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 selecting_states: StateSet.t;
146 transitions: (State.t, (QNameSet.t*Formula.t) list) Hashtbl.t;
151 let get_states a = a.states
152 let get_selecting_states a = a.selecting_states
154 let get_trans a tag states =
155 StateSet.fold (fun q acc0 ->
157 let trs = Hashtbl.find a.transitions q in
158 List.fold_left (fun acc1 (labs, phi) ->
159 if QNameSet.mem tag labs then
160 TransList.cons (Transition.make (q, labs, phi)) acc1
162 with Not_found -> acc0
163 ) states TransList.nil
167 let _pr_buff = Buffer.create 50
168 let _str_fmt = formatter_of_buffer _pr_buff
169 let _flush_str_fmt () = pp_print_flush _str_fmt ();
170 let s = Buffer.contents _pr_buff in
171 Buffer.clear _pr_buff; s
175 "Internal UID: %i@\n\
177 Selection states: %a@\n\
178 Alternating transitions:@\n"
180 StateSet.print a.states
181 StateSet.print a.selecting_states;
184 (fun q t acc -> List.fold_left (fun acc (s , f) -> (q,s,f)::acc) acc t)
188 let sorted_trs = List.stable_sort (fun (q1, s1, _) (q2, s2, _) ->
189 let c = State.compare q1 q2 in - (if c == 0 then QNameSet.compare s1 s2 else c))
192 let _ = _flush_str_fmt () in
193 let strs_strings, max_pre, max_all = List.fold_left (fun (accl, accp, acca) (q, s, f) ->
194 let s1 = State.print _str_fmt q; _flush_str_fmt () in
195 let s2 = QNameSet.print _str_fmt s; _flush_str_fmt () in
196 let s3 = Formula.print _str_fmt f; _flush_str_fmt () in
197 let pre = Pretty.length s1 + Pretty.length s2 in
198 let all = Pretty.length s3 in
199 ( (q, s1, s2, s3) :: accl, max accp pre, max acca all)
200 ) ([], 0, 0) sorted_trs
202 let line = Pretty.line (max_all + max_pre + 6) in
203 let prev_q = ref State.dummy in
204 fprintf fmt "%s@\n" line;
205 List.iter (fun (q, s1, s2, s3) ->
206 if !prev_q != q && !prev_q != State.dummy then fprintf fmt "%s@\n" line;
208 fprintf fmt "%s, %s" s1 s2;
209 fprintf fmt "%s" (Pretty.padding (max_pre - Pretty.length s1 - Pretty.length s2));
210 fprintf fmt " %s %s@\n" Pretty.right_arrow s3;
212 fprintf fmt "%s@\n" line
215 [complete transitions a] ensures that for each state q
216 and each symbols s in the alphabet, a transition q, s exists.
217 (adding q, s -> F when necessary).
220 let complete_transitions a =
221 StateSet.iter (fun q ->
222 let qtrans = Hashtbl.find a.transitions q in
224 List.fold_left (fun rem (labels, _) ->
225 QNameSet.diff rem labels) QNameSet.any qtrans
228 if QNameSet.is_empty rem then qtrans
230 (rem, Formula.false_) :: qtrans
232 Hashtbl.replace a.transitions q nqtrans
235 let cleanup_states a =
236 let memo = ref StateSet.empty in
238 if not (StateSet.mem q !memo) then begin
239 memo := StateSet.add q !memo;
240 let trs = try Hashtbl.find a.transitions q with Not_found -> [] in
241 List.iter (fun (_, phi) ->
242 StateSet.iter loop (Formula.get_states phi)) trs
245 StateSet.iter loop a.selecting_states;
246 let unused = StateSet.diff a.states !memo in
247 StateSet.iter (fun q -> Hashtbl.remove a.transitions q) unused;
250 (* [normalize_negations a] removes negative atoms in the formula
251 complementing the sub-automaton in the negative states.
252 [TODO check the meaning of negative upward arrows]
255 let normalize_negations auto =
256 let memo_state = Hashtbl.create 17 in
257 let todo = Queue.create () in
259 match Formula.expr f with
260 Boolean.True | Boolean.False -> if b then f else Formula.not_ f
261 | Boolean.Or(f1, f2) -> (if b then Formula.or_ else Formula.and_)(flip b f1) (flip b f2)
262 | Boolean.And(f1, f2) -> (if b then Formula.and_ else Formula.or_)(flip b f1) (flip b f2)
263 | Boolean.Atom(a, b') -> begin
264 match a.Atom.node with
266 if b == b' then begin
267 (* a appears positively, either no negation or double negation *)
268 if not (Hashtbl.mem memo_state (q,b)) then Queue.add (q,true) todo;
269 Formula.mk_atom (Move(m, q))
271 (* need to reverse the atom
272 either we have a positive state deep below a negation
273 or we have a negative state in a positive formula
274 b' = sign of the state
275 b = sign of the enclosing formula
279 (* does the inverted state of q exist ? *)
280 Hashtbl.find memo_state (q, false)
283 (* create a new state and add it to the todo queue *)
284 let nq = State.make () in
285 auto.states <- StateSet.add nq auto.states;
286 Hashtbl.add memo_state (q, false) nq;
287 Queue.add (q, false) todo; nq
289 Formula.mk_atom (Move (m,not_q))
291 | _ -> if b then f else Formula.not_ f
294 (* states that are not reachable from a selection stat are not interesting *)
295 StateSet.iter (fun q -> Queue.add (q, true) todo) auto.selecting_states;
297 while not (Queue.is_empty todo) do
298 let (q, b) as key = Queue.pop todo in
301 Hashtbl.find memo_state key
304 let nq = if b then q else
305 let nq = State.make () in
306 auto.states <- StateSet.add nq auto.states;
309 Hashtbl.add memo_state key nq; nq
311 let trans = Hashtbl.find auto.transitions q in
312 let trans' = List.map (fun (lab, f) -> lab, flip b f) trans in
313 Hashtbl.replace auto.transitions q' trans';
322 let next = Uid.make_maker ()
328 states = StateSet.empty;
329 selecting_states = StateSet.empty;
330 transitions = Hashtbl.create MED_H_SIZE;
337 Cache.N2.iteri (fun _ _ _ b -> if b then incr n2) auto.cache2;
338 Cache.N4.iteri (fun _ _ _ _ _ b -> if b then incr n4) auto.cache4;
339 Logger.msg `STATS "automaton %i, cache2: %i entries, cache6: %i entries"
340 (auto.id :> int) !n2 !n4;
341 let c2l, c2u = Cache.N2.stats auto.cache2 in
342 let c4l, c4u = Cache.N4.stats auto.cache4 in
344 "cache2: length: %i, used: %i, occupation: %f"
345 c2l c2u (float c2u /. float c2l);
347 "cache4: length: %i, used: %i, occupation: %f"
348 c4l c4u (float c4u /. float c4l)
353 let add_state a ?(selecting=false) q =
354 a.states <- StateSet.add q a.states;
355 if selecting then a.selecting_states <- StateSet.add q a.selecting_states
357 let add_trans a q s f =
358 if not (StateSet.mem q a.states) then add_state a q;
359 let trs = try Hashtbl.find a.transitions q with Not_found -> [] in
361 List.fold_left (fun (acup, atrs) (labs, phi) ->
362 let lab1 = QNameSet.inter labs s in
363 let lab2 = QNameSet.diff labs s in
365 if QNameSet.is_empty lab1 then []
366 else [ (lab1, Formula.or_ phi f) ]
369 if QNameSet.is_empty lab2 then []
370 else [ (lab2, Formula.or_ phi f) ]
372 (QNameSet.union acup labs, tr1@ tr2 @ atrs)
373 ) (QNameSet.empty, []) trs
375 let rem = QNameSet.diff s cup in
376 let ntrs = if QNameSet.is_empty rem then ntrs
377 else (rem, f) :: ntrs
379 Hashtbl.replace a.transitions q ntrs
382 complete_transitions a;
383 normalize_negations a;