1 (***********************************************************************)
5 (* Kim Nguyen, LRI UMR8623 *)
6 (* Université Paris-Sud & CNRS *)
8 (* Copyright 2010-2013 Université Paris-Sud and Centre National de la *)
9 (* Recherche Scientifique. All rights reserved. This file is *)
10 (* distributed under the terms of the GNU Lesser General Public *)
11 (* License, with the special exception on linking described in file *)
14 (***********************************************************************)
19 type move = [ `First_child
25 type predicate = Move of move * State.t
28 | Is of Tree.NodeKind.t
38 let equal n1 n2 = n1 = n2
39 let hash n = Hashtbl.hash n
42 include Hcons.Make(Node)
46 | Move (m, q) -> begin
48 `First_child -> fprintf ppf "%s" Pretty.down_arrow
49 | `Next_sibling -> fprintf ppf "%s" Pretty.right_arrow
50 | `Parent -> fprintf ppf "%s" Pretty.up_arrow
51 | `Previous_sibling -> fprintf ppf "%s" Pretty.left_arrow
52 | `Stay -> fprintf ppf "%s" Pretty.bullet
54 fprintf ppf "%a" State.print q
55 | Is_first_child -> fprintf ppf "%s?" Pretty.up_arrow
56 | Is_next_sibling -> fprintf ppf "%s?" Pretty.left_arrow
57 | Is k -> fprintf ppf "is-%a?" Tree.NodeKind.print k
58 | Has_first_child -> fprintf ppf "%s?" Pretty.down_arrow
59 | Has_next_sibling -> fprintf ppf "%s?" Pretty.right_arrow
66 include Boolean.Make(Atom)
68 let mk_atom a = atom_ (Atom.make a)
69 let is k = mk_atom (Is k)
71 let has_first_child = mk_atom Has_first_child
73 let has_next_sibling = mk_atom Has_next_sibling
75 let is_first_child = mk_atom Is_first_child
77 let is_next_sibling = mk_atom Is_next_sibling
79 let is_attribute = mk_atom (Is Attribute)
81 let is_element = mk_atom (Is Element)
83 let is_processing_instruction = mk_atom (Is ProcessingInstruction)
85 let is_comment = mk_atom (Is Comment)
87 let mk_move m q = mk_atom (Move(m,q))
90 (mk_move `First_child q)
95 (mk_move `Next_sibling q)
103 let previous_sibling q =
105 (mk_move `Previous_sibling q)
108 let stay q = mk_move `Stay q
113 | Boolean.Atom ({ Atom.node = Move(_,q) ; _ }, _) -> StateSet.add q acc
119 module Transition = Hcons.Make (struct
120 type t = State.t * QNameSet.t * Formula.t
121 let equal (a, b, c) (d, e, f) =
122 a == d && b == e && c == f
124 HASHINT4 (PRIME1, a, ((QNameSet.uid b) :> int), ((Formula.uid c) :> int))
128 module TransList : sig
129 include Hlist.S with type elt = Transition.t
130 val print : Format.formatter -> ?sep:string -> t -> unit
133 include Hlist.Make(Transition)
134 let print ppf ?(sep="\n") l =
136 let q, lab, f = Transition.node t in
137 fprintf ppf "%a, %a -> %a%s" State.print q QNameSet.print lab Formula.print f sep) l
144 mutable states : StateSet.t;
145 mutable starting_states : StateSet.t;
146 mutable selecting_states: StateSet.t;
147 transitions: (State.t, (QNameSet.t*Formula.t) list) Hashtbl.t;
152 let get_states a = a.states
153 let get_starting_states a = a.starting_states
154 let get_selecting_states a = a.selecting_states
156 let get_trans a tag states =
157 StateSet.fold (fun q acc0 ->
159 let trs = Hashtbl.find a.transitions q in
160 List.fold_left (fun acc1 (labs, phi) ->
161 if QNameSet.mem tag labs then
162 TransList.cons (Transition.make (q, labs, phi)) acc1
164 with Not_found -> acc0
165 ) states TransList.nil
169 let _pr_buff = Buffer.create 50
170 let _str_fmt = formatter_of_buffer _pr_buff
171 let _flush_str_fmt () = pp_print_flush _str_fmt ();
172 let s = Buffer.contents _pr_buff in
173 Buffer.clear _pr_buff; s
177 "Internal UID: %i@\n\
179 Starting states: %a@\n\
180 Selection states: %a@\n\
181 Alternating transitions:@\n"
183 StateSet.print a.states
184 StateSet.print a.starting_states
185 StateSet.print a.selecting_states;
188 (fun q t acc -> List.fold_left (fun acc (s , f) -> (q,s,f)::acc) acc t)
192 let sorted_trs = List.stable_sort (fun (q1, s1, _) (q2, s2, _) ->
193 let c = State.compare q1 q2 in - (if c == 0 then QNameSet.compare s1 s2 else c))
196 let _ = _flush_str_fmt () in
197 let strs_strings, max_pre, max_all = List.fold_left (fun (accl, accp, acca) (q, s, f) ->
198 let s1 = State.print _str_fmt q; _flush_str_fmt () in
199 let s2 = QNameSet.print _str_fmt s; _flush_str_fmt () in
200 let s3 = Formula.print _str_fmt f; _flush_str_fmt () in
201 let pre = Pretty.length s1 + Pretty.length s2 in
202 let all = Pretty.length s3 in
203 ( (q, s1, s2, s3) :: accl, max accp pre, max acca all)
204 ) ([], 0, 0) sorted_trs
206 let line = Pretty.line (max_all + max_pre + 6) in
207 let prev_q = ref State.dummy in
208 fprintf fmt "%s@\n" line;
209 List.iter (fun (q, s1, s2, s3) ->
210 if !prev_q != q && !prev_q != State.dummy then fprintf fmt "%s@\n" line;
212 fprintf fmt "%s, %s" s1 s2;
213 fprintf fmt "%s" (Pretty.padding (max_pre - Pretty.length s1 - Pretty.length s2));
214 fprintf fmt " %s %s@\n" Pretty.right_arrow s3;
216 fprintf fmt "%s@\n" line
219 [complete transitions a] ensures that for each state q
220 and each symbols s in the alphabet, a transition q, s exists.
221 (adding q, s -> F when necessary).
224 let complete_transitions a =
225 StateSet.iter (fun q ->
226 if StateSet.mem q a.starting_states then ()
228 let qtrans = try Hashtbl.find a.transitions q with Not_found -> eprintf "Not found here 226\n%!"; raise Not_found in
230 List.fold_left (fun rem (labels, _) ->
231 QNameSet.diff rem labels) QNameSet.any qtrans
234 if QNameSet.is_empty rem then qtrans
236 (rem, Formula.false_) :: qtrans
238 Hashtbl.replace a.transitions q nqtrans
241 (* [cleanup_states] remove states that do not lead to a
244 let cleanup_states a =
245 let memo = ref StateSet.empty in
247 if not (StateSet.mem q !memo) then begin
248 memo := StateSet.add q !memo;
249 let trs = try Hashtbl.find a.transitions q with Not_found -> [] in
250 List.iter (fun (_, phi) ->
251 StateSet.iter loop (Formula.get_states phi)) trs
254 StateSet.iter loop a.selecting_states;
255 let unused = StateSet.diff a.states !memo in
256 StateSet.iter (fun q -> Hashtbl.remove a.transitions q) unused;
259 (* [normalize_negations a] removes negative atoms in the formula
260 complementing the sub-automaton in the negative states.
261 [TODO check the meaning of negative upward arrows]
264 let normalize_negations auto =
265 let memo_state = Hashtbl.create 17 in
266 let todo = Queue.create () in
268 match Formula.expr f with
269 Boolean.True | Boolean.False -> if b then f else Formula.not_ f
270 | Boolean.Or(f1, f2) -> (if b then Formula.or_ else Formula.and_)(flip b f1) (flip b f2)
271 | Boolean.And(f1, f2) -> (if b then Formula.and_ else Formula.or_)(flip b f1) (flip b f2)
272 | Boolean.Atom(a, b') -> begin
273 match a.Atom.node with
275 if b == b' then begin
276 (* a appears positively, either no negation or double negation *)
277 if not (Hashtbl.mem memo_state (q,b)) then Queue.add (q,true) todo;
278 Formula.mk_atom (Move(m, q))
280 (* need to reverse the atom
281 either we have a positive state deep below a negation
282 or we have a negative state in a positive formula
283 b' = sign of the state
284 b = sign of the enclosing formula
288 (* does the inverted state of q exist ? *)
289 Hashtbl.find memo_state (q, false)
292 (* create a new state and add it to the todo queue *)
293 let nq = State.make () in
294 auto.states <- StateSet.add nq auto.states;
295 Hashtbl.add memo_state (q, false) nq;
296 Queue.add (q, false) todo; nq
298 Formula.mk_atom (Move (m,not_q))
300 | _ -> if b then f else Formula.not_ f
303 (* states that are not reachable from a selection stat are not interesting *)
304 StateSet.iter (fun q -> Queue.add (q, true) todo) auto.selecting_states;
306 while not (Queue.is_empty todo) do
307 let (q, b) as key = Queue.pop todo in
310 Hashtbl.find memo_state key
313 let nq = if b then q else
314 let nq = State.make () in
315 auto.states <- StateSet.add nq auto.states;
318 Hashtbl.add memo_state key nq; nq
320 let trans = try Hashtbl.find auto.transitions q with Not_found -> eprintf "Not_found here 318\n%!"; [] in
321 let trans' = List.map (fun (lab, f) -> lab, flip b f) trans in
322 Hashtbl.replace auto.transitions q' trans';
331 let next = Uid.make_maker ()
337 states = StateSet.empty;
338 starting_states = StateSet.empty;
339 selecting_states = StateSet.empty;
340 transitions = Hashtbl.create MED_H_SIZE;
347 Cache.N2.iteri (fun _ _ _ b -> if b then incr n2) auto.cache2;
348 Cache.N4.iteri (fun _ _ _ _ _ b -> if b then incr n4) auto.cache4;
349 Logger.msg `STATS "automaton %i, cache2: %i entries, cache6: %i entries"
350 (auto.id :> int) !n2 !n4;
351 let c2l, c2u = Cache.N2.stats auto.cache2 in
352 let c4l, c4u = Cache.N4.stats auto.cache4 in
354 "cache2: length: %i, used: %i, occupation: %f"
355 c2l c2u (float c2u /. float c2l);
357 "cache4: length: %i, used: %i, occupation: %f"
358 c4l c4u (float c4u /. float c4l)
363 let add_state a ?(starting=false) ?(selecting=false) q =
364 a.states <- StateSet.add q a.states;
365 if starting then a.starting_states <- StateSet.add q a.starting_states;
366 if selecting then a.selecting_states <- StateSet.add q a.selecting_states
368 let add_trans a q s f =
369 if not (StateSet.mem q a.states) then add_state a q;
370 let trs = try Hashtbl.find a.transitions q with Not_found -> [] in
372 List.fold_left (fun (acup, atrs) (labs, phi) ->
373 let lab1 = QNameSet.inter labs s in
374 let lab2 = QNameSet.diff labs s in
376 if QNameSet.is_empty lab1 then []
377 else [ (lab1, Formula.or_ phi f) ]
380 if QNameSet.is_empty lab2 then []
381 else [ (lab2, Formula.or_ phi f) ]
383 (QNameSet.union acup labs, tr1@ tr2 @ atrs)
384 ) (QNameSet.empty, []) trs
386 let rem = QNameSet.diff s cup in
387 let ntrs = if QNameSet.is_empty rem then ntrs
388 else (rem, f) :: ntrs
390 Hashtbl.replace a.transitions q ntrs
393 complete_transitions a;
394 normalize_negations a;