM2INT(number); xmlNodeSetPtr node_set; Data_Get_Struct(self, xmlNodeSet, node_set); if(i >= node_set->nodeNr || abs(i) > node_set->nodeNr) return Qnil; if(i < 0) i = i + node_set->nodeNr; return Nokogiri_wrap_xml_node(node_set->nodeTab[i]); } static void deallocate(xmlNodeSetPtr node_set) { /* * xmlXPathFreeNodeSet() contains an implicit assumption that it is being * called before any of its pointed-to nodes have been free()d. this * assumption lies in the operation where it dereferences nodeTab pointers * while searching for namespace nodes to free. * * however, since Ruby's GC mechanism cannot guarantee the strict order in * which ruby objects will be GC'd, nodes may be garbage collected before a * nodeset containing pointers to those nodes. (this is true regardless of * how we declare dependencies between objects with rb_gc_mark().) * * as a result, xmlXPathFreeNodeSet() will perform unsafe memory operations, * and calling it would be evil. * * on the bright side, though, Nokogiri's API currently does not cause * namespace nodes to be included in node sets, ever. * * armed with that fact, we examined xmlXPathFreeNodeSet() and related libxml * code and determined that, within the Nokogiri abstraction, we will not * leak memory if we simply free the node set's memory directly. that's only * quasi-evil! * * there's probably a lesson in here somewhere about intermingling, within a * single array, structs with different memory-ownership semantics. or more * generally, a lesson about building an API in C/C++ that does not contain * assumptions about the strict order in which memory will be released. hey, * that sounds like a great idea for a blog post! get to it! * * "In Valgrind We Trust." seriously. */ NOKOGIRI_DEBUG_START(node_set) ; if (node_set->nodeTab != NULL) xmlFree(node_set->nodeTab); xmlFree(node_set); NOKOGIRI_DEBUG_END(node_set) ; } static VALUE allocate(VALUE klass) { return Nokogiri_wrap_xml_node_set(xmlXPathNodeSetCreate(NULL)); } VALUE Nokogiri_wrap_xml_node_set(xmlNodeSetPtr node_set) { return Data_Wrap_Struct(cNokogiriXmlNodeSet, 0, deallocate, node_set); } VALUE cNokogiriXmlNodeSet ; void init_xml_node_set(void) { VALUE klass = cNokogiriXmlNodeSet = rb_eval_string("Nokogiri::XML::NodeSet"); rb_define_alloc_func(klass, allocate); rb_define_method(klass, "length", length, 0); rb_define_method(klass, "[]", index_at, 1); rb_define_method(klass, "push", push, 1); }