X-Git-Url: http://git.nguyen.vg/gitweb/?a=blobdiff_plain;f=ata.ml;h=9197cbca94ebb7b55354098bbc24aef3c30fcff6;hb=d046f2b9d8e39b30350399eafe122c30ff61c8c1;hp=eb4f39484897b0043845d0f3f08c532726bce9ab;hpb=9728f46b5f256250a4451c0a9bda30ce81be5b8b;p=SXSI%2Fxpathcomp.git diff --git a/ata.ml b/ata.ml index eb4f394..9197cbc 100644 --- a/ata.ml +++ b/ata.ml @@ -1,6 +1,6 @@ INCLUDE "debug.ml" INCLUDE "utils.ml" - +open Camlp4.Struct type jump_kind = [ `TAG of Tag.t | `CONTAINS of string | `NOTHING ] (* Todo : move elsewhere *) @@ -13,9 +13,9 @@ sig end = struct type t = int - let make = - let id = ref (-1) in - fun () -> incr id;!id + let make = let id = ref ~-1 in + fun () -> incr id; !id + let compare = (-) let equal = (==) external hash : t -> int = "%identity" @@ -58,8 +58,8 @@ struct match f.pos with | False -> 0 | True -> 1 - | Or (f1,f2) -> HASHINT3(PRIME2,f1.Node.id, f2.Node.id) - | And (f1,f2) -> HASHINT3(PRIME3,f1.Node.id,f2.Node.id) + | Or (f1,f2) -> HASHINT3(PRIME2,Uid.to_int f1.Node.id, Uid.to_int f2.Node.id) + | And (f1,f2) -> HASHINT3(PRIME3,Uid.to_int f1.Node.id, Uid.to_int f2.Node.id) | Atom(d,p,s) -> HASHINT4(PRIME4,hash_const_variant d,vb p,s) end @@ -194,16 +194,20 @@ end module Transition = struct - type node = State.t*bool*Formula.t*bool + type node = State.t*TagSet.t*bool*Formula.t*bool include Hcons.Make(struct type t = node - let hash (s,m,f,b) = HASHINT4(s,Formula.uid f,vb m,vb b) - let equal (s,b,f,m) (s',b',f',m') = - s == s' && b==b' && m==m' && Formula.equal f f' + let hash (s,ts,m,f,b) = HASHINT5(s,Uid.to_int (TagSet.uid ts), + Uid.to_int (Formula.uid f), + vb m,vb b) + let equal (s,ts,b,f,m) (s',ts',b',f',m') = + s == s' && ts == ts' && b==b' && m==m' && f == f' end) - let print ppf f = let (st,mark,form,b) = node f in - Format.fprintf ppf "%i %s" st (if mark then "⇒" else "→"); + let print ppf f = let (st,ts,mark,form,b) = node f in + Format.fprintf ppf "(%i, " st; + TagSet.print ppf ts; + Format.fprintf ppf ") %s" (if mark then "⇒" else "→"); Formula.print ppf form; Format.fprintf ppf "%s%!" (if b then " (b)" else "") @@ -212,7 +216,7 @@ module Transition = struct let ( ?< ) x = x let ( >< ) state (l,mark) = state,(l,mark,false) let ( ><@ ) state (l,mark) = state,(l,mark,true) - let ( >=> ) (state,(label,mark,bur)) form = (state,label,(make (state,mark,form,bur))) + let ( >=> ) (state,(label,mark,bur)) form = (state,label,(make (state,label,mark,form,bur))) end end @@ -254,7 +258,7 @@ let dump ppf a = if y-x == 0 then TagSet.compare tsy tsx else y-x) l in let maxh,maxt,l_print = List.fold_left ( - fun (maxh,maxt,l) ((ts,q),(_,b,f,_)) -> + fun (maxh,maxt,l) ((ts,q),(_,_,b,f,_)) -> let s = if TagSet.is_finite ts then "{" ^ (TagSet.fold (fun t a -> a ^ " '" ^ (Tag.to_string t)^"'") ts "") ^" }" @@ -282,7 +286,9 @@ module FormTable = Hashtbl.Make(struct let equal (f1,s1,t1) (f2,s2,t2) = f1 == f2 && s1 == s2 && t1 == t2 let hash (f,s,t) = - HASHINT3(Formula.uid f ,StateSet.uid s,StateSet.uid t) + HASHINT3(Uid.to_int (Formula.uid f), + Uid.to_int (StateSet.uid s), + Uid.to_int (StateSet.uid t)) end) module F = Formula @@ -326,39 +332,69 @@ let eval_form_bool = in loop f -module FTable = Hashtbl.Make( struct - type t = Formlist.t*StateSet.t*StateSet.t - let equal (f1,s1,t1) (f2,s2,t2) = - f1 == f2 && s1 == s2 && t1 == t2;; - let hash (f,s,t) = HASHINT3(Formlist.uid f ,StateSet.uid s,StateSet.uid t);; - end) +module FTable = Hashtbl.Make(struct + type t = Tag.t*Formlist.t*StateSet.t*StateSet.t + let equal (tg1,f1,s1,t1) (tg2,f2,s2,t2) = + tg1 == tg2 && f1 == f2 && s1 == s2 && t1 == t2;; + let hash (tg,f,s,t) = + HASHINT4(tg, Uid.to_int (Formlist.uid f), + Uid.to_int (StateSet.uid s), + Uid.to_int (StateSet.uid t)) + end) let h_f = FTable.create BIG_H_SIZE - -let eval_formlist s1 s2 fl = +type merge_conf = NO | ONLY1 | ONLY2 | ONLY12 | MARK | MARK1 | MARK2 | MARK12 +(* 000 001 010 011 100 101 110 111 *) +let eval_formlist tag s1 s2 fl = let rec loop fl = try - FTable.find h_f (fl,s1,s2) + FTable.find h_f (tag,fl,s1,s2) with | Not_found -> match Formlist.node fl with | Formlist.Cons(f,fll) -> - let q,mark,f,_ = Transition.node f in - let b,b1,b2 = eval_form_bool f s1 s2 in + let q,ts,mark,f,_ = Transition.node f in + let b,b1,b2 = + if TagSet.mem tag ts then eval_form_bool f s1 s2 else (false,false,false) + in let (s,(b',b1',b2',amark)) as res = loop fll in let r = if b then (StateSet.add q s, (b, b1'||b1,b2'||b2,mark||amark)) else res - in FTable.add h_f (fl,s1,s2) r;r + in FTable.add h_f (tag,fl,s1,s2) r;r | Formlist.Nil -> StateSet.empty,(false,false,false,false) - in loop fl - + in + let r,conf = loop fl + in + r,(match conf with + | (false,_,_,_) -> NO + | (_,false,false,false) -> NO + | (_,true,false,false) -> ONLY1 + | (_,false,true,false) -> ONLY2 + | (_,true,true,false) -> ONLY12 + | (_,false,false,true) -> MARK + | (_,true,false,true) -> MARK1 + | (_,false,true,true) -> MARK2 + | _ -> MARK12) + +let bool_of_merge conf = + match conf with + | NO -> false,false,false,false + | ONLY1 -> true,true,false,false + | ONLY2 -> true,false,true,false + | ONLY12 -> true,true,true,false + | MARK -> true,false,false,true + | MARK1 -> true,true,false,true + | MARK2 -> true,false,true,true + | MARK12 -> true,true,true,true + + let tags_of_state a q = Hashtbl.fold (fun p l acc -> if p == q then List.fold_left (fun acc (ts,t) -> - let _,_,_,aux = Transition.node t in + let _,_,_,_,aux = Transition.node t in if aux then acc else TagSet.cup ts acc) acc l @@ -382,7 +418,7 @@ let tags_of_state a q = module type ResultSet = sig type t - type elt = [` Tree] Tree.node + type elt = [` Tree ] Tree.node val empty : t val cons : elt -> t -> t val concat : t -> t -> t @@ -390,13 +426,16 @@ let tags_of_state a q = val fold : ( elt -> 'a -> 'a) -> t -> 'a -> 'a val map : ( elt -> elt) -> t -> t val length : t -> int - val merge : (bool*bool*bool*bool) -> elt -> t -> t -> t + val merge : merge_conf -> elt -> t -> t -> t + val mk_quick_tag_loop : (elt -> elt -> 'a*t array) -> 'a -> int -> Tree.t -> Tag.t -> (elt -> elt -> 'a*t array) + val mk_quick_star_loop : (elt -> elt -> 'a*t array) -> 'a -> int -> Tree.t -> (elt -> elt -> 'a*t array) end module Integer : ResultSet = struct type t = int type elt = [`Tree] Tree.node + let empty = 0 let cons _ x = x+1 let concat x y = x + y @@ -404,7 +443,8 @@ let tags_of_state a q = let fold _ _ _ = failwith "fold not implemented" let map _ _ = failwith "map not implemented" let length x = x - let merge (rb,rb1,rb2,mark) t res1 res2 = + let merge2 conf t res1 res2 = + let rb,rb1,rb2,mark = conf in if rb then let res1 = if rb1 then res1 else 0 and res2 = if rb2 then res2 else 0 @@ -412,9 +452,27 @@ let tags_of_state a q = if mark then 1+res1+res2 else res1+res2 else 0 + let merge conf t res1 res2 = + match conf with + NO -> 0 + | ONLY1 -> res1 + | ONLY2 -> res2 + | ONLY12 -> res1+res2 + | MARK -> 1 + | MARK1 -> res1+1 + | MARK2 -> res2+1 + | MARK12 -> res1+res2+1 + + let mk_quick_tag_loop _ sl ss tree tag = (); + fun t ctx -> + (sl, Array.make ss (Tree.subtree_tags tree tag t)) + let mk_quick_star_loop _ sl ss tree = (); + fun t ctx -> + (sl, Array.make ss (Tree.subtree_elements tree t)) + end - module IdSet : ResultSet = + module IdSet : ResultSet= struct type elt = [`Tree] Tree.node type node = Nil @@ -456,51 +514,173 @@ let tags_of_state a q = in { l with node = loop l.node } - let merge (rb,rb1,rb2,mark) t res1 res2 = - if rb then - let res1 = if rb1 then res1 else empty - and res2 = if rb2 then res2 else empty - in - if mark then { node = Cons(t,(Concat(res1.node,res2.node))); - length = res1.length + res2.length + 1;} - else - { node = (Concat(res1.node,res2.node)); - length = res1.length + res2.length ;} - else empty - - + let merge conf t res1 res2 = + match conf with + NO -> empty + | MARK -> cons t empty + | ONLY1 -> res1 + | ONLY2 -> res2 + | ONLY12 -> { node = (Concat(res1.node,res2.node)); + length = res1.length + res2.length ;} + | MARK12 -> { node = Cons(t,(Concat(res1.node,res2.node))); + length = res1.length + res2.length + 1;} + | MARK1 -> { node = Cons(t,res1.node); + length = res1.length + 1;} + | MARK2 -> { node = Cons(t,res2.node); + length = res2.length + 1;} + + let mk_quick_tag_loop f _ _ _ _ = f + let mk_quick_star_loop f _ _ _ = f end - module GResult = struct - type t + module GResult(Doc : sig val doc : Tree.t end) = struct + type bits type elt = [` Tree] Tree.node - external create_empty : int -> t = "caml_result_set_create" - external set : t -> int -> t = "caml_result_set_set" - external next : t -> int -> int = "caml_result_set_next" - external clear : t -> int -> int -> unit = "caml_result_set_clear" - let empty = create_empty 100000000 + external create_empty : int -> bits = "caml_result_set_create" "noalloc" + external set : bits -> int -> unit = "caml_result_set_set" "noalloc" + external next : bits -> int -> int = "caml_result_set_next" "noalloc" + external count : bits -> int = "caml_result_set_count" "noalloc" + external clear : bits -> elt -> elt -> unit = "caml_result_set_clear" "noalloc" + + external set_tag_bits : bits -> Tag.t -> Tree.t -> elt -> elt = "caml_set_tag_bits" "noalloc" + type t = + { segments : elt list; + bits : bits; + } + + let ebits = + let size = (Tree.subtree_size Doc.doc Tree.root) in + create_empty (size*2+1) + + let empty = { segments = []; + bits = ebits } - let cons e t = set t (Obj.magic e) - let concat _ t = t + let cons e t = + let rec loop l = match l with + | [] -> { bits = (set t.bits (Obj.magic e);t.bits); + segments = [ e ] } + | p::r -> + if Tree.is_binary_ancestor Doc.doc e p then + loop r + else + { bits = (set t.bits (Obj.magic e);t.bits); + segments = e::l } + in + loop t.segments + + let concat t1 t2 = + if t2.segments == [] then t1 + else + if t1.segments == [] then t2 + else + let h2 = List.hd t2.segments in + let rec loop l = match l with + | [] -> t2.segments + | p::r -> + if Tree.is_binary_ancestor Doc.doc p h2 then + l + else + p::(loop r) + in + { bits = t1.bits; + segments = loop t1.segments + } + let iter f t = let rec loop i = if i == -1 then () - else (f (Obj.magic i);loop (next t i)) - in loop 0 + else (f ((Obj.magic i):elt);loop (next t.bits i)) + in loop (next t.bits 0) - let fold _ _ _ = failwith "noop" + let fold f t acc = + let rec loop i acc = + if i == -1 then acc + else loop (next t.bits i) (f ((Obj.magic i):elt) acc) + in loop (next t.bits 0) acc + let map _ _ = failwith "noop" - let length t = let cpt = ref ~-1 in - iter (fun _ -> incr cpt) t; !cpt + (*let length t = let cpt = ref 0 in + iter (fun _ -> incr cpt) t; !cpt *) + let length t = count t.bits + let clear_bits t = + let rec loop l = match l with + [] -> () + | idx::ll -> + clear t.bits idx (Tree.closing Doc.doc idx); loop ll + in + loop t.segments;empty + let merge (rb,rb1,rb2,mark) elt t1 t2 = - if mark then (set t1 (Obj.magic elt) ; t1) else t1 - + if rb then +(* let _ = Printf.eprintf "Lenght before merging is %i %i\n" + (List.length t1.segments) (List.length t2.segments) + in *) + match t1.segments,t2.segments with + [],[] -> if mark then cons elt empty else empty + | [_],[] when rb1 -> if mark then cons elt t1 else t1 + | [], [_] when rb2 -> if mark then cons elt t2 else t2 + | [_],[_] when rb1 && rb2 -> if mark then cons elt empty else + concat t1 t2 + | _ -> + let t1 = if rb1 then t1 else clear_bits t1 + and t2 = if rb2 then t2 else clear_bits t2 + in + (if mark then cons elt (concat t1 t2) + else concat t1 t2) + else + let _ = clear_bits t1 in + clear_bits t2 + + let merge conf t t1 t2 = + match t1.segments,t2.segments,conf with + | _,_,NO -> let _ = clear_bits t1 in clear_bits t2 + | [],[],(MARK1|MARK2|MARK12|MARK) -> cons t empty + | [],[],_ -> empty + | [_],[],(ONLY1|ONLY12) -> t1 + | [_],[],(MARK1|MARK12) -> cons t t1 + | [],[_],(ONLY2|ONLY12) -> t2 + | [],[_],(MARK2|MARK12) -> cons t t2 + | [_],[_],ONLY12 -> concat t1 t2 + | [_],[_],MARK12 -> cons t empty + | _,_,MARK -> let _ = clear_bits t2 in cons t (clear_bits t1) + | _,_,ONLY1 -> let _ = clear_bits t2 in t1 + | _,_,ONLY2 -> let _ = clear_bits t1 in t2 + | _,_,ONLY12 -> concat t1 t2 + | _,_,MARK1 -> let _ = clear_bits t2 in cons t t1 + | _,_,MARK2 -> let _ = clear_bits t1 in cons t t2 + | _,_,MARK12 -> cons t (concat t1 t2) + + let mk_quick_tag_loop _ sl ss tree tag = (); + fun t _ -> + let res = empty in + let first = set_tag_bits empty.bits tag tree t in + let res = + if first == Tree.nil then res else + cons first res + in + (sl, Array.make ss res) + + let mk_quick_star_loop f _ _ _ = f end module Run (RS : ResultSet) = struct - module SList = Hlist.Make (StateSet) - + module SList = struct + include Hlist.Make (StateSet) + let print ppf l = + Format.fprintf ppf "[ "; + begin + match l.Node.node with + | Nil -> () + | Cons(s,ll) -> + StateSet.print ppf s; + iter (fun s -> Format.fprintf ppf "; "; + StateSet.print ppf s) ll + end; + Format.fprintf ppf "]%!" + + + end IFDEF DEBUG @@ -512,113 +692,14 @@ END let mk_fun f s = D_IGNORE_(register_funname f s,f) let mk_app_fun f arg s = let g = f arg in D_IGNORE_(register_funname g ((get_funname f) ^ " " ^ s), g) + let mk_app_fun f arg _ = f arg let mk_app_fun2 f arg1 arg2 s = let g = f arg1 arg2 in D_IGNORE_(register_funname g ((get_funname f) ^ " " ^ s), g) +(* let mk_app_fun2 f arg1 arg2 s = Printf.eprintf "Building f2 %s\n%!" s; f arg1 arg2 *) let string_of_ts tags = (Ptset.Int.fold (fun t a -> a ^ " " ^ (Tag.to_string t) ) tags "{")^ " }" -(* - module Algebra = - struct - type jump = [ `LONG | `CLOSE | `NIL ] - type t = jump*Ptset.Int.t - - let merge_jump (j1,l1) (j2,l2) = - match j1,j2 with - | _ when j1 = j2 -> (j1,Ptset.Int.union l1 l2) - | _,`NIL -> j1,l1 - | `NIL,_ -> j2,l2 - | _,_ -> (`CLOSE, Ptset.Int.union l1 l2) - - let merge_jump_list = function - | [] -> `NIL,Ptset.Int.empty - | p::r -> List.fold_left (merge_jump) p r - - let labels a s = - Hashtbl.fold - ( - fun q l acc -> - if (q == s) - then - - (List.fold_left - (fun acc (ts,f) -> - let _,_,_,bur = Transition.node f in - if bur then acc else TagSet.cup acc ts) - acc l) - else acc ) a.trans TagSet.empty - exception Found - - let is_rec a s access = - List.exists - (fun (_,t) -> let _,_,f,_ = Transition.node t in - StateSet.mem s (access f)) (Hashtbl.find a.trans s) - - - let decide a c_label l_label dir_states access = - - let l = StateSet.fold - (fun s l -> - let s_rec= is_rec a s access in - let tlabels,jmp = - if s_rec then l_label,`LONG - else c_label,`CLOSE in - let slabels = TagSet.positive ((TagSet.cap (labels a s) tlabels)) - in - (if Ptset.Int.is_empty slabels - then `NIL,Ptset.Int.empty - else jmp,slabels)::l) dir_states [] - in merge_jump_list l - - - - - - end - - - let choose_jump tagset qtags1 qtagsn a f_nil f_t1 f_s1 f_tn f_sn f_notext f_maytext = - let tags1,hastext1,fin1 = inter_text tagset (tags a qtags1) in - let tagsn,hastextn,finn = inter_text tagset (tags a qtagsn) in - (*if (hastext1||hastextn) then (`ANY,f_text) (* jumping to text nodes doesn't work really well *) - else*) - if (Ptset.Int.is_empty tags1) && (Ptset.Int.is_empty tagsn) then (`NIL,f_nil) - else if (Ptset.Int.is_empty tagsn) then - if (Ptset.Int.is_singleton tags1) - then (* TaggedChild/Sibling *) - let tag = (Ptset.Int.choose tags1) in (`TAG(tag),mk_app_fun f_t1 tag (Tag.to_string tag)) - else (* SelectChild/Sibling *) - (`ANY,mk_app_fun f_s1 tags1 (string_of_ts tags1)) - else if (Ptset.Int.is_empty tags1) then - if (Ptset.Int.is_singleton tagsn) - then (* TaggedDesc/Following *) - let tag = (Ptset.Int.choose tagsn) in (`TAG(tag),mk_app_fun f_tn tag (Tag.to_string tag)) - else (* SelectDesc/Following *) - (`ANY,mk_app_fun f_sn tagsn (string_of_ts tagsn)) - else if (hastext1||hastextn) then (`ANY,f_maytext) - else (`ANY,f_notext) - - let choose_jump_down tree a b c d = - choose_jump a b c d - (mk_fun (fun _ -> Tree.nil) "Tree.mk_nil") - (mk_fun (Tree.tagged_child tree) "Tree.tagged_child") - (mk_fun (Tree.select_child tree) "Tree.select_child") - (mk_fun (Tree.tagged_desc tree) "Tree.tagged_desc") - (mk_fun (Tree.select_desc tree) "Tree.select_desc") - (mk_fun (Tree.first_element tree) "Tree.first_element") - (mk_fun (Tree.first_child tree) "Tree.first_child") - - let choose_jump_next tree a b c d = - choose_jump a b c d - (mk_fun (fun _ _ -> Tree.nil) "Tree.mk_nil2") - (mk_fun (Tree.tagged_sibling_ctx tree) "Tree.tagged_sibling_ctx") - (mk_fun (Tree.select_sibling_ctx tree) "Tree.select_sibling_ctx") - (mk_fun (Tree.tagged_foll_ctx tree) "Tree.tagged_foll_ctx") - (mk_fun (Tree.select_foll_ctx tree) "Tree.select_foll_ctx") - (mk_fun (Tree.next_element_ctx tree) "Tree.node_element_ctx") - (mk_fun (Tree.next_sibling_ctx tree) "Tree.node_sibling_ctx") -*) module Algebra = struct type jump = [ `NIL | `ANY |`ANYNOTEXT | `JUMP ] @@ -660,7 +741,7 @@ END (List.fold_left (fun acc (ts,f) -> - let _,_,_,bur = Transition.node f in + let _,_,_,_,bur = Transition.node f in if bur then acc else TagSet.cup acc ts) acc l) else acc ) a.trans TagSet.empty @@ -668,10 +749,14 @@ END let is_rec a s access = List.exists - (fun (_,t) -> let _,_,f,_ = Transition.node t in + (fun (_,t) -> let _,_,_,f,_ = Transition.node t in StateSet.mem s ((fun (_,_,x) -> x) (access (Formula.st f)))) (Hashtbl.find a.trans s) - + let is_final_marking a s = + List.exists (fun (_,t) -> let _,_,m,f,_ = Transition.node t in m&& (Formula.is_true f)) + (Hashtbl.find a.trans s) + + let decide a c_label l_label dir_states dir = let l = StateSet.fold @@ -718,13 +803,13 @@ END let tag = Ptset.Int.choose ll in (`TAG(tag),mk_app_fun f_tn tag (Tag.to_string tag)) else - (`ANY,mk_app_fun f_sn ll (string_of_ts ll)) + (`MANY(ll),mk_app_fun f_sn ll (string_of_ts ll)) else if Ptset.Int.is_empty ll then if Ptset.Int.is_singleton cl then let tag = Ptset.Int.choose cl in (`TAG(tag),mk_app_fun f_t1 tag (Tag.to_string tag)) else - (`ANY,mk_app_fun f_s1 cl (string_of_ts cl)) + (`MANY(cl),mk_app_fun f_s1 cl (string_of_ts cl)) else (`ANY,mk_app_fun2 f_s1n cl ll ((string_of_ts cl) ^ " " ^ (string_of_ts ll))) @@ -735,8 +820,8 @@ END (mk_fun (fun _ -> Tree.nil) "Tree.mk_nil") (mk_fun (Tree.tagged_child tree) "Tree.tagged_child") (mk_fun (Tree.select_child tree) "Tree.select_child") - (mk_fun (Tree.tagged_desc tree) "Tree.tagged_desc") - (mk_fun (Tree.select_desc tree) "Tree.select_desc") + (mk_fun (Tree.tagged_descendant tree) "Tree.tagged_desc") + (mk_fun (Tree.select_descendant tree) "Tree.select_desc") (mk_fun (fun _ _ -> Tree.first_child tree) "[FIRSTCHILD]Tree.select_child_desc") (mk_fun (Tree.first_element tree) "Tree.first_element") (mk_fun (Tree.first_child tree) "Tree.first_child") @@ -744,194 +829,316 @@ END let choose_jump_next tree d = choose_jump d (mk_fun (fun _ _ -> Tree.nil) "Tree.mk_nil2") - (mk_fun (Tree.tagged_sibling_ctx tree) "Tree.tagged_sibling_ctx") - (mk_fun (Tree.select_sibling_ctx tree) "Tree.select_sibling_ctx") - (mk_fun (Tree.tagged_foll_ctx tree) "Tree.tagged_foll_ctx") - (mk_fun (Tree.select_foll_ctx tree) "Tree.select_foll_ctx") - (mk_fun (fun _ _ -> Tree.next_sibling_ctx tree) "[NEXTSIBLING]Tree.select_sibling_foll_ctx") - (mk_fun (Tree.next_element_ctx tree) "Tree.next_element_ctx") - (mk_fun (Tree.next_sibling_ctx tree) "Tree.node_sibling_ctx") + (mk_fun (Tree.tagged_following_sibling_below tree) "Tree.tagged_sibling_ctx") + (mk_fun (Tree.select_following_sibling_below tree) "Tree.select_sibling_ctx") + (mk_fun (Tree.tagged_following_below tree) "Tree.tagged_foll_ctx") + (mk_fun (Tree.select_following_below tree) "Tree.select_foll_ctx") + (mk_fun (fun _ _ -> Tree.next_sibling_below tree) "[NEXTSIBLING]Tree.select_sibling_foll_ctx") + (mk_fun (Tree.next_element_below tree) "Tree.next_element_ctx") + (mk_fun (Tree.next_sibling_below tree) "Tree.node_sibling_ctx") - module SetTagKey = - struct - type t = Tag.t*SList.t - let equal (t1,s1) (t2,s2) = t1 == t2 && s1 == s2 - let hash (t,s) = HASHINT2(t,s.SList.Node.id) + + + + module CodeCache = + struct + let get = Array.unsafe_get + let set = Array.set + + type fun_tree = [`Tree] Tree.node -> [`Tree] Tree.node -> SList.t -> Tag.t -> bool -> SList.t*RS.t array + type t = fun_tree array array + + let dummy = fun _ _ _ _ _ -> failwith "Uninitializd CodeCache" + let default_line = Array.create 256 dummy (* 256 = max_tag *) + let create n = Array.create n default_line + let init f = + for i = 0 to (Array.length default_line) - 1 + do + default_line.(i) <- f + done + + let get_fun h slist tag = + get (get h (Uid.to_int slist.SList.Node.id)) tag + + let set_fun (h : t) slist tag (data : fun_tree) = + let tab = get h (Uid.to_int slist.SList.Node.id) in + let line = if tab == default_line then + let x = Array.copy tab in + (set h (Uid.to_int slist.SList.Node.id) x;x) + else tab + in + set line tag data + end - - module CachedTransTable = Hashtbl.Make(SetTagKey) - let td_trans = CachedTransTable.create 4093 - - + + let td_trans = CodeCache.create 10000 (* should be number of tags *number of states^2 + in the document *) + let empty_size n = let rec loop acc = function 0 -> acc | n -> loop (SList.cons StateSet.empty acc) (n-1) in loop SList.nil n + + module Fold2Res = struct + let get = Array.unsafe_get + let set = Array.set + external field1 : Obj.t -> int = "%field1" + type t = Obj.t array array array array + let dummy_val = Obj.repr ((),2,()) + + let default_line3 = Array.create 10000 dummy_val + let default_line2 = Array.create 10000 default_line3 + let default_line1 = Array.create 10000 default_line2 + + let create n = Array.create n default_line1 + + let find h tag fl s1 s2 : SList.t*bool*(merge_conf array) = + let l1 = get h tag in + let l2 = get l1 (Uid.to_int fl.Formlistlist.Node.id) in + let l3 = get l2 (Uid.to_int s1.SList.Node.id) in + Obj.magic (get l3 (Uid.to_int s2.SList.Node.id)) + + let is_valid b = (Obj.magic b) != 2 + let get_replace tab idx default = + let e = get tab idx in + if e == default then + let ne = Array.copy e in (set tab idx ne;ne) + else e + + let add h tag fl s1 s2 (data: SList.t*bool*(merge_conf array)) = + let l1 = get_replace h tag default_line1 in + let l2 = get_replace l1 (Uid.to_int fl.Formlistlist.Node.id) default_line2 in + let l3 = get_replace l2 (Uid.to_int s1.SList.Node.id) default_line3 in + set l3 (Uid.to_int s2.SList.Node.id) (Obj.repr data) + end - module Fold2Res = Hashtbl.Make(struct - type t = Formlistlist.t*SList.t*SList.t - let hash (f,s,t) = HASHINT3(f.Formlistlist.Node.id, - s.SList.Node.id, - t.SList.Node.id) - let equal (a,b,c) (d,e,f) = a==d && b == e && c == f - end) - - let h_fold2 = Fold2Res.create BIG_H_SIZE + + let h_fold2 = Fold2Res.create 256 let top_down ?(noright=false) a tree t slist ctx slot_size = let pempty = empty_size slot_size in let rempty = Array.make slot_size RS.empty in (* evaluation starts from the right so we put sl1,res1 at the end *) - let eval_fold2_slist fll t (sl2,res2) (sl1,res1) = + let eval_fold2_slist fll t tag (sl2,res2) (sl1,res1) = let res = Array.copy rempty in - try - let r,b,btab = Fold2Res.find h_fold2 (fll,sl1,sl2) in + let r,b,btab = Fold2Res.find h_fold2 tag fll sl1 sl2 in + if Fold2Res.is_valid b then + begin + if b then for i=0 to slot_size - 1 do + res.(0) <- RS.merge btab.(0) t res1.(0) res2.(0); + done; + r,res + end + else + begin + let btab = Array.make slot_size NO in + let rec fold l1 l2 fll i aq ab = + match fll.Formlistlist.Node.node, + l1.SList.Node.node, + l2.SList.Node.node + with + | Formlistlist.Cons(fl,fll), + SList.Cons(s1,ll1), + SList.Cons(s2,ll2) -> + let r',conf = eval_formlist tag s1 s2 fl in + let _ = btab.(i) <- conf + in + fold ll1 ll2 fll (i+1) (SList.cons r' aq) ((conf!=NO)||ab) + | _ -> aq,ab + in + let r,b = fold sl1 sl2 fll 0 SList.nil false in + Fold2Res.add h_fold2 tag fll sl1 sl2 (r,b,btab); if b then for i=0 to slot_size - 1 do res.(i) <- RS.merge btab.(i) t res1.(i) res2.(i); done; - r,res - with - Not_found -> - let btab = Array.make slot_size (false,false,false,false) in - let rec fold l1 l2 fll i aq ab = - match fll.Formlistlist.Node.node, - l1.SList.Node.node, - l2.SList.Node.node - with - | Formlistlist.Cons(fl,fll), - SList.Cons(s1,ll1), - SList.Cons(s2,ll2) -> - let r',((b,_,_,_) as flags) = eval_formlist s1 s2 fl in - let _ = btab.(i) <- flags - in - fold ll1 ll2 fll (i+1) (SList.cons r' aq) (b||ab) - | _ -> aq,ab - in - let r,b = fold sl1 sl2 fll 0 SList.nil false in - Fold2Res.add h_fold2 (fll,sl1,sl2) (r,b,btab); - if b then for i=0 to slot_size - 1 do - res.(i) <- RS.merge btab.(i) t res1.(i) res2.(i); - done; - r,res + r,res; + end in let null_result = (pempty,Array.copy rempty) in - let rec loop t slist ctx= - if t == Tree.nil then null_result else get_trans t slist (Tree.tag tree t) ctx - and loop_tag tag t slist ctx = - if t == Tree.nil then null_result else get_trans t slist tag ctx - and loop_no_right t slist ctx = - if t == Tree.nil then null_result else get_trans ~noright:true t slist (Tree.tag tree t) ctx - and get_trans ?(noright=false) t slist tag ctx = - let cont = - try - CachedTransTable.find td_trans (tag,slist) - with - | Not_found -> - let fl_list,llist,rlist,ca,da,sa,fa = - SList.fold - (fun set (fll_acc,lllacc,rllacc,ca,da,sa,fa) -> (* For each set *) - let fl,ll,rr,ca,da,sa,fa = - StateSet.fold - (fun q acc -> - List.fold_left - (fun ((fl_acc,ll_acc,rl_acc,c_acc,d_acc,s_acc,f_acc) as acc) - (ts,t) -> - if (TagSet.mem tag ts) - then - let _,_,f,_ = Transition.node t in - let (child,desc,below),(sibl,foll,after) = Formula.st f in - (Formlist.cons t fl_acc, - StateSet.union ll_acc below, - StateSet.union rl_acc after, - StateSet.union child c_acc, - StateSet.union desc d_acc, - StateSet.union sibl s_acc, - StateSet.union foll f_acc) - else acc ) acc ( - try Hashtbl.find a.trans q - with - Not_found -> Printf.eprintf "Looking for state %i, doesn't exist!!!\n%!" - q;[] - ) - - ) set (Formlist.nil,StateSet.empty,StateSet.empty,ca,da,sa,fa) - in (Formlistlist.cons fl fll_acc), (SList.cons ll lllacc), (SList.cons rr rllacc),ca,da,sa,fa) - slist (Formlistlist.nil,SList.nil,SList.nil,StateSet.empty,StateSet.empty,StateSet.empty,StateSet.empty) - in - (* Logic to chose the first and next function *) - let tags_child,tags_below,tags_siblings,tags_after = Tree.tags tree tag in - let d_f = Algebra.decide a tags_child tags_below (StateSet.union ca da) true in - let d_n = Algebra.decide a tags_siblings tags_after (StateSet.union sa fa) false in -(* let _ = Printf.eprintf "Tags below %s are : \n" (Tag.to_string tag) in - let _ = Ptset.Int.iter (fun i -> Printf.eprintf "%s " (Tag.to_string i)) tags_below in - let _ = Printf.eprintf "\n%!" in *) -(* let tags_below = Ptset.Int.remove tag tags_below in *) - let f_kind,first = choose_jump_down tree d_f - and n_kind,next = if noright then (`NIL, fun _ _ -> Tree.nil ) - else choose_jump_next tree d_n in - let empty_res = null_result in - let cont = - match f_kind,n_kind with - | `NIL,`NIL -> - (fun t _ -> eval_fold2_slist fl_list t empty_res empty_res ) - | _,`NIL -> ( - match f_kind with - |`TAG(tag) -> - (fun t _ -> eval_fold2_slist fl_list t empty_res - (loop_tag tag (first t) llist t)) - | `ANY -> - (fun t _ -> eval_fold2_slist fl_list t empty_res - (loop (first t) llist t)) - | _ -> assert false) - - | `NIL,_ -> ( - match n_kind with - |`TAG(tag) -> - (fun t ctx -> eval_fold2_slist fl_list t - (loop_tag tag (next t ctx) rlist ctx) empty_res) - - | `ANY -> - (fun t ctx -> eval_fold2_slist fl_list t - (loop (next t ctx) rlist ctx) empty_res) - - | _ -> assert false) - - | `TAG(tag1),`TAG(tag2) -> - (fun t ctx -> eval_fold2_slist fl_list t - (loop_tag tag2 (next t ctx) rlist ctx) - (loop_tag tag1 (first t) llist t)) - - | `TAG(tag),`ANY -> - (fun t ctx -> eval_fold2_slist fl_list t - (loop (next t ctx) rlist ctx) - (loop_tag tag (first t) llist t)) - | `ANY,`TAG(tag) -> - (fun t ctx -> - eval_fold2_slist fl_list t - (loop_tag tag (next t ctx) rlist ctx) - (loop (first t) llist t) ) - | `ANY,`ANY -> - (fun t ctx -> - eval_fold2_slist fl_list t - (loop (next t ctx) rlist ctx) - (loop (first t) llist t) ) - | _ -> assert false - in - let cont = D_IF_( (fun t ctx -> - let a,b = cont t ctx in - register_trace tree t (slist,a,fl_list,first,next,ctx); - (a,b) - ) ,cont) - in - (CachedTransTable.add td_trans (tag,slist) cont;cont) - in cont t ctx - - in - (if noright then loop_no_right else loop) t slist ctx + let empty_res = null_result in + + let rec loop t ctx slist _ = + if t == Tree.nil then null_result else + let tag = Tree.tag tree t in + (CodeCache.get_fun td_trans slist tag) t ctx slist tag false + (* get_trans t ctx slist tag false + (CodeCache.get_opcode td_trans slist tag) + *) + and loop_tag t ctx slist tag = + if t == Tree.nil then null_result else + (CodeCache.get_fun td_trans slist tag) t ctx slist tag false + (* get_trans t ctx slist tag false + (CodeCache.get_opcode td_trans slist tag) *) + + and loop_no_right t ctx slist _ = + if t == Tree.nil then null_result else + let tag = Tree.tag tree t in + (CodeCache.get_fun td_trans slist tag) t ctx slist tag true + (* get_trans t ctx slist tag true + (CodeCache.get_opcode td_trans slist tag) *) + (* + and get_trans t ctx slist tag noright opcode = + match opcode with + | OpCode.K0 fll -> + eval_fold2_slist fll t tag empty_res empty_res + + | OpCode.K1 (fll,first,llist,tag1) -> + eval_fold2_slist fll t tag empty_res + (loop_tag (first t) t llist tag1) + + | OpCode.K2 (fll,first,llist) -> + eval_fold2_slist fll t tag empty_res + (loop (first t) t llist) + + | OpCode.K3 (fll,next,rlist,tag2) -> + eval_fold2_slist fll t tag + (loop_tag (next t ctx) ctx rlist tag2) + empty_res + | OpCode.K4 (fll,next,rlist) -> + eval_fold2_slist fll t tag + (loop (next t ctx) ctx rlist) + empty_res + + | OpCode.K5 (fll,next,rlist,tag2,first,llist,tag1) -> + eval_fold2_slist fll t tag + (loop_tag (next t ctx) ctx rlist tag2) + (loop_tag (first t) t llist tag1) + + | OpCode.K6 (fll,next,rlist,first,llist,tag1) -> + eval_fold2_slist fll t tag + (loop (next t ctx) ctx rlist) + (loop_tag (first t) t llist tag1) + + | OpCode.K7 (fll,next,rlist,tag2,first,llist) -> + eval_fold2_slist fll t tag + (loop_tag (next t ctx) ctx rlist tag2) + (loop (first t) t llist) + + | OpCode.K8 (fll,next,rlist,first,llist) -> + eval_fold2_slist fll t tag + (loop (next t ctx) ctx rlist) + (loop (first t) t llist) + + | OpCode.KDefault _ -> + mk_trans t ctx tag slist noright + *) + and mk_trans t ctx slist tag noright = + let fl_list,llist,rlist,ca,da,sa,fa = + SList.fold + (fun set (fll_acc,lllacc,rllacc,ca,da,sa,fa) -> (* For each set *) + let fl,ll,rr,ca,da,sa,fa = + StateSet.fold + (fun q acc -> + List.fold_left + (fun ((fl_acc,ll_acc,rl_acc,c_acc,d_acc,s_acc,f_acc) as acc) + (ts,t) -> + if (TagSet.mem tag ts) + then + let _,_,_,f,_ = t.Transition.node in + let (child,desc,below),(sibl,foll,after) = Formula.st f in + (Formlist.cons t fl_acc, + StateSet.union ll_acc below, + StateSet.union rl_acc after, + StateSet.union child c_acc, + StateSet.union desc d_acc, + StateSet.union sibl s_acc, + StateSet.union foll f_acc) + else acc ) acc ( + try Hashtbl.find a.trans q + with + Not_found -> Printf.eprintf "Looking for state %i, doesn't exist!!!\n%!" + q;[] + ) + + ) set (Formlist.nil,StateSet.empty,StateSet.empty,ca,da,sa,fa) + in (Formlistlist.cons fl fll_acc), (SList.cons ll lllacc), (SList.cons rr rllacc),ca,da,sa,fa) + slist (Formlistlist.nil,SList.nil,SList.nil,StateSet.empty,StateSet.empty,StateSet.empty,StateSet.empty) + in + (* Logic to chose the first and next function *) + let tags_child,tags_below,tags_siblings,tags_after = Tree.tags tree tag in + let d_f = Algebra.decide a tags_child tags_below (StateSet.union ca da) true in + let d_n = Algebra.decide a tags_siblings tags_after (StateSet.union sa fa) false in + let f_kind,first = choose_jump_down tree d_f + and n_kind,next = if noright then (`NIL, fun _ _ -> Tree.nil ) + else choose_jump_next tree d_n in + let empty_res = null_result in + let fll = fl_list in + let cont = + match f_kind,n_kind with + | `NIL,`NIL -> (*OpCode.K0(fl_list) *) + fun t _ _ tag _ -> eval_fold2_slist fll t tag empty_res empty_res + + | _,`NIL -> ( + match f_kind with + |`TAG(tag1) -> (*OpCode.K1(fl_list,first,llist,tag1) *) + fun t _ _ tag _ -> eval_fold2_slist fll t tag empty_res + (loop_tag (first t) t llist tag1) + | _ -> (* OpCode.K2(fl_list,first,llist) *) + fun t _ _ tag _ -> eval_fold2_slist fll t tag empty_res + (loop (first t) t llist tag) + ) + | `NIL,_ -> ( + match n_kind with + |`TAG(tag2) -> (*OpCode.K3(fl_list,next,rlist,tag2) *) + fun t ctx _ tag _ -> + eval_fold2_slist fll t tag + (loop_tag (next t ctx) ctx rlist tag2) + empty_res + + | _ -> (*OpCode.K4(fl_list,next,rlist) *) + fun t ctx _ tag _ -> + eval_fold2_slist fll t tag + (loop (next t ctx) ctx rlist tag) + empty_res + + ) + + | `TAG(tag1),`TAG(tag2) -> (*OpCode.K5(fl_list,next,rlist,tag2,first,llist,tag1) *) + fun t ctx _ tag _ -> + eval_fold2_slist fll t tag + (loop_tag (next t ctx) ctx rlist tag2) + (loop_tag (first t) t llist tag1) + + | `TAG(tag1),`ANY -> (* OpCode.K6(fl_list,next,rlist,first,llist,tag1) *) + fun t ctx _ tag _ -> + eval_fold2_slist fll t tag + (loop (next t ctx) ctx rlist tag) + (loop_tag (first t) t llist tag1) + + | `ANY,`TAG(tag2) -> (* OpCode.K7(fl_list,next,rlist,tag2,first,llist) *) + fun t ctx _ tag _ -> + eval_fold2_slist fll t tag + (loop_tag (next t ctx) ctx rlist tag2) + (loop (first t) t llist tag) + + + | _,_ -> (*OpCode.K8(fl_list,next,rlist,first,llist) *) + (*if SList.equal slist rlist && SList.equal slist llist + then + let rec loop t ctx = + if t == Tree.nil then empty_res else + let r1 = loop (first t) t + and r2 = loop (next t ctx) ctx + in + eval_fold2_slist fl_list t (Tree.tag tree t) r2 r1 + in loop + else *) + fun t ctx _ tag _ -> + eval_fold2_slist fll t tag + (loop (next t ctx) ctx rlist tag) + (loop (first t) t llist tag) + + + + in + CodeCache.set_fun td_trans slist tag cont; + cont t ctx slist tag noright + in + let _ = CodeCache.init mk_trans in + (if noright then loop_no_right else loop) t ctx slist Tag.dummy - let run_top_down a tree = let init = SList.cons a.init SList.nil in let _,res = top_down a tree Tree.root init Tree.root 1 @@ -958,7 +1165,7 @@ END 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.Int.uid s); + { hash = HASHINT2(c.hash,Uid.to_int (Ptset.Int.uid s)); sets = Ptss.add s c.sets; results = IMap.add s r c.results } @@ -992,7 +1199,7 @@ END in let h,s = Ptss.fold - (fun s (ah,ass) -> (HASHINT2(ah,Ptset.Int.uid s), + (fun s (ah,ass) -> (HASHINT2(ah, Uid.to_int (Ptset.Int.uid s)), Ptss.add s ass)) (Ptss.union c1.sets c2.sets) (0,Ptss.empty) in @@ -1004,21 +1211,23 @@ END let h_fold = Hashtbl.create 511 - let fold_f_conf t slist fl_list conf dir= + let fold_f_conf tree t slist fl_list conf dir= + let tag = Tree.tag tree t in let rec loop sl fl acc = match SList.node sl,fl with |SList.Nil,[] -> acc |SList.Cons(s,sll), formlist::fll -> - let r',(rb,rb1,rb2,mark) = + let r',mcnf = let key = SList.hash sl,Formlist.hash formlist,dir in try Hashtbl.find h_fold key with Not_found -> let res = - if dir then eval_formlist s Ptset.Int.empty formlist - else eval_formlist Ptset.Int.empty s formlist + if dir then eval_formlist tag s Ptset.Int.empty formlist + else eval_formlist tag Ptset.Int.empty s formlist in (Hashtbl.add h_fold key res;res) in + let (rb,rb1,rb2,mark) = bool_of_merge mcnf in if rb && ((dir&&rb1)|| ((not dir) && rb2)) then let acc = @@ -1037,7 +1246,7 @@ END let h_trans = Hashtbl.create 4096 let get_up_trans slist ptag a tree = - let key = (HASHINT2(SList.uid slist,ptag)) in + let key = (HASHINT2(Uid.to_int slist.SList.Node.id ,ptag)) in try Hashtbl.find h_trans key with @@ -1086,7 +1295,7 @@ END let slist = Configuration.Ptss.fold (fun e a -> SList.cons e a) conf.Configuration.sets SList.nil in let fl_list = get_up_trans slist ptag a parent in let slist = SList.rev (slist) in - let newconf = fold_f_conf parent slist fl_list conf dir in + let newconf = fold_f_conf tree parent slist fl_list conf dir in let accu,newconf = Configuration.IMap.fold (fun s res (ar,nc) -> if Ptset.Int.intersect s init then ( RS.concat res ar ,nc) @@ -1129,7 +1338,7 @@ END in let init = List.fold_left (fun acc (_,t) -> - let _,_,f,_ = Transition.node t in + let _,_,_,f,_ = Transition.node t in let _,_,l = fst ( Formula.st f ) in StateSet.union acc l) StateSet.empty trlist @@ -1138,7 +1347,7 @@ END match k with | `TAG (tag) -> (*Tree.tagged_lowest t tag, fun tree -> Tree.tagged_next tree tag*) - (Tree.tagged_desc tree tag t, let jump = Tree.tagged_foll_ctx tree tag + (Tree.tagged_descendant tree tag t, let jump = Tree.tagged_following_below tree tag in fun n -> jump n t ) | `CONTAINS(_) -> (Tree.text_below tree t,let jump = Tree.text_next tree in fun n -> jump n t) @@ -1162,8 +1371,15 @@ END 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 top_down_count a t = let module RI = Run(Integer) in let r = Integer.length (RI.run_top_down a t) + in (*RI.TransCache.dump RI.td_trans; *)r + 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) + let bottom_up a t k = let module RI = Run(IdSet) in (RI.run_bottom_up a t k) + module Test (Doc : sig val doc : Tree.t end) = + struct + module Results = GResult(Doc) + let top_down a t = let module R = Run(Results) in (R.run_top_down a t) + end