X-Git-Url: http://git.nguyen.vg/gitweb/?a=blobdiff_plain;ds=sidebyside;f=ata.ml;h=bb38863bad0285e25b696f1dfe85528557bf2854;hb=22e128466565745a4e74f3b9823e7884ee5c6157;hp=fc29e98fc96c916a05a8710a675a6d08a34a0231;hpb=4ef83e49878441a5146af5326c1b49ced4002756;p=SXSI%2Fxpathcomp.git diff --git a/ata.ml b/ata.ml index fc29e98..bb38863 100644 --- a/ata.ml +++ b/ata.ml @@ -1,12 +1,29 @@ -(* Todo refactor and remove this alias *) INCLUDE "debug.ml" -module Tree = Tree.Binary +INCLUDE "utils.ml" +type jump_kind = [ `TAG of Tag.t | `CONTAINS of string | `NOTHING ] +let cpt_trans = ref 0 +let miss_trans = ref 0 +let cpt_eval = ref 0 +let miss_eval = ref 0 let gen_id = let id = ref (-1) in fun () -> incr id;!id +let h_union = Hashtbl.create 4097 + +let pt_cup s1 s2 = + (* special case, since this is a union we want hash(s1,s2) = hash(s2,s1) *) + let x = Ptset.hash s1 + and y = Ptset.hash s2 in + let h = if x < y then HASHINT2(x,y) else HASHINT2(y,x) in + try + Hashtbl.find h_union h + with + | Not_found -> let s = Ptset.union s1 s2 + in + Hashtbl.add h_union h s;s module State = struct @@ -18,13 +35,7 @@ let mk_state = State.mk type state = State.t -type predicate = [ `Left of (Tree.t -> bool) | `Right of (Tree.t -> bool) | - `True - ] -let eval_pred t = - function `True -> true - | `Left f | `Right f -> f t type formula_expr = | False | True @@ -35,19 +46,20 @@ and formula = { fid: int; fkey : int; pos : formula_expr; neg : formula; - st : (Ptset.t*Ptset.t)*(Ptset.t*Ptset.t); + st : (Ptset.t*Ptset.t*Ptset.t)*(Ptset.t*Ptset.t*Ptset.t); size: int; } external hash_const_variant : [> ] -> int = "%identity" -external int_bool : bool -> int = "%identity" +external vb : bool -> int = "%identity" let hash_node_form t = match t with | False -> 0 | True -> 1 - | And(f1,f2) -> (2+17*f1.fkey + 37*f2.fkey) land max_int - | Or(f1,f2) -> (3+101*f1.fkey + 253*f2.fkey) land max_int - | Atom(v,b,s) -> ((hash_const_variant v) + (3846*(int_bool b) +257) + (s lsl 13 - s)) land max_int + | And(f1,f2) -> (2+17*f1.fkey + 37*f2.fkey) (*land max_int *) + | Or(f1,f2) -> (3+101*f1.fkey + 253*f2.fkey) (*land max_int *) + | Atom(v,b,s) -> HASHINT3(hash_const_variant v,(3846*(vb b) +257),s) + module FormNode = @@ -70,12 +82,12 @@ module WH = Weak.Make(FormNode) let f_pool = WH.create 107 -let empty_pair = Ptset.empty,Ptset.empty -let empty_quad = empty_pair,empty_pair +let empty_triple = Ptset.empty,Ptset.empty,Ptset.empty +let empty_hex = empty_triple,empty_triple let true_,false_ = - let rec t = { fid = 1; pos = True; fkey=1; neg = f ; st = empty_quad; size =1; } - and f = { fid = 0; pos = False; fkey=0; neg = t; st = empty_quad; size = 1; } + let rec t = { fid = 1; pos = True; fkey=1; neg = f ; st = empty_hex; size =1; } + and f = { fid = 0; pos = False; fkey=0; neg = t; st = empty_hex; size = 1; } in WH.add f_pool f; WH.add f_pool t; @@ -107,21 +119,21 @@ let cons pos neg s1 s2 size1 size2 = let atom_ d p s = let si = Ptset.singleton s in let ss = match d with - | `Left -> (si,Ptset.empty),empty_pair - | `Right -> empty_pair,(si,Ptset.empty) - | `LLeft -> (Ptset.empty,si),empty_pair - | `RRight -> empty_pair,(Ptset.empty,si) + | `Left -> (si,Ptset.empty,si),empty_triple + | `Right -> empty_triple,(si,Ptset.empty,si) + | `LLeft -> (Ptset.empty,si,si),empty_triple + | `RRight -> empty_triple,(Ptset.empty,si,si) in fst (cons (Atom(d,p,s)) (Atom(d,not p,s)) ss ss 1 1) -let union_quad ((l1,ll1),(r1,rr1)) ((l2,ll2),(r2,rr2)) = - (Ptset.union l1 l2 ,Ptset.union ll1 ll2), - (Ptset.union r1 r2 ,Ptset.union rr1 rr2) +let union_hex ((l1,ll1,lll1),(r1,rr1,rrr1)) ((l2,ll2,lll2),(r2,rr2,rrr2)) = + (pt_cup l1 l2 ,pt_cup ll1 ll2,pt_cup lll1 lll2), + (pt_cup r1 r2 ,pt_cup rr1 rr2,pt_cup rrr1 rrr2) let merge_states f1 f2 = let sp = - union_quad f1.st f2.st + union_hex f1.st f2.st and sn = - union_quad f1.neg.st f2.neg.st + union_hex f1.neg.st f2.neg.st in sp,sn @@ -161,29 +173,38 @@ let and_ f1 f2 = let not_ f = f.neg +let k_hash (s,t) = HASHINT2(Ptset.hash s,Tag.hash t) -module HTagSetKey = +module HTagSetKey = struct type t = Ptset.t*Tag.t - let int_hash key = key lsl 31 lor (key lsl 8) let equal (s1,s2) (t1,t2) = (s2 == t2) && Ptset.equal s1 t1 - let hash (s,t) = int_hash (Ptset.hash s) lxor ( int_hash (Tag.hash t)) + let hash = k_hash end + module HTagSet = Hashtbl.Make(HTagSetKey) -type t = { +type skiplist = Nothing | All + | Zero of skiplist + | One of skiplist | Two of skiplist | Three of skiplist + | Four of skiplist | Five of skiplist | Six of skiplist + | Seven of skiplist | Eight of skiplist | Nine of skiplist + + +type formlist = Nil | Cons of state*formula*int*bool*formlist + +type 'a t = { id : int; mutable states : Ptset.t; init : Ptset.t; mutable final : Ptset.t; universal : Ptset.t; + starstate : Ptset.t option; (* Transitions of the Alternating automaton *) - phi : (state,(TagSet.t*(bool*formula*predicate)) list) Hashtbl.t; - delta : (state*Tag.t, (bool*formula*predicate)) Hashtbl.t; -(* delta : (state,(bool*formula*predicate) TagMap.t) Hashtbl.t; *) - sigma : (bool*formula*(predicate list*predicate list)*bool) HTagSet.t; - } - + phi : (state,(TagSet.t*(bool*formula*bool)) list) Hashtbl.t; + sigma : (int,('a t -> Tree.t -> Tree.t -> Ptset.t*'a)) Hashtbl.t; +} + module Pair (X : Set.OrderedType) (Y : Set.OrderedType) = struct type t = X.t*Y.t @@ -196,7 +217,7 @@ type t = { module PL = Set.Make (Pair (Ptset) (Ptset)) - let pr_st ppf l = Format.fprintf ppf "{"; + let pr_st ppf l = Format.fprintf ppf "{"; begin match l with | [] -> () @@ -226,51 +247,6 @@ type t = { | `LLeft -> "⇓₁" | `RRight -> "⇓₂") s - let dnf_hash = Hashtbl.create 17 - - let rec dnf_aux f = match f.pos with - | False -> PL.empty - | True -> PL.singleton (Ptset.empty,Ptset.empty) - | Atom((`Left|`LLeft),_,s) -> PL.singleton (Ptset.singleton s,Ptset.empty) - | Atom((`Right|`RRight),_,s) -> PL.singleton (Ptset.empty,Ptset.singleton s) - | Or(f1,f2) -> PL.union (dnf f1) (dnf f2) - | And(f1,f2) -> - let pl1 = dnf f1 - and pl2 = dnf f2 - in - PL.fold (fun (s1,s2) acc -> - PL.fold ( fun (s1', s2') acc' -> - (PL.add - ((Ptset.union s1 s1'), - (Ptset.union s2 s2')) acc') ) - pl2 acc ) - pl1 PL.empty - - - and dnf f = - try - Hashtbl.find dnf_hash f.fid - with - Not_found -> - let d = dnf_aux f in - Hashtbl.add dnf_hash f.fid d;d - - - let can_top_down f = - let nf = dnf f in - if (PL.cardinal nf > 3)then None - else match PL.elements nf with - | [(s1,s2); (t1,t2); (u1,u2)] when - Ptset.is_empty s1 && Ptset.is_empty s2 && Ptset.is_empty t1 && Ptset.is_empty u2 - -> Some(true,t2,u1) - | [(t1,t2); (u1,u2)] when Ptset.is_empty t1 && Ptset.is_empty u2 - -> Some(false,t2,u1) - | _ -> None - - - let equal_form f1 f2 = - (f1.fid == f2.fid) || (FormNode.equal f1 f2) || (PL.equal (dnf f1) (dnf f2)) - let dump ppf a = Format.fprintf ppf "Automaton (%i) :\n" a.id; Format.fprintf ppf "States : "; pr_st ppf (Ptset.elements a.states); @@ -296,28 +272,38 @@ type t = { Format.fprintf ppf "\n")l; Format.fprintf ppf "NFA transitions :\n------------------------------\n"; - HTagSet.iter (fun (qs,t) (b,f,_,_) -> - pr_st ppf (Ptset.elements qs); - Format.fprintf ppf ",%s %s " (Tag.to_string t) (if b then "=>" else "->"); - pr_frm ppf f; - Format.fprintf ppf "(fid=%i) left=" f.fid; - let (l,ll),(r,rr) = f.st in - pr_st ppf (Ptset.elements l); - Format.fprintf ppf ", "; - pr_st ppf (Ptset.elements ll); +(* HTagSet.iter (fun (qs,t) (disp,b,_,flist,_,_) -> + let (ls,lls,_),(rs,rrs,_) = + List.fold_left (fun ((a1,b1,c1),(a2,b2,c2)) (_,f) -> + let (x1,y1,z1),(x2,y2,z2) = f.st in + ((Ptset.union x1 a1),(Ptset.union y1 b1),(Ptset.union c1 z1)), + ((Ptset.union x2 a2),(Ptset.union y2 b2),(Ptset.union c2 z2))) + ((Ptset.empty,Ptset.empty,Ptset.empty), + (Ptset.empty,Ptset.empty,Ptset.empty)) + flist + in + pr_st ppf (Ptset.elements qs); + Format.fprintf ppf ",%s %s " (Tag.to_string t) (if b then "=>" else "->"); + List.iter (fun (q,f) -> + Format.fprintf ppf "\n%i," q; + pr_frm ppf f) flist; + Format.fprintf ppf "\nleft="; + pr_st ppf (Ptset.elements ls); + Format.fprintf ppf " , "; + pr_st ppf (Ptset.elements lls); Format.fprintf ppf ", right="; - pr_st ppf (Ptset.elements r); + pr_st ppf (Ptset.elements rs); Format.fprintf ppf ", "; - pr_st ppf (Ptset.elements rr); - Format.fprintf ppf "\n"; - ) a.sigma; - Format.fprintf ppf "=======================================\n" + pr_st ppf (Ptset.elements rrs); + Format.fprintf ppf ", first=%s, next=%s\n\n" disp.flabel disp.nlabel; + ) a.sigma; *) + Format.fprintf ppf "=======================================\n%!" module Transitions = struct - type t = state*TagSet.t*bool*formula*predicate + type t = state*TagSet.t*bool*formula*bool let ( ?< ) x = x - let ( >< ) state (l,b) = state,(l,b,`True) - let ( ><@ ) state (l,b,p) = state,(l,b,p) + let ( >< ) state (l,b) = state,(l,b,false) + let ( ><@ ) state (l,b) = state,(l,b,true) let ( >=> ) (state,(label,mark,pred)) form = (state,label,mark,form,pred) let ( +| ) f1 f2 = or_ f1 f2 let ( *& ) f1 f2 = and_ f1 f2 @@ -328,93 +314,41 @@ type t = { type transition = Transitions.t let equal_trans (q1,t1,m1,f1,_) (q2,t2,m2,f2,_) = - (q1 == q2) && (TagSet.equal t1 t2) && (m1 == m2) && (equal_form f1 f2) - - module TS = - struct - type node = Nil | Cons of Tree.t * node | Concat of node*node - and t = { node : node; size : int } - let node n s = { node=n; size = s } - - let empty = node Nil 0 - - let cons e t = node (Cons(e,t.node)) (t.size+1) - let concat t1 t2 = node (Concat (t1.node,t2.node)) (t1.size+t2.size) - let append = cons -(* let append e t = node (Concat(t.node,Cons(e,Nil))) (t.size+1) *) + (q1 == q2) && (TagSet.equal t1 t2) && (m1 == m2) (*&& (equal_form f1 f2) *) - let to_list_rev t = - let rec aux acc l rest = - match l with - | Nil -> begin - match rest with - | Nil -> acc - | Cons(e,t) -> aux (e::acc) t Nil - | Concat(t1,t2) -> aux acc t1 t2 - end - | Cons(e,r) -> aux (e::acc) r rest - | Concat(t1,t2) -> aux acc t1 (Concat(t2,rest)) - in - aux [] t.node Nil - - let length = function { size = s } -> s - - let iter f { node = n } = - let rec loop = function - | Nil -> () - | Cons(e,n) -> let _ = f e in loop n - | Concat(n1,n2) -> let _ = loop n1 in loop n2 - in loop n - - let rev_iter f { node = n } = - let rec loop = function - | Nil -> () - | Cons(e,n) -> let _ = loop n in f e - | Concat(n1,n2) -> let _ = loop n2 in loop n1 - in loop n - - - let find f { node = n } = - let rec loop = function - | Nil -> raise Not_found - | Cons(e,n) -> if f e then e else loop n - | Concat(n1,n2) -> try - loop n1 - with - | Not_found -> loop n2 - in - loop n - end -(* - module BottomUpJumpNew = struct - -*) - module HFEval = Hashtbl.Make( - struct - type t = int*Ptset.t*Ptset.t - let equal (a,b,c) (d,e,f) = - a==d && (Ptset.equal b e) && (Ptset.equal c f) - let hash (a,b,c) = - a+17*(Ptset.hash b) + 31*(Ptset.hash c) - end) - - let hfeval = HFEval.create 4097 - + module HFEval = Hashtbl.Make( + struct + type t = int*Ptset.t*Ptset.t + let equal (a,b,c) (d,e,f) = + a==d && (Ptset.equal b e) && (Ptset.equal c f) + let hash (a,b,c) = + HASHINT3(a,Ptset.hash b,Ptset.hash c) + end) + + + + let hfeval = HFEval.create 4097 let eval_form_bool f s1 s2 = let rec eval f = match f.pos with - | Atom((`Left|`LLeft),b,q) -> if b == (Ptset.mem q s1) then (true,true,false) else false,false,false - | Atom((`Right|`RRight),b,q) -> if b == (Ptset.mem q s2) then (true,false,true) else false,false,false - (* test some inlining *) + (* test some inlining *) | True -> true,true,true | False -> false,false,false | _ -> try - HFEval.find hfeval (f.fid,s1,s2) + HFEval.find hfeval (f.fid,s1,s2) with - | Not_found -> let r = + | Not_found -> let r = match f.pos with + | Atom((`Left|`LLeft),b,q) -> + if b == (Ptset.mem q s1) + then (true,true,false) + else false,false,false + | Atom(_,b,q) -> + if b == (Ptset.mem q s2) + then (true,false,true) + else false,false,false | Or(f1,f2) -> let b1,rl1,rr1 = eval f1 in @@ -440,344 +374,649 @@ type t = { in eval f - let fstate_pool = Hashtbl.create 11 - - let merge_pred a b = match a,b with - | Some(f1), Some(f2) -> Some(fun x -> f1 x || f2 x) - | None,None -> None - | None,Some(_) -> b - | Some(_),None -> a - - let acc_pred p l1 l2 = match p with - | `Left _ -> p::l1,l2 - | `Right _ -> l1,p::l2 - | _ -> l1,l2 - - - let merge_trans t a tag q acc = - List.fold_left (fun (accf,accm,acchtrue) (ts,(m,f,pred)) -> - if TagSet.mem tag ts - then - let tmpf,hastrue = - if is_true f then - let newfinal = - try Hashtbl.find fstate_pool f.fid with - | Not_found -> let s = mk_state() in - a.states <- Ptset.add s a.states; - a.final <- Ptset.add s a.final; - Hashtbl.add fstate_pool f.fid s;s - in - (atom_ `Left true newfinal),true - else f,false in - (or_ tmpf accf,accm||m,acchtrue||hastrue) - else (accf,accm,acchtrue) - ) acc (try Hashtbl.find a.phi q with Not_found -> []) - - let get_trans t a tag r = - try - let mark,f,predl,has_true = - HTagSet.find a.sigma (r,tag) - in f.st,f,mark,has_true,r - with - Not_found -> - let f,mark,has_true,accq = - Ptset.fold (fun q (accf,accm,acchtrue,accq) -> - let naccf,naccm,nacctrue = - merge_trans t a tag q (accf,accm,acchtrue ) - in - if is_false naccf then (naccf,naccm,nacctrue,accq) - else (naccf,naccm,nacctrue,Ptset.add q accq) - ) - r (false_,false,false,Ptset.empty) - in - HTagSet.add a.sigma (accq,tag) (mark,f,([],[]),has_true); - f.st,f,mark,has_true,accq - - let h_union = Hashtbl.create 4097 - - let pt_cup s1 s2 = - let h = (Ptset.hash s1,Ptset.hash s2) in - try - Hashtbl.find h_union h - with - | Not_found -> let s = Ptset.union s1 s2 + let form_list_fold_left f acc fl = + let rec loop acc fl = + match fl with + | Nil -> acc + | Cons(s,frm,h,m,fll) -> loop (f acc s frm h m) fll + in + loop acc fl + + let h_formlist = Hashtbl.create 4096 + let rec eval_formlist ?(memo=true) s1 s2 fl = + match fl with + | Nil -> Ptset.empty,false,false,false,false + | Cons(q,f,h,mark,fll) -> + let k = (h,Ptset.hash s1,Ptset.hash s2,mark) in - Hashtbl.add h_union h s;s - - - + + try + if memo then Hashtbl.find h_formlist k + else (raise Not_found) + with + Not_found -> + let s,b',b1',b2',amark = eval_formlist (~memo:memo) s1 s2 fll in + let b,b1,b2 = eval_form_bool f s1 s2 in + let r = if b then (Ptset.add q s, b, b1'||b1,b2'||b2,mark||amark) + else s,b',b1',b2',amark + in(* + Format.fprintf Format.err_formatter "\nEvaluating formula (%i) %i %s" h q (if mark then "=>" else "->"); + pr_frm (Format.err_formatter) f; + Format.fprintf Format.err_formatter " in context "; + pr_st Format.err_formatter (Ptset.elements s1); + Format.fprintf Format.err_formatter ", "; + pr_st Format.err_formatter (Ptset.elements s2); + Format.fprintf Format.err_formatter " result is %b\n%!" b; *) + (Hashtbl.add h_formlist k r;r) + + + let tags_of_state a q = Hashtbl.fold (fun p l acc -> if p == q then List.fold_left - (fun acc (ts,_) -> - pt_cup (TagSet.positive ts) acc) acc l - else acc) a.phi Ptset.empty + (fun acc (ts,(_,_,aux)) -> + if aux then acc else + TagSet.cup ts acc) acc l + else acc) a.phi TagSet.empty + - let h_tags_states = Hashtbl.create 4096 + + let tags a qs = + let ts = Ptset.fold (fun q acc -> TagSet.cup acc (tags_of_state a q)) qs TagSet.empty + in + if TagSet.is_finite ts + then `Positive(TagSet.positive ts) + else `Negative(TagSet.negative ts) + + let inter_text a b = + match b with + | `Positive s -> let r = Ptset.inter a s in (r,Ptset.mem Tag.pcdata r, true) + | `Negative s -> let r = Ptset.diff a s in (r, Ptset.mem Tag.pcdata r, false) + + let mk_nil_ctx x _ = Tree.mk_nil x + let next_sibling_ctx x _ = Tree.next_sibling x + let r_ignore _ x = x + let set_get_tag r t = r := (fun _ -> t) + + module type ResultSet = + sig + type t + val empty : t + val cons : Tree.t -> t -> t + val concat : t -> t -> t + val iter : (Tree.t -> unit) -> t -> unit + val fold : (Tree.t -> 'a -> 'a) -> t -> 'a -> 'a + val map : (Tree.t -> Tree.t) -> t -> t + val length : t -> int + end + + module Integer : ResultSet = + struct + type t = int + let empty = 0 + let cons _ x = x+1 + let concat x y = x + y + let iter _ _ = failwith "iter not implemented" + let fold _ _ _ = failwith "fold not implemented" + let map _ _ = failwith "map not implemented" + let length x = x + end + + module IdSet : ResultSet = + struct + type node = Nil + | Cons of Tree.t * node + | Concat of node*node + + and t = { node : node; + length : int } + + let empty = { node = Nil; length = 0 } + + let cons e t = { node = Cons(e,t.node); length = t.length+1 } + let concat t1 t2 = { node = Concat(t1.node,t2.node); length = t1.length+t2.length } + let append e t = { node = Concat(t.node,Cons(e,Nil)); length = t.length+1 } + + let fold f l acc = + let rec loop acc t = match t with + | Nil -> acc + | Cons (e,t) -> loop (f e acc) t + | Concat (t1,t2) -> loop (loop acc t1) t2 + in + loop acc l.node + + let length l = l.length + + + let iter f l = + let rec loop = function + | Nil -> () + | Cons (e,t) -> f e; loop t + | Concat(t1,t2) -> loop t1;loop t2 + in loop l.node + + let map f l = + let rec loop = function + | Nil -> Nil + | Cons(e,t) -> Cons(f e, loop t) + | Concat(t1,t2) -> Concat(loop t1,loop t2) + in + { l with node = loop l.node } + + end + module Run (RS : ResultSet) = + struct + let fmt = Format.err_formatter + let pr x = Format.fprintf fmt x + module Formlist = + struct + type t = formlist + let nil : t = Nil + let cons q f i m l = Cons(q,f,i,m,l) + let hash = function Nil -> 0 | Cons(_,_,i,_,_) -> max_int land i + let pr fmt l = + let rec loop = function + | Nil -> () + | Cons(q,f,_,m,l) -> + Format.fprintf fmt "%i %s" q (if m then "=>" else "->"); + pr_frm fmt f; + Format.fprintf fmt "\n%!"; + loop l + in + loop l + end + + type ptset_list = Nil | Cons of Ptset.t*int*ptset_list + let hpl l = match l with + | Nil -> 0 + | Cons (_,i,_) -> i - let tags a qs = - try - Hashtbl.find h_tags_states (Ptset.hash qs) - with - | Not_found -> - let l = Ptset.fold (fun q acc -> pt_cup acc (tags_of_state a q)) qs Ptset.empty - in - Hashtbl.add h_tags_states (Ptset.hash qs) l;l - - let time cpt acc f x = - let t1 = Unix.gettimeofday () in - let r = f x in - let t2 = Unix.gettimeofday () in - let t = (1000. *.(t2 -. t1)) in - acc:=!acc+.t; - incr cpt; - r + let cons s l = Cons (s,(Ptset.hash s) + 65599 * (hpl l), l) + let rec empty_size n = + if n == 0 then Nil + else cons Ptset.empty (empty_size (n-1)) - let h_time = Hashtbl.create 4096 - let calls = ref 0 + let fold_pl f l acc = + let rec loop l acc = match l with + Nil -> acc + | Cons(s,h,pl) -> loop pl (f s h acc) + in + loop l acc + let map_pl f l = + let rec loop = + function Nil -> Nil + | Cons(s,h,ll) -> cons (f s) (loop ll) + in loop l + let iter_pl f l = + let rec loop = + function Nil -> () + | Cons(s,h,ll) -> (f s);(loop ll) + in loop l + + let rev_pl l = + let rec loop acc l = match l with + | Nil -> acc + | Cons(s,_,ll) -> loop (cons s acc) ll + in + loop Nil l - let rtime s f x = - - let cpt,atime = + let rev_map_pl f l = + let rec loop acc l = + match l with + | Nil -> acc + | Cons(s,_,ll) -> loop (cons (f s) acc) ll + in + loop Nil l + + let td_trans = Hashtbl.create 4096 + + + let choose_jump tagset qtags1 qtagsn a f_nil f_text f_t1 f_s1 f_tn f_sn f_notext = + let tags1,hastext1,fin1 = inter_text tagset (tags a qtags1) in + let tagsn,hastextn,finn = inter_text tagset (tags a qtagsn) in +(* Format.fprintf Format.err_formatter "Tags below states "; + pr_st Format.err_formatter (Ptset.elements qtags1); + Format.fprintf Format.err_formatter " are { "; + Ptset.iter (fun t -> Format.fprintf Format.err_formatter "%s " (Tag.to_string t)) tags1; + Format.fprintf Format.err_formatter "}, %b,%b\n%!" hastext1 fin1; + + Format.fprintf Format.err_formatter "Tags below states "; + pr_st Format.err_formatter (Ptset.elements qtagsn); + Format.fprintf Format.err_formatter " are { "; + Ptset.iter (fun t -> Format.fprintf Format.err_formatter "%s " (Tag.to_string t)) tagsn; + Format.fprintf Format.err_formatter "}, %b,%b\n%!" hastextn finn; +*) + if (hastext1||hastextn) then f_text (* jumping to text nodes doesn't work really well *) + else if (Ptset.is_empty tags1) && (Ptset.is_empty tagsn) then f_nil + else if (Ptset.is_empty tagsn) then + if (Ptset.is_singleton tags1) then f_t1 (Ptset.choose tags1) (* TaggedChild/Sibling *) + else f_s1 tags1 (* SelectChild/Sibling *) + else if (Ptset.is_empty tags1) then + if (Ptset.is_singleton tagsn) then f_tn (Ptset.choose tagsn) (* TaggedDesc/Following *) + else f_sn tagsn (* SelectDesc/Following *) + else f_notext + + let choose_jump_down a b c d = + choose_jump a b c d + (Tree.mk_nil) + (Tree.text_below) + (*fun x -> let i,j = Tree.doc_ids x in + let res = Tree.text_below x in + Printf.printf "Calling text_below %s (tag=%s), docids= (%i,%i), res=%s\n" + (Tree.dump_node x) (Tag.to_string (Tree.tag x)) i j (Tree.dump_node res); + res*) + (fun _ -> Tree.node_child ) (* !! no tagged_child in Tree.ml *) + (fun _ -> Tree.node_child ) (* !! no select_child in Tree.ml *) + (Tree.tagged_desc) + (fun _ -> Tree.node_child ) (* !! no select_desc *) + (Tree.node_child) + + let choose_jump_next a b c d = + choose_jump a b c d + (fun t _ -> Tree.mk_nil t) + (Tree.text_next) + (*fun x y -> let i,j = Tree.doc_ids x in + let res = Tree.text_next x y in + Printf.printf "Calling text_next %s (tag=%s) ctx=%s, docids= (%i,%i), res=%s\n" + (Tree.dump_node x) (Tag.to_string (Tree.tag x)) (Tree.dump_node y) i j (Tree.dump_node res); + res*) + + (fun _ -> Tree.node_sibling_ctx) (* !! no tagged_sibling in Tree.ml *) + (fun _ -> Tree.node_sibling_ctx) (* !! no select_child in Tree.ml *) + (Tree.tagged_foll_below) + (fun _ -> Tree.node_sibling_ctx) (* !! no select_foll *) + (Tree.node_sibling_ctx) + + + let get_trans slist tag a t = try - Hashtbl.find h_time s + Hashtbl.find td_trans (tag,hpl slist) with - | _ -> (ref 0, ref 0.) - in - let r = time cpt atime f x - in - Hashtbl.replace h_time s (cpt,atime); - r - - let rec accepting_among_time a t r ctx = - incr calls; - let orig = r in - let rest = Ptset.inter r a.final in - let r = Ptset.diff r rest in - if Ptset.is_empty r then rest,TS.empty else - if Tree.is_node t - then - let among,result,form = - let ((ls,lls),(rs,rrs)),formula,mark,has_true,r' = - let tag = rtime "Tree.tag" Tree.tag t in - rtime "get_trans" (get_trans t a tag) r - in - let tl = rtime "tags" (tags a) ls - and tr = rtime "tags" (tags a) rs - and tll = rtime "tags" (tags a) lls - and trr = rtime "tags" (tags a) rrs - in - let first = - if Ptset.mem Tag.pcdata (pt_cup tl tll) - then - rtime "Tree.text_below" (Tree.text_below) t - else - let etl = Ptset.is_empty tl - and etll = Ptset.is_empty tll - in - if etl && etll - then Tree.mk_nil t - else - if etl then rtime "Tree.tagged_desc_only" (Tree.tagged_desc_only t) tll - else if etll then rtime "Tree.first_child" (Tree.first_child) t - else (* add child only *) - rtime "Tree.tagged_below" (Tree.tagged_below t tl) tll - and next = - if Ptset.mem Tag.pcdata (pt_cup tr trr) - then - rtime "Tree.text_next" (Tree.text_next t) ctx - else - let etr = Ptset.is_empty tr - and etrr = Ptset.is_empty trr - in - if etr && etrr - then Tree.mk_nil t - else - if etr then rtime "Tree.tagged_foll_only" (Tree.tagged_foll_only t trr) ctx - else if etrr then rtime "Tree.next_sibling" (Tree.next_sibling) t - else (* add ns only *) - rtime "Tree.tagged_next" (Tree.tagged_next t tr trr) ctx - - in - let s1,res1 = accepting_among_time a first (pt_cup ls lls) t - and s2,res2 = accepting_among_time a next (pt_cup rs rrs) ctx + | Not_found -> + let fl_list,llist,rlist,ca,da,sa,fa = + fold_pl + (fun set _ (fll_acc,lllacc,rllacc,ca,da,sa,fa) -> (* For each set *) + let fl,ll,rr,ca,da,sa,fa = + Ptset.fold + (fun q acc -> + fst ( + List.fold_left + (fun (((fl_acc,ll_acc,rl_acc,c_acc,d_acc,s_acc,f_acc),h_acc) as acc) + (ts,(m,f,_)) -> + if (TagSet.mem tag ts) + then + let (child,desc,below),(sibl,foll,after) = f.st in + let h_acc = HASHINT3(h_acc,f.fid,HASHINT2(q,vb m)) in + ((Formlist.cons q f h_acc m fl_acc, + Ptset.union ll_acc below, + Ptset.union rl_acc after, + Ptset.union child c_acc, + Ptset.union desc d_acc, + Ptset.union sibl s_acc, + Ptset.union foll f_acc), + h_acc) + else acc ) (acc,0) ( + try Hashtbl.find a.phi q + with + Not_found -> Printf.eprintf "Looking for state %i, doesn't exist!!!\n%!" + q;[] + )) + + ) set (Formlist.nil,Ptset.empty,Ptset.empty,ca,da,sa,fa) + in fl::fll_acc, cons ll lllacc, cons rr rllacc,ca,da,sa,fa) + slist ([],Nil,Nil,Ptset.empty,Ptset.empty,Ptset.empty,Ptset.empty) in - let rb,rb1,rb2 = rtime "eval_form_bool" (eval_form_bool formula s1) s2 in - if rb - then - let res1 = if rb1 then res1 else TS.empty - and res2 = if rb2 then res2 else TS.empty - in r', rtime "TS.concat" (TS.concat res2) (if mark then rtime "TS.append" (TS.append t) res1 else res1),formula - else Ptset.empty,TS.empty,formula - - in + (* Logic to chose the first and next function *) + let tags_below,tags_after = Tree.tags t tag in + let first = choose_jump_down tags_below ca da a + and next = choose_jump_next tags_after sa fa a in + let v = (fl_list,llist,rlist,first,next) in + Hashtbl.add td_trans (tag, hpl slist) v; v + + let merge rb rb1 rb2 mark t res1 res2 = + if rb + then + let res1 = if rb1 then res1 else RS.empty + and res2 = if rb2 then res2 else RS.empty + in + if mark then RS.cons t (RS.concat res1 res2) + else RS.concat res1 res2 + else RS.empty - among,result - - else orig,TS.empty - - - let run_time a t = - let st,res = accepting_among_time a t a.init t in - let _ = Printf.eprintf "\n Timings\n"; - let total_time = Hashtbl.fold (fun fname ({ contents=cpt }, {contents=atime}) (total_time) -> - Printf.eprintf "%s\t %i calls, %f ms accumulated time, %f ms mean time\n" - fname cpt atime (atime /. (float_of_int cpt)); - total_time +. atime ) h_time 0. + let top_down ?(noright=false) a t slist ctx slot_size = + let pempty = empty_size slot_size in + let eval_fold2_slist fll sl1 sl2 res1 res2 t = + let res = Array.copy res1 in + let rec fold l1 l2 fll i aq = match l1,l2,fll with + | Cons(s1,_,ll1), Cons(s2, _ ,ll2),fl::fll -> + let r',rb,rb1,rb2,mark = eval_formlist s1 s2 fl in +(* let _ = pr "Evaluation context : "; pr_st fmt (Ptset.elements s1); + pr_st fmt (Ptset.elements s2); + pr "Formlist (%i) : " (Formlist.hash fl); + Formlist.pr fmt fl; + pr "Results : "; pr_st fmt (Ptset.elements r'); + pr ", %b %b %b %b\n%!" rb rb1 rb2 mark + in *) + let _ = res.(i) <- merge rb rb1 rb2 mark t res1.(i) res2.(i) + in + fold ll1 ll2 fll (i+1) (cons r' aq) + | Nil, Nil,[] -> aq,res + | _ -> assert false + in + fold sl1 sl2 fll 0 Nil in - Printf.eprintf "total calls %i, total monitored time %f ms\n%!" !calls total_time - in - if Ptset.is_empty (st) then TS.empty else res - - - - let rec accepting_among a t r ctx = - let orig = r in - let rest = Ptset.inter r a.final in - let r = Ptset.diff r rest in - if Ptset.is_empty r then rest,TS.empty else - if Tree.is_node t - then - let among,result,form = - let ((ls,lls),(rs,rrs)),formula,mark,has_true,r' = - let tag = Tree.tag t in - get_trans t a tag r - in - let tl = tags a ls - and tr = tags a rs - and tll = tags a lls - and trr = tags a rrs - in - let first = - if Ptset.mem Tag.pcdata (pt_cup tl tll) - then - Tree.text_below t - else - let etl = Ptset.is_empty tl - and etll = Ptset.is_empty tll - in - if etl && etll - then Tree.mk_nil t - else - if etl then Tree.tagged_desc_only t tll - else if etll then Tree.first_child t - else (* add child only *) - Tree.tagged_below t tl tll - and next = - if Ptset.mem Tag.pcdata (pt_cup tr trr) - then - Tree.text_next t ctx - else - let etr = Ptset.is_empty tr - and etrr = Ptset.is_empty trr - in - if etr && etrr - then Tree.mk_nil t - else - if etr then Tree.tagged_foll_only t trr ctx - else if etrr then Tree.next_sibling t - else (* add ns only *) - Tree.tagged_next t tr trr ctx - - in - let s1,res1 = accepting_among a first (pt_cup ls lls) t - and s2,res2 = accepting_among a next (pt_cup rs rrs) ctx - in - let rb,rb1,rb2 = eval_form_bool formula s1 s2 in - if rb - then - let res1 = if rb1 then res1 else TS.empty - and res2 = if rb2 then res2 else TS.empty - in r', TS.concat res2 (if mark then TS.append t res1 else res1),formula - else Ptset.empty,TS.empty,formula - - in - among,result - - else orig,TS.empty + let null_result() = (pempty,Array.make slot_size RS.empty) in + let rec loop t slist ctx = + let (a,b) = + if Tree.is_nil t then null_result() + else + let tag = Tree.tag t in + let fl_list,llist,rlist,first,next = get_trans slist tag a t in +(* let _ = pr "For tag %s,node %s, returning formulae list: \n%!" + (Tag.to_string tag) (Tree.dump_node t); + List.iter (fun f -> Formlist.pr fmt f;pr "\n%!") fl_list + in*) + let sl1,res1 = loop (first t) llist t in + let sl2,res2 = loop (next t ctx) rlist ctx in + eval_fold2_slist fl_list sl1 sl2 res1 res2 t + in +(* let _ = pr "Inside topdown call: tree was %s, tag = %s" (Tree.dump_node t) (if Tree.is_nil t then "###" + else Tag.to_string (Tree.tag t)); + iter_pl (fun s -> (pr_st fmt (Ptset.elements s))) a; + Array.iter (fun i -> pr "%i" (RS.length i)) b; + pr "\n%!"; in*) (a,b) + + in + let loop_no_right t slist ctx = + if Tree.is_nil t then null_result() + else + let tag = Tree.tag t in + let fl_list,llist,rlist,first,next = get_trans slist tag a t in + let sl1,res1 = loop (first t) llist t in + let sl2,res2 = null_result() in + eval_fold2_slist fl_list sl1 sl2 res1 res2 t + in + (if noright then loop_no_right else loop) t slist ctx + + let run_top_down a t = + let init = cons a.init Nil in + let _,res = top_down a t init t 1 + in res.(0) + ;; + + module Configuration = + struct + module Ptss = Set.Make(Ptset) + module IMap = Map.Make(Ptset) + type t = { hash : int; + sets : Ptss.t; + results : RS.t IMap.t } + let empty = { hash = 0; + sets = Ptss.empty; + results = IMap.empty; + } + let is_empty c = Ptss.is_empty c.sets + let add c s r = + if Ptss.mem s c.sets then + { c with results = IMap.add s (RS.concat r (IMap.find s c.results)) c.results} + else + { hash = HASHINT2(c.hash,Ptset.hash s); + sets = Ptss.add s c.sets; + results = IMap.add s r c.results + } - - let run a t = - let st,res = accepting_among a t a.init t in - if Ptset.is_empty (st) then TS.empty else res - - let rec accepting_among_count a t r ctx = - let orig = r in - let rest = Ptset.inter r a.final in - let r = Ptset.diff r rest in - if Ptset.is_empty r then rest,0 else - if Tree.is_node t - then - let ((ls,lls),(rs,rrs)),formula,mark,has_true,r' = - let tag = Tree.tag t in - get_trans t a tag r - in - let tl = tags a ls - and tr = tags a rs - and tll = tags a lls - and trr = tags a rrs - in - let first = - if Ptset.mem Tag.pcdata (pt_cup tl tll) - then - Tree.text_below t - else - let etl = Ptset.is_empty tl - and etll = Ptset.is_empty tll - in - if etl && etll - then Tree.mk_nil t - else - if etl then Tree.tagged_desc_only t tll - else if etll then Tree.first_child t - else (* add child only *) - Tree.tagged_below t tl tll - and next = - if Ptset.mem Tag.pcdata (pt_cup tr trr) - then - Tree.text_next t ctx - else - let etr = Ptset.is_empty tr - and etrr = Ptset.is_empty trr - in - if etr && etrr - then Tree.mk_nil t - else - if etr then Tree.tagged_foll_only t trr ctx - else if etrr then Tree.next_sibling t - else (* add ns only *) - Tree.tagged_next t tr trr ctx - + let pr fmt c = Format.fprintf fmt "{"; + Ptss.iter (fun s -> pr_st fmt (Ptset.elements s); + Format.fprintf fmt " ") c.sets; + Format.fprintf fmt "}\n%!"; + IMap.iter (fun k d -> + pr_st fmt (Ptset.elements k); + Format.fprintf fmt "-> %i\n" (RS.length d)) c.results; + Format.fprintf fmt "\n%!" + + let merge c1 c2 = + let acc1 = IMap.fold (fun s r acc -> + IMap.add s + (try + RS.concat r (IMap.find s acc) + with + | Not_found -> r) acc) c1.results IMap.empty in - let s1,res1 = accepting_among_count a first (pt_cup ls lls) t - and s2,res2 = accepting_among_count a next (pt_cup rs rrs) ctx + let imap = + IMap.fold (fun s r acc -> + IMap.add s + (try + RS.concat r (IMap.find s acc) + with + | Not_found -> r) acc) c2.results acc1 in - let rb,rb1,rb2 = eval_form_bool formula s1 s2 in - if rb - then - let res1 = if rb1 then res1 else 0 - and res2 = if rb2 then res2 else 0 - in r', res2 + (if mark then 1 + res1 else res1) - else Ptset.empty,0 - - - - else orig,0 + let h,s = + Ptss.fold + (fun s (ah,ass) -> (HASHINT2(ah,Ptset.hash s), + Ptss.add s ass)) + (Ptss.union c1.sets c2.sets) (0,Ptss.empty) + in + { hash = h; + sets =s; + results = imap } + + end + + let h_fold = Hashtbl.create 511 + + let fold_f_conf t slist fl_list conf dir= + let rec loop sl fl acc = + match sl,fl with + |Nil,[] -> acc + | Cons(s,hs,sll), formlist::fll -> + let r',rb,rb1,rb2,mark = + try + Hashtbl.find h_fold (hs,Formlist.hash formlist,dir) + with + Not_found -> let res = + if dir then eval_formlist ~memo:false s Ptset.empty formlist + else eval_formlist ~memo:false Ptset.empty s formlist + in (Hashtbl.add h_fold (hs,Formlist.hash formlist,dir) res;res) + in(* + let _ = pr "Evaluating on set (%s) with tree %s=%s" + (if dir then "left" else "right") + (Tag.to_string (Tree.tag t)) + (Tree.dump_node t) ; + pr_st fmt (Ptset.elements s); + pr ", formualae (with hash %i): \n" (Formlist.hash formlist); + Formlist.pr fmt formlist; + pr "result is "; + pr_st fmt (Ptset.elements r'); + pr " %b %b %b %b \n%!" rb rb1 rb2 mark ; + in *) + if rb && ((dir&&rb1)|| ((not dir) && rb2)) + then + let acc = + let old_r = + try Configuration.IMap.find s conf.Configuration.results + with Not_found -> RS.empty + in + Configuration.add acc r' (if mark then RS.cons t old_r else old_r) + in + loop sll fll acc + else loop sll fll acc + | _ -> assert false + in + loop slist fl_list Configuration.empty - - let run_count a t = - let st,res = accepting_among_count a t a.init t in - if Ptset.is_empty (st) then 0 else res + let h_trans = Hashtbl.create 4096 + let get_up_trans slist ptag a tree = + let key = (HASHINT2(hpl slist,Tag.hash ptag)) in + try + Hashtbl.find h_trans key + with + | Not_found -> + let f_list,_ = + Hashtbl.fold (fun q l acc -> + List.fold_left (fun (fl_acc,h_acc) (ts,(m,f,_)) -> + if TagSet.mem ptag ts + then + let h_acc = HASHINT3(h_acc,f.fid,HASHINT2(q,vb m)) in + (Formlist.cons q f h_acc m fl_acc, + h_acc) + else (fl_acc,h_acc)) + acc l) + a.phi (Formlist.nil,0) + in + let res = fold_pl (fun _ _ acc -> f_list::acc) slist [] + in + (Hashtbl.add h_trans key res;res) + + + let h_tdconf = Hashtbl.create 511 + let rec bottom_up a tree conf next jump_fun root dotd init accu = + if (not dotd) && (Configuration.is_empty conf ) then +(* let _ = pr "Returning early from %s, with accu %i, next is %s\n%!" + (Tree.dump_node tree) (Obj.magic accu) (Tree.dump_node next) + in *) + accu,conf,next + else +(* let _ = + pr "Going bottom up for tree with tag %s configuration is" + (if Tree.is_nil tree then "###" else Tag.to_string (Tree.tag tree)); + Configuration.pr fmt conf + in *) + let below_right = Tree.is_below_right tree next in + (* let _ = Format.fprintf Format.err_formatter "below_right %s %s = %b\n%!" + (Tree.dump_node tree) (Tree.dump_node next) below_right + in *) + let accu,rightconf,next_of_next = + if below_right then (* jump to the next *) +(* let _ = pr "Jumping to %s tag %s\n%!" (Tree.dump_node next) (Tag.to_string (Tree.tag next)) in *) + bottom_up a next conf (jump_fun next) jump_fun (Tree.next_sibling tree) true init accu + else accu,Configuration.empty,next + in +(* let _ = if below_right then pr "Returning from jump to next = %s\n" (Tree.dump_node next)in *) + let sub = + if dotd then + if below_right then (* only recurse on the left subtree *) +(* let _ = pr "Topdown on left subtree\n%!" in *) + prepare_topdown a tree true + else +(* let _ = pr "Topdown on whole tree\n%!" in *) + prepare_topdown a tree false + else conf + in + let conf,next = + (Configuration.merge rightconf sub, next_of_next) + in + if Tree.equal tree root then +(* let _ = pr "Stopping at root, configuration after topdown is:" ; + Configuration.pr fmt conf; + pr "\n%!" + in *) accu,conf,next + else + let parent = Tree.binary_parent tree in + let ptag = Tree.tag parent in + let dir = Tree.is_left tree in + let slist = Configuration.Ptss.fold (fun e a -> cons e a) conf.Configuration.sets Nil in + let fl_list = get_up_trans slist ptag a parent in + let slist = rev_pl (slist) in +(* let _ = pr "Current conf is : %s " (Tree.dump_node tree); + Configuration.pr fmt conf; + pr "\n" + in *) + let newconf = fold_f_conf parent slist fl_list conf dir in +(* let _ = pr "New conf before pruning is (dir=%b):" dir; + Configuration.pr fmt newconf ; + pr "accu is %i\n" (RS.length accu); + in *) + let accu,newconf = Configuration.IMap.fold (fun s res (ar,nc) -> + if Ptset.intersect s init then + ( RS.concat res ar ,nc) + else (ar,Configuration.add nc s res)) + (newconf.Configuration.results) (accu,Configuration.empty) + in +(* let _ = pr "New conf after pruning is (dir=%b):" dir; + Configuration.pr fmt newconf ; + pr "accu is %i\n" (RS.length accu); + in *) + bottom_up a parent newconf next jump_fun root false init accu + + and prepare_topdown a t noright = + let tag = Tree.tag t in +(* pr "Going top down on tree with tag %s = %s " + (if Tree.is_nil t then "###" else (Tag.to_string(Tree.tag t))) (Tree.dump_node t); *) + let r = + try + Hashtbl.find h_tdconf tag + with + | Not_found -> + let res = Hashtbl.fold (fun q l acc -> + if List.exists (fun (ts,_) -> TagSet.mem tag ts) l + then Ptset.add q acc + else acc) a.phi Ptset.empty + in Hashtbl.add h_tdconf tag res;res + in +(* let _ = pr ", among "; + pr_st fmt (Ptset.elements r); + pr "\n%!"; + in *) + let r = cons r Nil in + let set,res = top_down (~noright:noright) a t r t 1 in + let set = match set with + | Cons(x,_,Nil) ->x + | _ -> assert false + in +(* pr "Result of topdown run is %!"; + pr_st fmt (Ptset.elements set); + pr ", number is %i\n%!" (RS.length res.(0)); *) + Configuration.add Configuration.empty set res.(0) + + + + let run_bottom_up a t k = + let trlist = Hashtbl.find a.phi (Ptset.choose a.init) + in + let init = List.fold_left + (fun acc (_,(_,f,_)) -> + Ptset.union acc (let (_,_,l) = fst (f.st) in l)) + Ptset.empty trlist + in + let tree1,jump_fun = + match k with + | `TAG (tag) -> + (*Tree.tagged_lowest t tag, fun tree -> Tree.tagged_next tree tag*) + (Tree.tagged_desc tag t, fun tree -> Tree.tagged_foll_below tag tree t) + | `CONTAINS(_) -> (Tree.text_below t,fun tree -> Tree.text_next tree t) + | _ -> assert false + in + let tree2 = jump_fun tree1 in + let rec loop tree next acc = +(* let _ = pr "\n_________________________\nNew iteration\n" in + let _ = pr "Jumping to %s\n%!" (Tree.dump_node tree) in *) + let acc,conf,next_of_next = bottom_up a tree + Configuration.empty next jump_fun (Tree.root tree) true init acc + in + (* let _ = pr "End of first iteration, conf is:\n%!"; + Configuration.pr fmt conf + in *) + let acc = Configuration.IMap.fold + ( fun s res acc -> if Ptset.intersect init s + then RS.concat res acc else acc) conf.Configuration.results acc + in + if Tree.is_nil next_of_next (*|| Tree.equal next next_of_next *)then + acc + else loop next_of_next (jump_fun next_of_next) acc + in + loop tree1 tree2 RS.empty + end + + let top_down_count a t = let module RI = Run(Integer) in Integer.length (RI.run_top_down a t) + let top_down a t = let module RI = Run(IdSet) in (RI.run_top_down a t) + let bottom_up_count a t k = let module RI = Run(Integer) in Integer.length (RI.run_bottom_up a t k) -(* - end -*)