1 (******************************************************************************)
2 (* SXSI : XPath evaluator *)
3 (* Kim Nguyen (Kim.Nguyen@nicta.com.au) *)
4 (* Copyright NICTA 2008 *)
5 (* Distributed under the terms of the LGPL (see LICENCE) *)
6 (******************************************************************************)
10 external init_lib : unit -> unit = "sxsi_cpp_init"
12 exception CPlusPlusError of string
14 let () = Callback.register_exception "CPlusPlusError" (CPlusPlusError "")
18 type node = [ `Tree ] Node.t
22 external register_tag : tree -> string -> Tag.t = "caml_xml_tree_register_tag"
24 external tag_name : tree -> Tag.t -> string = "caml_xml_tree_get_tag_name"
26 let tag t = (); fun s ->
29 | "<@>" -> Tag.attribute
30 | "" -> Tag.document_node
31 | "<@$>" -> Tag.attribute_data
32 | _ -> register_tag t s
36 if t == Tag.pcdata then "<$>"
37 else if t == Tag.attribute_data then "<@$>"
38 else if t == Tag.attribute then "<@>"
39 else if t == Tag.nullt then "<!NIL!>"
46 Tag.to_string = to_string t;
47 Tag.translate = translate
53 external create : unit -> t = "caml_xml_tree_builder_create"
54 external open_document : t -> bool -> int -> bool -> int -> unit = "caml_xml_tree_builder_open_document"
55 external close_document : t -> tree = "caml_xml_tree_builder_close_document"
56 external open_tag : t -> string -> unit = "caml_xml_tree_builder_open_tag"
57 external close_tag : t -> string -> unit = "caml_xml_tree_builder_close_tag"
58 external text : t -> string -> unit = "caml_xml_tree_builder_text"
64 (c == '\n' || c == '\t' || c == ' ') && loop len (i+1)
68 loop (String.length s) 0
72 let event_counter = ref 0 in
75 if !event_counter land 0xffffff == 0 then
76 Logger.print Format.err_formatter "Current position: %i@\n@?" (Expat.get_current_byte_index parser_))
80 if Buffer.length t > 0 then begin
81 let s = Buffer.contents t in
90 let output_attr b name value =
91 let atname = "<@>" ^ name in
98 let start_element_handler parser_ b t tag attr_list =
99 display_count parser_;
106 List.iter (fun (name, value) -> output_attr b name value) l;
110 let end_element_handler parser_ b t tag =
111 display_count parser_;
115 let character_data_handler parser_ _ t text =
116 display_count parser_;
117 Buffer.add_string t text
119 let create_parser () =
120 let buf = Buffer.create 512 in
121 let build = create () in
122 let parser_ = Expat.parser_create ~encoding:None in
126 Logger.print Format.err_formatter "Finished parsing@\n";
127 Logger.print Format.err_formatter "Starting index construction@\n";
128 let r = close_document build in
129 Logger.print Format.err_formatter "Finished index construction@\n";
132 Expat.set_start_element_handler parser_ (start_element_handler parser_ build buf);
133 Expat.set_end_element_handler parser_ (end_element_handler parser_ build buf);
134 Expat.set_character_data_handler parser_ (character_data_handler parser_ build buf);
135 Logger.print Format.err_formatter "Started parsing@\n";
136 open_document build !Options.index_empty_texts !Options.sample_factor
137 !Options.disable_text_collection !Options.text_index_type;
142 let parser_, finalizer = create_parser () in
143 Expat.parse parser_ s;
146 let parse_file file =
147 let in_chan = open_in file in
148 let buffer = String.create 4096 in
149 let parser_, finalizer = create_parser () in
153 let read = input in_chan buffer 0 4096 in
154 if read == 0 then raise End_of_file else
155 Expat.parse_sub parser_ buffer 0 read;
159 | End_of_file -> close_in in_chan
169 type bit_vector = string
171 external bool_of_int : int -> bool = "%identity"
172 external int_of_bool : bool -> int = "%identity"
174 let bit_vector_unsafe_get v i =
176 (((Char.code (String.unsafe_get v (i lsr 3))) lsr (i land 7)) land 1)
177 let chr (c:int) : char = Obj.magic (c land 0xff)
178 let bit_vector_unsafe_set v i b =
180 let c = Char.code v.[j] in
181 let bit = int_of_bool b in
182 let mask = bit lsl (i land 7) in
183 if b then v.[j] <- chr (c lor mask) else v.[j] <- (chr (c land (lnot mask)))
185 let bit_vector_create n =
186 let len = if n <= 0 then 0 else (n - 1) / 8 + 1 in
187 String.make len '\000'
191 elements: Ptset.Int.t;
192 attributes: Ptset.Int.t;
193 attribute_array : Tag.t array;
194 children : Ptset.Int.t array;
195 siblings : Ptset.Int.t array;
196 descendants: Ptset.Int.t array;
197 followings: Ptset.Int.t array;
201 let tag_operations t = mk_tag_ops t.doc
203 external parse_xml_uri : string -> int -> bool -> bool -> int -> tree = "caml_call_shredder_uri"
204 external parse_xml_string : string -> int -> bool -> bool -> int -> tree = "caml_call_shredder_string"
206 external tree_print_xml_fast3 : tree -> [`Tree ] Node.t -> Unix.file_descr -> unit = "caml_xml_tree_print"
207 let print_xml t n fd =
208 tree_print_xml_fast3 t.doc n fd
211 external tree_save : tree -> Unix.file_descr -> string -> unit = "caml_xml_tree_save"
212 external tree_load : Unix.file_descr -> string -> bool -> int -> tree = "caml_xml_tree_load"
214 external nullt : unit -> 'a Node.t = "caml_xml_tree_nullt"
216 let nil : [`Tree ] Node.t = Node.nil
217 let root : [`Tree ] Node.t = Node.null
221 external unordered_set_alloc : int -> unordered_set = "caml_unordered_set_alloc"
222 external unordered_set_length : unordered_set -> int = "caml_unordered_set_length"
223 external unordered_set_insert : unordered_set -> int -> unit = "caml_unordered_set_set" "noalloc"
225 module HPtset = Hashtbl.Make(Ptset.Int)
227 let vector_htbl = HPtset.create MED_H_SIZE
229 let unordered_set_of_set s =
231 HPtset.find vector_htbl s
234 let v = unordered_set_alloc (Ptset.Int.cardinal s) in
235 let _ = Ptset.Int.iter (fun e -> unordered_set_insert v e) s in
236 HPtset.add vector_htbl s v; v
238 let ptset_to_vector = unordered_set_of_set
240 (** tree interface *)
242 external tree_root : tree -> [`Tree] Node.t = "caml_xml_tree_root" "noalloc"
245 external tree_first_child : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_first_child" "noalloc"
246 let first_child t n = tree_first_child t.doc n
248 external tree_first_element : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_first_element" "noalloc"
249 let first_element t n = tree_first_element t.doc n
251 external tree_tagged_child : tree -> [`Tree] Node.t -> Tag.t -> [`Tree] Node.t = "caml_xml_tree_tagged_child" "noalloc"
252 let tagged_child t n tag = tree_tagged_child t.doc n tag
254 external tree_select_child : tree -> [`Tree ] Node.t -> unordered_set -> [`Tree] Node.t = "caml_xml_tree_select_child" "noalloc"
255 let select_child t n tag_set = tree_select_child t.doc n tag_set
257 external tree_last_child : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_last_child" "noalloc"
258 let last_child t n = tree_last_child t.doc n
261 external tree_next_sibling : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_next_sibling" "noalloc"
262 let next_sibling t n = tree_next_sibling t.doc n
264 external tree_next_element : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_next_element" "noalloc"
265 let next_element t n = tree_next_element t.doc n
267 external tree_tagged_sibling : tree -> [`Tree] Node.t -> Tag.t -> [`Tree] Node.t = "caml_xml_tree_tagged_sibling" "noalloc"
268 let tagged_sibling t n tag = tree_tagged_sibling t.doc n tag
270 external tree_select_sibling : tree -> [`Tree ] Node.t -> unordered_set -> [`Tree] Node.t = "caml_xml_tree_select_sibling" "noalloc"
271 let select_sibling t n tag_set = tree_select_sibling t.doc n tag_set
273 external tree_prev_sibling : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_prev_sibling" "noalloc"
274 let prev_sibling t n = tree_prev_sibling t.doc n
278 external tree_tagged_descendant : tree -> [`Tree ] Node.t -> Tag.t -> [`Tree ] Node.t = "caml_xml_tree_tagged_descendant" "noalloc"
279 let tagged_descendant t n tag = tree_tagged_descendant t.doc n tag
281 external tree_tagged_next : tree -> [`Tree ] Node.t -> Tag.t -> [`Tree ] Node.t = "caml_xml_tree_tagged_next" "noalloc"
282 let tagged_next t n tag = tree_tagged_next t.doc n tag
284 external tree_select_descendant : tree -> [`Tree ] Node.t -> unordered_set -> [`Tree] Node.t = "caml_xml_tree_select_descendant" "noalloc"
285 let select_descendant t n tag_set = tree_select_descendant t.doc n tag_set
287 external tree_tagged_following_before : tree -> [`Tree ] Node.t -> Tag.t -> [`Tree ] Node.t -> [`Tree ] Node.t = "caml_xml_tree_tagged_following_before" "noalloc"
288 let tagged_following_before t n tag ctx = tree_tagged_following_before t.doc n tag ctx
290 external tree_select_following_before : tree -> [`Tree ] Node.t -> unordered_set -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_select_following_before" "noalloc"
291 let select_following_before t n tag_set ctx = tree_select_following_before t.doc n tag_set ctx
293 external tree_parent : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_parent" "noalloc"
294 let parent t n = tree_parent t.doc n
296 external tree_tag : tree -> [`Tree] Node.t -> Tag.t = "caml_xml_tree_tag" "noalloc"
297 let tag t n = tree_tag t.doc n
299 external tree_is_first_child : tree -> [ `Tree ] Node.t -> bool = "caml_xml_tree_is_first_child" "noalloc"
300 let is_first_child t n = tree_is_first_child t.doc n
302 external tree_is_right_descendant : tree -> [ `Tree ] Node.t -> [`Tree] Node.t -> bool =
303 "caml_xml_tree_is_right_descendant" "noalloc"
305 let is_right_descendant t n1 n2 = tree_is_right_descendant t.doc n1 n2
308 let node_tags t = Ptset.Int.add Tag.document_node t.descendants.(Tag.document_node)
310 let attribute_tags t = t.attributes
312 let element_tags t = t.elements
315 t.children.(tag), t.descendants.(tag), t.siblings.(tag), t.followings.(tag)
318 let dump_tag_table t =
320 let printer ppf set =
321 Logger.print ppf "%s: %a"
322 (Tag.to_string !tag) TagSet.print (TagSet.inj_positive set);
325 let set_printer msg set =
327 Logger.print err_formatter "%s :@\n" msg;
328 Pretty.pp_print_array ~sep:pp_force_newline printer err_formatter set;
329 Logger.print err_formatter "-----------------------------@\n";
331 set_printer "Child tags" t.children;
332 set_printer "Descendant tags" t.descendants;
333 set_printer "Sibling tags" t.siblings;
334 set_printer "Following tags" t.followings
336 external tree_subtree_tags : tree -> [`Tree] Node.t -> Tag.t -> int = "caml_xml_tree_subtree_tags" "noalloc"
337 let subtree_tags t n tag = tree_subtree_tags t.doc n tag
339 external tree_subtree_size : tree -> [`Tree] Node.t -> int = "caml_xml_tree_subtree_size" "noalloc"
340 let subtree_size t n = tree_subtree_size t.doc n
342 let subtree_elements t node =
343 let size = tree_subtree_size t.doc node - 1 in
345 else let size = size - (tree_subtree_tags t.doc node Tag.pcdata) in
346 if size < 2 then size else
347 let acc = ref size in
348 for i = 0 to Array.length t.attribute_array - 1 do
349 acc := !acc - tree_subtree_tags t.doc node t.attribute_array.(i)
353 external tree_closing : tree -> [`Tree] Node.t -> [`Tree] Node.t = "caml_xml_tree_closing" "noalloc"
354 let closing t n = tree_closing t.doc n
356 external tree_num_tags : tree -> int = "caml_xml_tree_num_tags" "noalloc"
357 let num_tags t = tree_num_tags t.doc
359 external tree_size : tree -> int = "caml_xml_tree_size" "noalloc"
360 let size t = tree_size t.doc
365 let rec loop left node acc_d total_d num_leaves =
367 (acc_d+total_d,if left then num_leaves+1 else num_leaves)
369 let d,td = loop true (tree_first_child tree node) (acc_d+1) total_d num_leaves in
370 loop false (tree_next_sibling tree node) (acc_d) d td
372 let a,b = loop true root 0 0 0
374 Logger.print err_formatter "Average depth: %f, number of leaves %i@\n@?" ((float_of_int a)/. (float_of_int b)) b
382 external hash : t -> int = "%identity"
383 external uid : t -> Uid.t = "%identity"
384 external equal : t -> t -> bool = "%eq"
385 external make : t -> int = "%identity"
386 external node : t -> int = "%identity"
387 external stats : unit -> unit = "%identity"
390 let to_ptset s = fold (Ptset.Int.add) s Ptset.Int.empty
394 Hashtbl.Make(struct type t = TagS.t * TagS.t
396 HASHINT2(Uid.to_int x.TagS.Node.id,
397 Uid.to_int y.TagS.Node.id)
404 Hashtbl.Make(struct type t = Tag.t * TagS.t
406 HASHINT2(x, Uid.to_int y.TagS.Node.id)
413 let add_cache = TagTSCache.create 1023
414 let union_cache = TSTSCache.create 1023
415 let subset_cache = TSTSCache.create 1023
418 TSTSCache.clear union_cache;
419 TSTSCache.clear subset_cache;
420 TagTSCache.clear add_cache
423 (x == y) || (x == TagS.empty) ||
424 if y == TagS.empty then false
428 TSTSCache.find subset_cache key
431 let z = TagS.subset x y in
432 TSTSCache.add subset_cache key z; z
434 let order ((x, y) as z) =
435 if x.TagS.Node.id <= y.TagS.Node.id then z
439 if _subset x y then y
440 else if _subset y x then x
442 let key = order (x, y) in
444 TSTSCache.find union_cache key
447 let z = TagS.union x y in
448 TSTSCache.add union_cache key z; z
453 TagTSCache.find add_cache key
456 let z = TagS.add t s in
457 TagTSCache.add add_cache key z;z
459 let child_sibling_labels tree =
460 let table_c = Array.create (tree_num_tags tree) TagS.empty in
461 let table_n = Array.copy table_c in
463 if node == nil then TagS.empty
465 let children = loop (tree_first_child tree node) in
466 let tag = tree_tag tree node in
468 let tc = table_c.(tag) in
469 if _subset children tc then ()
470 else table_c.(tag) <- _union tc children
472 let siblings = loop (tree_next_sibling tree node) in
474 let tn = table_n.(tag) in
475 if _subset siblings tn then ()
476 else table_n.(tag) <- _union tn siblings
483 let descendant_labels tree =
484 let table_d = Array.create (tree_num_tags tree) TagS.empty in
486 if node == nil then TagS.empty else
487 let d1 = loop (tree_first_child tree node) in
488 let d2 = loop (tree_next_sibling tree node) in
489 let tag = tree_tag tree node in
491 let td = table_d.(tag) in
492 if _subset d1 td then ()
493 else table_d.(tag) <- _union td d1;
495 _add tag (_union d1 d2)
500 let collect_labels tree =
501 let table_f = Array.create (tree_num_tags tree) TagS.empty in
502 let table_n = Array.copy table_f in
503 let table_c = Array.copy table_f in
504 let table_d = Array.copy table_f in
505 let rec loop node foll_siblings descendants followings =
506 if node == nil then foll_siblings, descendants, followings else
507 let tag = tree_tag tree node in
509 let tf = table_f.(tag) in
510 if _subset followings tf then ()
511 else table_f.(tag) <- _union tf followings in
513 let tn = table_n.(tag) in
514 if _subset foll_siblings tn then ()
515 else table_n.(tag) <- _union tn foll_siblings in
516 let children, n_descendants, n_followings =
517 loop (tree_last_child tree node) TagS.empty TagS.empty followings
520 let tc = table_c.(tag) in
521 if _subset children tc then ()
522 else table_c.(tag) <- _union tc children
525 let td = table_d.(tag) in
526 if _subset n_descendants td then ()
527 else table_d.(tag) <- _union td n_descendants
529 loop (tree_prev_sibling tree node)
530 (_add tag foll_siblings)
531 (_add tag (_union n_descendants descendants))
532 (_add tag n_followings)
534 ignore (loop root TagS.empty TagS.empty TagS.empty);
535 table_f, table_n, table_c, table_d
538 let is_nil t = t == nil
539 let is_node t = t != nil
540 let is_root t = t == root
543 Logger.print err_formatter "Initializing tag structure@\n";
544 let _ = Tag.init (mk_tag_ops t) in
545 Logger.print err_formatter "Starting tag table construction@\n";
546 let f, n, c, d = time collect_labels t ~msg:"Building tag relationship table" in
547 let c = Array.map TagS.to_ptset c in
548 let n = Array.map TagS.to_ptset n in
549 let f = Array.map TagS.to_ptset f in
550 let d = Array.map TagS.to_ptset d in
551 let () = clear_cache () in
552 let attributes = Ptset.Int.add Tag.attribute d.(Tag.attribute) in
553 let elements = Ptset.Int.add Tag.document_node
554 (Ptset.Int.remove Tag.pcdata
555 (Ptset.Int.diff d.(Tag.document_node) attributes))
558 attributes = attributes;
559 attribute_array = Array.of_list (Ptset.Int.elements attributes);
569 let parse_xml_uri str = node_of_t (TreeBuilder.parse_file str)
570 let parse_xml_string str = node_of_t (TreeBuilder.parse_string str)
572 let size t = tree_size t.doc;;
574 let magic_string = "SXSI_INDEX"
575 let version_string = "3"
578 Unix.lseek fd 0 Unix.SEEK_CUR
580 let pr_pos fd = Logger.print err_formatter "At position %i@\n" (pos fd)
583 let sl = String.length s in
584 let ssl = Printf.sprintf "%020i" sl in
585 ignore (Unix.write fd ssl 0 20);
586 ignore (Unix.write fd s 0 (String.length s))
588 let rec really_read fd buffer start length =
589 if length <= 0 then () else
590 match Unix.read fd buffer start length with
591 0 -> raise End_of_file
592 | r -> really_read fd buffer (start + r) (length - r);;
595 let buffer = String.create 20 in
596 let _ = really_read fd buffer 0 20 in
597 let size = int_of_string buffer in
598 let buffer = String.create size in
599 let _ = really_read fd buffer 0 size in
602 let save_tag_table channel t =
603 let t = Array.map (fun s -> Array.of_list (Ptset.Int.elements s)) t in
604 Marshal.to_channel channel t []
607 let fd = Unix.openfile str [ Unix.O_WRONLY;Unix.O_TRUNC;Unix.O_CREAT] 0o644 in
608 let out_c = Unix.out_channel_of_descr fd in
609 let index_prefix = Filename.chop_suffix str ".srx" in
610 let _ = set_binary_mode_out out_c true in
611 output_string out_c magic_string;
612 output_char out_c '\n';
613 output_string out_c version_string;
614 output_char out_c '\n';
615 save_tag_table out_c t.children;
616 save_tag_table out_c t.siblings;
617 save_tag_table out_c t.descendants;
618 save_tag_table out_c t.followings;
619 (* we need to move the fd to the correct position *)
621 ignore (Unix.lseek fd (pos_out out_c) Unix.SEEK_SET);
622 tree_save t.doc fd index_prefix;
625 let load_tag_table channel =
626 let table : int array array = Marshal.from_channel channel in
627 Array.map (fun a -> Ptset.Int.from_list (Array.to_list a)) table
629 let load ?(sample=64) ?(load_text=true) str =
630 let fd = Unix.openfile str [ Unix.O_RDONLY ] 0o644 in
631 let in_c = Unix.in_channel_of_descr fd in
632 let index_prefix = Filename.chop_suffix str ".srx" in
633 let _ = set_binary_mode_in in_c true in
635 (let ms = input_line in_c in if ms <> magic_string then failwith "Invalid index file");
636 (let vs = input_line in_c in if vs <> version_string then failwith "Invalid version file");
637 let c = load_tag_table in_c in
638 let s = load_tag_table in_c in
639 let d = load_tag_table in_c in
640 let f = load_tag_table in_c in
643 let c, s, d, f = time ~msg:"Loading tag table"(load_table) () in
644 ignore(Unix.lseek fd (pos_in in_c) Unix.SEEK_SET);
645 let xml_tree = tree_load fd index_prefix load_text sample in
646 let () = Tag.init (Obj.magic xml_tree) in
647 let attributes = Ptset.Int.add Tag.attribute d.(Tag.attribute) in
648 let elements = Ptset.Int.add Tag.document_node
649 (Ptset.Int.remove Tag.pcdata
650 (Ptset.Int.diff d.(Tag.document_node) attributes))
652 let tree = { doc = xml_tree;
653 attributes = attributes;
654 attribute_array = Array.of_list (Ptset.Int.elements attributes);
667 let equal a b = a == b
671 | i -> Printf.sprintf "Node (%i)" i
673 let dump_node t = nts (Node.to_int t)
677 type query_result = { bv : bit_vector;
681 external tree_flush : tree -> Unix.file_descr -> unit = "caml_xml_tree_flush"
682 let flush t fd = tree_flush t.doc fd
684 external text_prefix : tree -> string -> bool -> query_result = "caml_text_collection_prefix_bv"
685 let text_prefix t s b = text_prefix t.doc s b
687 external text_suffix : tree -> string -> bool -> query_result = "caml_text_collection_suffix_bv"
688 let text_suffix t s b = text_suffix t.doc s b
690 external text_equals : tree -> string -> bool -> query_result = "caml_text_collection_equals_bv"
691 let text_equals t s b = text_equals t.doc s b
693 external text_contains : tree -> string -> bool -> query_result = "caml_text_collection_contains_bv"
694 let text_contains t s b = text_contains t.doc s b
697 module Predicate = Hcons.Make (
700 type t = (_t -> node -> bool) ref
701 let hash t = Hashtbl.hash t
702 let equal t1 t2 = t1 == t2
705 let string_of_query query =
707 | `Prefix -> "starts-with"
708 | `Suffix -> "ends-with"
709 | `Equals -> "equals"
710 | `Contains -> "contains"
713 let query_fun = function
714 | `Prefix -> text_prefix
715 | `Suffix -> text_suffix
716 | `Equals -> text_equals
717 | `Contains -> text_contains
720 let _pred_cache = Hashtbl.create 17
722 let mk_pred query s =
723 let f = query_fun query in
724 let memo = ref (fun _ _ -> failwith "Undefined") in
725 memo := begin fun tree node ->
727 try Hashtbl.find _pred_cache (query,s) with
729 time ~count:1 ~msg:(Printf.sprintf "Computing text query %s(%s)"
730 (string_of_query query) s)
733 let bv = results.bv in
734 memo := begin fun _ n ->
735 bit_vector_unsafe_get bv (Node.to_int n)
737 bit_vector_unsafe_get bv (Node.to_int node)
742 let full_text_prefix t s = (text_prefix t s true).pos
744 let full_text_suffix t s = (text_suffix t s true).pos
746 let full_text_equals t s = (text_equals t s true).pos
748 let full_text_contains t s = (text_contains t s true).pos
750 let full_text_query q t s =
751 let res = (query_fun q) t s true in
752 Hashtbl.replace _pred_cache (q,s) res;