X-Git-Url: http://git.nguyen.vg/gitweb/?p=tatoo.git;a=blobdiff_plain;f=src%2Fauto%2Fata.ml;h=1015513f2563db0d5ebc669c5d47681c3a8f3004;hp=5001ebc2c4505cbba23c154e714c45fa9f5140d7;hb=ce09a30489dce8ac9e389c8c1525a34d1e02354e;hpb=30bc0bb1291426e5e26eb2dee1ffc41e4c246349 diff --git a/src/auto/ata.ml b/src/auto/ata.ml index 5001ebc..1015513 100644 --- a/src/auto/ata.ml +++ b/src/auto/ata.ml @@ -14,126 +14,216 @@ (***********************************************************************) (* - Time-stamp: + Time-stamp: *) +INCLUDE "utils.ml" open Format open Utils -type move = [ `Left | `Right | `Up1 | `Up2 | `Epsilon ] -type state_ctx = { left : StateSet.t; - right : StateSet.t; - up1 : StateSet.t; - up2 : StateSet.t; - epsilon : StateSet.t} -type ctx_ = { mutable positive : state_ctx; - mutable negative : state_ctx } -type pred_ = move * bool * State.t - -module Move : (Formula.PREDICATE with type data = pred_ and type ctx = ctx_ ) = +type predicate = | First_child + | Next_sibling + | Parent + | Previous_sibling + | Stay + | Is_first_child + | Is_next_sibling + | Is of (Tree.Common.NodeKind.t) + | Has_first_child + | Has_next_sibling + +let is_move p = match p with +| First_child | Next_sibling +| Parent | Previous_sibling | Stay -> true +| _ -> false + + +type atom = predicate * bool * State.t + +module Atom : (Formula.ATOM with type data = atom) = struct module Node = struct - type t = move * bool * State.t + type t = atom let equal n1 n2 = n1 = n2 let hash n = Hashtbl.hash n end - type ctx = ctx_ - let make_ctx a b c d e = - { left = a; right = b; up1 = c; up2 = d; epsilon = e } - include Hcons.Make(Node) let print ppf a = - let _ = flush_str_formatter() in - let fmt = str_formatter in - - let m, b, s = a.node in - let dir,num = - match m with - | `Left -> Pretty.down_arrow, Pretty.subscript 1 - | `Right -> Pretty.down_arrow, Pretty.subscript 2 - | `Epsilon -> Pretty.epsilon, "" - | `Up1 -> Pretty.up_arrow, Pretty.subscript 1 - | `Up2 -> Pretty.up_arrow, Pretty.subscript 2 - in - fprintf fmt "%s%s" dir num; - State.print fmt s; - let str = flush_str_formatter() in - if b then fprintf ppf "%s" str - else Pretty.pp_overline ppf str - - let neg p = - let l, b, s = p.node in - make (l, not b, s) - - let eval ctx p = - let l, b, s = p.node in - let nctx = if b then ctx.positive else ctx.negative in - StateSet.mem s begin - match l with - `Left -> nctx.left - | `Right -> nctx.right - | `Up1 -> nctx.up1 - | `Up2 -> nctx.up2 - | `Epsilon -> nctx.epsilon - end + let p, b, q = a.node in + if not b then fprintf ppf "%s" Pretty.lnot; + match p with + | First_child -> fprintf ppf "FC(%a)" State.print q + | Next_sibling -> fprintf ppf "NS(%a)" State.print q + | Parent -> fprintf ppf "FC%s(%a)" Pretty.inverse State.print q + | Previous_sibling -> fprintf ppf "NS%s(%a)" Pretty.inverse State.print q + | Stay -> fprintf ppf "%s(%a)" Pretty.epsilon State.print q + | Is_first_child -> fprintf ppf "FC%s?" Pretty.inverse + | Is_next_sibling -> fprintf ppf "NS%s?" Pretty.inverse + | Is k -> fprintf ppf "is-%a?" Tree.Common.NodeKind.print k + | Has_first_child -> fprintf ppf "FC?" + | Has_next_sibling -> fprintf ppf "NS?" + + let neg a = + let p, b, q = a.node in + make (p, not b, q) + + +end + +module SFormula = +struct + include Formula.Make(Atom) + open Tree.Common.NodeKind + let mk_atom a b c = atom_ (Atom.make (a,b,c)) + let mk_kind k = mk_atom (Is k) true State.dummy + let has_first_child = + (mk_atom Has_first_child true State.dummy) + + let has_next_sibling = + (mk_atom Has_next_sibling true State.dummy) + + let is_first_child = + (mk_atom Is_first_child true State.dummy) + + let is_next_sibling = + (mk_atom Is_next_sibling true State.dummy) + + let is_attribute = + (mk_atom (Is Attribute) true State.dummy) + + let is_element = + (mk_atom (Is Element) true State.dummy) + + let is_processing_instruction = + (mk_atom (Is ProcessingInstruction) true State.dummy) + + let is_comment = + (mk_atom (Is Comment) true State.dummy) + + let first_child q = + and_ + (mk_atom First_child true q) + has_first_child + + let next_sibling q = + and_ + (mk_atom Next_sibling true q) + has_next_sibling + + let parent q = + and_ + (mk_atom Parent true q) + is_first_child + + let previous_sibling q = + and_ + (mk_atom Previous_sibling true q) + is_next_sibling + + let stay q = + (mk_atom Stay true q) + + let get_states phi = + fold (fun phi acc -> + match expr phi with + | Formula.Atom a -> let _, _, q = Atom.node a in + if q != State.dummy then StateSet.add q acc else acc + | _ -> acc + ) phi StateSet.empty + end -module SFormula = Formula.Make(Move) type t = { id : Uid.t; mutable states : StateSet.t; - mutable top_states : StateSet.t; - mutable bottom_states: StateSet.t; mutable selection_states: StateSet.t; transitions: (State.t, (QNameSet.t*SFormula.t) list) Hashtbl.t; } - - let next = Uid.make_maker () let create () = { id = next (); states = StateSet.empty; - top_states = StateSet.empty; - bottom_states = StateSet.empty; selection_states = StateSet.empty; transitions = Hashtbl.create 17; } + +module Transition = Hcons.Make (struct + type t = State.t * QNameSet.t * SFormula.t + let equal (a, b, c) (d, e, f) = + a == d && b == e && c == f + let hash (a, b, c) = + HASHINT4 (PRIME1, a, ((QNameSet.uid b) :> int), ((SFormula.uid c) :> int)) +end) + +module TransList : sig + include Hlist.S with type elt = Transition.t + val print : Format.formatter -> t -> unit +end = + struct + include Hlist.Make(Transition) + let print ppf l = + iter (fun t -> + let q, lab, f = Transition.node t in + fprintf ppf "%a, %a -> %a
" State.print q QNameSet.print lab SFormula.print f) l + end + +let get_trans a states tag = + StateSet.fold (fun q acc0 -> + try + let trs = Hashtbl.find a.transitions q in + List.fold_left (fun acc1 (labs, phi) -> + if QNameSet.mem tag labs then TransList.cons (Transition.make (q, labs, phi)) acc1 else acc1) acc0 trs + with Not_found -> acc0 + ) states TransList.nil + +(* + [add_trans a q labels f] adds a transition [(q,labels) -> f] to the + automaton [a] but ensures that transitions remains pairwise disjoint +*) + let add_trans a q s f = let trs = try Hashtbl.find a.transitions q with Not_found -> [] in - let rem, ntrs = - List.fold_left (fun (rem, atrs) ((labs, phi) as tr) -> - let nlabs = QNameSet.inter labs rem in - if QNameSet.is_empty nlabs then - (rem, tr :: atrs) - else - let nrem = QNameSet.diff rem labs in - nrem, (nlabs, SFormula.or_ phi f)::atrs - ) (s, []) trs + let cup, ntrs = + List.fold_left (fun (acup, atrs) (labs, phi) -> + let lab1 = QNameSet.inter labs s in + let lab2 = QNameSet.diff labs s in + let tr1 = + if QNameSet.is_empty lab1 then [] + else [ (lab1, SFormula.or_ phi f) ] + in + let tr2 = + if QNameSet.is_empty lab2 then [] + else [ (lab2, SFormula.or_ phi f) ] + in + (QNameSet.union acup labs, tr1@ tr2 @ atrs) + ) (QNameSet.empty, []) trs in + let rem = QNameSet.diff s cup in let ntrs = if QNameSet.is_empty rem then ntrs else (rem, f) :: ntrs in Hashtbl.replace a.transitions q ntrs +let _pr_buff = Buffer.create 50 +let _str_fmt = formatter_of_buffer _pr_buff +let _flush_str_fmt () = pp_print_flush _str_fmt (); + let s = Buffer.contents _pr_buff in + Buffer.clear _pr_buff; s let print fmt a = fprintf fmt - "Unique ID: %i@\n\ - States %a@\n\ - Top states: %a@\n\ - Bottom states: %a@\n\ + "\nInternal UID: %i@\n\ + States: %a@\n\ Selection states: %a@\n\ Alternating transitions:@\n" (a.id :> int) StateSet.print a.states - StateSet.print a.top_states - StateSet.print a.bottom_states StateSet.print a.selection_states; let trs = Hashtbl.fold @@ -141,22 +231,137 @@ let print fmt a = a.transitions [] in - let sorted_trs = List.stable_sort (fun (q1, s1, phi1) (q2, s2, phi2) -> + let sorted_trs = List.stable_sort (fun (q1, s1, _) (q2, s2, _) -> let c = State.compare q1 q2 in - (if c == 0 then QNameSet.compare s1 s2 else c)) trs in - let sfmt = str_formatter in - let _ = flush_str_formatter () in - let strs_strings, maxs = List.fold_left (fun (accl, accm) (q, s, f) -> - let s1 = State.print sfmt q; flush_str_formatter () in - let s2 = QNameSet.print sfmt s; flush_str_formatter () in - let s3 = SFormula.print sfmt f; flush_str_formatter () in - ( (s1, s2, s3) :: accl, - max - accm (2 + String.length s1 + String.length s2)) - ) ([], 0) sorted_trs + let _ = _flush_str_fmt () in + let strs_strings, max_pre, max_all = List.fold_left (fun (accl, accp, acca) (q, s, f) -> + let s1 = State.print _str_fmt q; _flush_str_fmt () in + let s2 = QNameSet.print _str_fmt s; _flush_str_fmt () in + let s3 = SFormula.print _str_fmt f; _flush_str_fmt () in + let pre = Pretty.length s1 + Pretty.length s2 in + let all = Pretty.length s3 in + ( (q, s1, s2, s3) :: accl, max accp pre, max acca all) + ) ([], 0, 0) sorted_trs + in + let line = Pretty.line (max_all + max_pre + 6) in + let prev_q = ref State.dummy in + List.iter (fun (q, s1, s2, s3) -> + if !prev_q != q && !prev_q != State.dummy then fprintf fmt " %s\n%!" line; + prev_q := q; + fprintf fmt " %s, %s" s1 s2; + fprintf fmt "%s" (Pretty.padding (max_pre - Pretty.length s1 - Pretty.length s2)); + fprintf fmt " %s %s@\n%!" Pretty.right_arrow s3; + ) strs_strings; + fprintf fmt " %s\n%!" line + +(* + [complete transitions a] ensures that for each state q + and each symbols s in the alphabet, a transition q, s exists. + (adding q, s -> F when necessary). +*) + +let complete_transitions a = + StateSet.iter (fun q -> + let qtrans = Hashtbl.find a.transitions q in + let rem = + List.fold_left (fun rem (labels, _) -> + QNameSet.diff rem labels) QNameSet.any qtrans + in + let nqtrans = + if QNameSet.is_empty rem then qtrans + else + (rem, SFormula.false_) :: qtrans + in + Hashtbl.replace a.transitions q nqtrans + ) a.states + +let cleanup_states a = + let memo = ref StateSet.empty in + let rec loop q = + if not (StateSet.mem q !memo) then begin + memo := StateSet.add q !memo; + let trs = try Hashtbl.find a.transitions q with Not_found -> [] in + List.iter (fun (_, phi) -> + StateSet.iter loop (SFormula.get_states phi)) trs + end + in + StateSet.iter loop a.selection_states; + let unused = StateSet.diff a.states !memo in + eprintf "Unused states %a\n%!" StateSet.print unused; + StateSet.iter (fun q -> Hashtbl.remove a.transitions q) unused; + a.states <- !memo + +(* [normalize_negations a] removes negative atoms in the formula + complementing the sub-automaton in the negative states. + [TODO check the meaning of negative upward arrows] +*) + +let normalize_negations auto = + eprintf "Automaton before normalize_trans:\n"; + print err_formatter auto; + eprintf "--------------------\n%!"; + + let memo_state = Hashtbl.create 17 in + let todo = Queue.create () in + let rec flip b f = + match SFormula.expr f with + Formula.True | Formula.False -> if b then f else SFormula.not_ f + | Formula.Or(f1, f2) -> (if b then SFormula.or_ else SFormula.and_)(flip b f1) (flip b f2) + | Formula.And(f1, f2) -> (if b then SFormula.and_ else SFormula.or_)(flip b f1) (flip b f2) + | Formula.Atom(a) -> begin + let l, b', q = Atom.node a in + if q == State.dummy then if b then f else SFormula.not_ f + else + if b == b' then begin + (* a appears positively, either no negation or double negation *) + if not (Hashtbl.mem memo_state (q,b)) then Queue.add (q,true) todo; + SFormula.atom_ (Atom.make (l, true, q)) + end else begin + (* need to reverse the atom + either we have a positive state deep below a negation + or we have a negative state in a positive formula + b' = sign of the state + b = sign of the enclosing formula + *) + let not_q = + try + (* does the inverted state of q exist ? *) + Hashtbl.find memo_state (q, false) + with + Not_found -> + (* create a new state and add it to the todo queue *) + let nq = State.make () in + auto.states <- StateSet.add nq auto.states; + Hashtbl.add memo_state (q, false) nq; + Queue.add (q, false) todo; nq + in + SFormula.atom_ (Atom.make (l, true, not_q)) + end + end in - List.iter (fun (s1, s2, s3) -> - fprintf fmt "%s, %s" s1 s2; - fprintf fmt "%s" (Pretty.padding (maxs - String.length s1 - String.length s2 - 2)); - fprintf fmt "%s %s@\n" Pretty.right_arrow s3) strs_strings + (* states that are not reachable from a selection stat are not interesting *) + StateSet.iter (fun q -> Queue.add (q, true) todo) auto.selection_states; + + while not (Queue.is_empty todo) do + let (q, b) as key = Queue.pop todo in + let q' = + try + Hashtbl.find memo_state key + with + Not_found -> + let nq = if b then q else + let nq = State.make () in + auto.states <- StateSet.add nq auto.states; + nq + in + Hashtbl.add memo_state key nq; nq + in + let trans = Hashtbl.find auto.transitions q in + let trans' = List.map (fun (lab, f) -> lab, flip b f) trans in + Hashtbl.replace auto.transitions q' trans'; + done; + cleanup_states auto + +