src/HOL/Tools/res_atp.ML
author paulson
Wed Jan 31 14:03:31 2007 +0100 (2007-01-31)
changeset 22217 a5d983f7113f
parent 22193 62753ae847a2
child 22382 dbf09db0a40d
permissions -rw-r--r--
Tidying; more debugging information. New reference unwanted_types.
     1 (*  Author: Jia Meng, Cambridge University Computer Laboratory, NICTA
     2     ID: $Id$
     3     Copyright 2004 University of Cambridge
     4 
     5 ATPs with TPTP format input.
     6 *)
     7 
     8 (*Currently unused, during debugging*)
     9 signature RES_ATP =
    10 sig
    11   val prover: string ref
    12   val custom_spass: string list ref
    13   val destdir: string ref
    14   val helper_path: string -> string -> string
    15   val problem_name: string ref
    16   val time_limit: int ref
    17   val set_prover: string -> unit
    18 
    19   datatype mode = Auto | Fol | Hol
    20   val linkup_logic_mode : mode ref
    21   val write_subgoal_file: bool -> mode -> Proof.context -> thm list -> thm list -> int -> string
    22   val vampire_time: int ref
    23   val eprover_time: int ref
    24   val spass_time: int ref
    25   val run_vampire: int -> unit
    26   val run_eprover: int -> unit
    27   val run_spass: int -> unit
    28   val vampireLimit: unit -> int
    29   val eproverLimit: unit -> int
    30   val spassLimit: unit -> int
    31   val atp_method: (Proof.context -> thm list -> int -> tactic) ->
    32     Method.src -> Proof.context -> Proof.method
    33   val cond_rm_tmp: string -> unit
    34   val hol_full_types: unit -> unit
    35   val hol_partial_types: unit -> unit
    36   val hol_const_types_only: unit -> unit
    37   val hol_no_types: unit -> unit
    38   val hol_typ_level: unit -> ResHolClause.type_level
    39   val include_all: bool ref
    40   val run_relevance_filter: bool ref
    41   val run_blacklist_filter: bool ref
    42   val blacklist: string list ref
    43   val add_all : unit -> unit
    44   val add_claset : unit -> unit
    45   val add_simpset : unit -> unit
    46   val add_clasimp : unit -> unit
    47   val add_atpset : unit -> unit
    48   val rm_all : unit -> unit
    49   val rm_claset : unit -> unit
    50   val rm_simpset : unit -> unit
    51   val rm_atpset : unit -> unit
    52   val rm_clasimp : unit -> unit
    53   val is_fol_thms : thm list -> bool
    54 end;
    55 
    56 structure ResAtp =
    57 struct
    58 
    59 fun timestamp s = Output.debug (fn () => ("At " ^ Time.toString (Time.now()) ^ ": " ^ s));
    60 
    61 (********************************************************************)
    62 (* some settings for both background automatic ATP calling procedure*)
    63 (* and also explicit ATP invocation methods                         *)
    64 (********************************************************************)
    65 
    66 (*** background linkup ***)
    67 val call_atp = ref false;
    68 val hook_count = ref 0;
    69 val time_limit = ref 60;
    70 val prover = ref "";
    71 
    72 fun set_prover atp =
    73   case String.map Char.toLower atp of
    74       "e" =>
    75           (ReduceAxiomsN.max_new := 100;
    76            ReduceAxiomsN.theory_const := false;
    77            prover := "E")
    78     | "spass" =>
    79           (ReduceAxiomsN.max_new := 40;
    80            ReduceAxiomsN.theory_const := true;
    81            prover := "spass")
    82     | "vampire" =>
    83           (ReduceAxiomsN.max_new := 60;
    84            ReduceAxiomsN.theory_const := false;
    85            prover := "vampire")
    86     | _ => error ("No such prover: " ^ atp);
    87 
    88 val _ = set_prover "E"; (* use E as the default prover *)
    89 
    90 val custom_spass =   (*specialized options for SPASS*)
    91       ref ["-Auto=0","-FullRed=0","-IORe","-IOFc","-RTaut","-RFSub","-RBSub"];
    92 val destdir = ref "";   (*Empty means write files to /tmp*)
    93 val problem_name = ref "prob";
    94 
    95 (*Return the path to a "helper" like SPASS or tptp2X, first checking that
    96   it exists.  FIXME: modify to use Path primitives and move to some central place.*)
    97 fun helper_path evar base =
    98   case getenv evar of
    99       "" => error  ("Isabelle environment variable " ^ evar ^ " not defined")
   100     | home =>
   101         let val path = home ^ "/" ^ base
   102         in  if File.exists (File.explode_platform_path path) then path
   103             else error ("Could not find the file " ^ path)
   104         end;
   105 
   106 fun probfile_nosuffix _ =
   107   if !destdir = "" then File.platform_path (File.tmp_path (Path.basic (!problem_name)))
   108   else if File.exists (File.explode_platform_path (!destdir))
   109   then !destdir ^ "/" ^ !problem_name
   110   else error ("No such directory: " ^ !destdir);
   111 
   112 fun prob_pathname n = probfile_nosuffix n ^ "_" ^ Int.toString n;
   113 
   114 
   115 (*** ATP methods ***)
   116 val vampire_time = ref 60;
   117 val eprover_time = ref 60;
   118 val spass_time = ref 60;
   119 
   120 fun run_vampire time =
   121     if (time >0) then vampire_time:= time
   122     else vampire_time:=60;
   123 
   124 fun run_eprover time =
   125     if (time > 0) then eprover_time:= time
   126     else eprover_time:=60;
   127 
   128 fun run_spass time =
   129     if (time > 0) then spass_time:=time
   130     else spass_time:=60;
   131 
   132 
   133 fun vampireLimit () = !vampire_time;
   134 fun eproverLimit () = !eprover_time;
   135 fun spassLimit () = !spass_time;
   136 
   137 val hol_full_types = ResHolClause.full_types;
   138 val hol_partial_types = ResHolClause.partial_types;
   139 val hol_const_types_only = ResHolClause.const_types_only;
   140 val hol_no_types = ResHolClause.no_types;
   141 fun hol_typ_level () = ResHolClause.find_typ_level ();
   142 fun is_typed_hol () =
   143     let val tp_level = hol_typ_level()
   144     in
   145         not (tp_level = ResHolClause.T_NONE)
   146     end;
   147 
   148 fun atp_input_file () =
   149     let val file = !problem_name
   150     in
   151         if !destdir = "" then File.platform_path (File.tmp_path (Path.basic file))
   152         else if File.exists (File.explode_platform_path (!destdir))
   153         then !destdir ^ "/" ^ file
   154         else error ("No such directory: " ^ !destdir)
   155     end;
   156 
   157 val include_all = ref true;
   158 val include_simpset = ref false;
   159 val include_claset = ref false;
   160 val include_atpset = ref true;
   161 
   162 (*Tests show that follow_defs gives VERY poor results with "include_all"*)
   163 fun add_all() = (include_all:=true; ReduceAxiomsN.follow_defs := false);
   164 fun rm_all() = include_all:=false;
   165 
   166 fun add_simpset() = include_simpset:=true;
   167 fun rm_simpset() = include_simpset:=false;
   168 
   169 fun add_claset() = include_claset:=true;
   170 fun rm_claset() = include_claset:=false;
   171 
   172 fun add_clasimp() = (include_simpset:=true;include_claset:=true);
   173 fun rm_clasimp() = (include_simpset:=false;include_claset:=false);
   174 
   175 fun add_atpset() = include_atpset:=true;
   176 fun rm_atpset() = include_atpset:=false;
   177 
   178 
   179 (**** relevance filter ****)
   180 val run_relevance_filter = ReduceAxiomsN.run_relevance_filter;
   181 val run_blacklist_filter = ref true;
   182 
   183 (******************************************************************)
   184 (* detect whether a given problem (clauses) is FOL/HOL/HOLC/HOLCS *)
   185 (******************************************************************)
   186 
   187 datatype logic = FOL | HOL | HOLC | HOLCS;
   188 
   189 fun string_of_logic FOL = "FOL"
   190   | string_of_logic HOL = "HOL"
   191   | string_of_logic HOLC = "HOLC"
   192   | string_of_logic HOLCS = "HOLCS";
   193 
   194 fun is_fol_logic FOL = true
   195   | is_fol_logic  _ = false
   196 
   197 (*HOLCS will not occur here*)
   198 fun upgrade_lg HOLC _ = HOLC
   199   | upgrade_lg HOL HOLC = HOLC
   200   | upgrade_lg HOL _ = HOL
   201   | upgrade_lg FOL lg = lg;
   202 
   203 (* check types *)
   204 fun has_bool_hfn (Type("bool",_)) = true
   205   | has_bool_hfn (Type("fun",_)) = true
   206   | has_bool_hfn (Type(_, Ts)) = exists has_bool_hfn Ts
   207   | has_bool_hfn _ = false;
   208 
   209 fun is_hol_fn tp =
   210     let val (targs,tr) = strip_type tp
   211     in
   212         exists (has_bool_hfn) (tr::targs)
   213     end;
   214 
   215 fun is_hol_pred tp =
   216     let val (targs,tr) = strip_type tp
   217     in
   218         exists (has_bool_hfn) targs
   219     end;
   220 
   221 exception FN_LG of term;
   222 
   223 fun fn_lg (t as Const(f,tp)) (lg,seen) =
   224     if is_hol_fn tp then (upgrade_lg HOL lg, insert (op =) t seen) else (lg, insert (op =) t seen)
   225   | fn_lg (t as Free(f,tp)) (lg,seen) =
   226     if is_hol_fn tp then (upgrade_lg HOL lg, insert (op =) t seen) else (lg, insert (op =) t seen)
   227   | fn_lg (t as Var(f,tp)) (lg,seen) =
   228     if is_hol_fn tp then (upgrade_lg HOL lg,insert (op =) t seen) else (lg,insert (op =) t seen)
   229   | fn_lg (t as Abs(_,_,_)) (lg,seen) = (upgrade_lg HOLC lg,insert (op =) t seen)
   230   | fn_lg f _ = raise FN_LG(f);
   231 
   232 
   233 fun term_lg [] (lg,seen) = (lg,seen)
   234   | term_lg (tm::tms) (FOL,seen) =
   235       let val (f,args) = strip_comb tm
   236           val (lg',seen') = if f mem seen then (FOL,seen)
   237                             else fn_lg f (FOL,seen)
   238       in
   239         if is_fol_logic lg' then ()
   240         else Output.debug (fn () => ("Found a HOL term: " ^ Display.raw_string_of_term f));
   241         term_lg (args@tms) (lg',seen')
   242       end
   243   | term_lg _ (lg,seen) = (lg,seen)
   244 
   245 exception PRED_LG of term;
   246 
   247 fun pred_lg (t as Const(P,tp)) (lg,seen)=
   248       if is_hol_pred tp then (upgrade_lg HOL lg, insert (op =) t seen)
   249       else (lg,insert (op =) t seen)
   250   | pred_lg (t as Free(P,tp)) (lg,seen) =
   251       if is_hol_pred tp then (upgrade_lg HOL lg, insert (op =) t seen)
   252       else (lg,insert (op =) t seen)
   253   | pred_lg (t as Var(_,_)) (lg,seen) = (upgrade_lg HOL lg, insert (op =) t seen)
   254   | pred_lg P _ = raise PRED_LG(P);
   255 
   256 
   257 fun lit_lg (Const("Not",_) $ P) (lg,seen) = lit_lg P (lg,seen)
   258   | lit_lg P (lg,seen) =
   259       let val (pred,args) = strip_comb P
   260           val (lg',seen') = if pred mem seen then (lg,seen)
   261                             else pred_lg pred (lg,seen)
   262       in
   263         if is_fol_logic lg' then ()
   264         else Output.debug (fn () => ("Found a HOL predicate: " ^ Display.raw_string_of_term pred));
   265         term_lg args (lg',seen')
   266       end;
   267 
   268 fun lits_lg [] (lg,seen) = (lg,seen)
   269   | lits_lg (lit::lits) (FOL,seen) =
   270       let val (lg,seen') = lit_lg lit (FOL,seen)
   271       in
   272         if is_fol_logic lg then ()
   273         else Output.debug (fn () => ("Found a HOL literal: " ^ Display.raw_string_of_term lit));
   274         lits_lg lits (lg,seen')
   275       end
   276   | lits_lg lits (lg,seen) = (lg,seen);
   277 
   278 fun dest_disj_aux (Const("Trueprop",_) $ t) disjs = dest_disj_aux t disjs
   279   | dest_disj_aux (Const ("op |", _) $ t $ t') disjs = dest_disj_aux t (dest_disj_aux t' disjs)
   280   | dest_disj_aux t disjs = t::disjs;
   281 
   282 fun dest_disj t = dest_disj_aux t [];
   283 
   284 fun logic_of_clause tm = lits_lg (dest_disj tm);
   285 
   286 fun logic_of_clauses [] (lg,seen) = (lg,seen)
   287   | logic_of_clauses (cls::clss) (FOL,seen) =
   288     let val (lg,seen') = logic_of_clause cls (FOL,seen)
   289         val _ =
   290           if is_fol_logic lg then ()
   291           else Output.debug (fn () => ("Found a HOL clause: " ^ Display.raw_string_of_term cls))
   292     in
   293         logic_of_clauses clss (lg,seen')
   294     end
   295   | logic_of_clauses (cls::clss) (lg,seen) = (lg,seen);
   296 
   297 fun problem_logic_goals_aux [] (lg,seen) = lg
   298   | problem_logic_goals_aux (subgoal::subgoals) (lg,seen) =
   299     problem_logic_goals_aux subgoals (logic_of_clauses subgoal (lg,seen));
   300 
   301 fun problem_logic_goals subgoals = problem_logic_goals_aux subgoals (FOL,[]);
   302 
   303 fun is_fol_thms ths = ((fst(logic_of_clauses (map prop_of ths) (FOL,[]))) = FOL);
   304 
   305 (***************************************************************)
   306 (* Retrieving and filtering lemmas                             *)
   307 (***************************************************************)
   308 
   309 (*** white list and black list of lemmas ***)
   310 
   311 (*The rule subsetI is frequently omitted by the relevance filter.*)
   312 val whitelist = ref [subsetI];
   313 
   314 (*Names of theorems (not theorem lists! See multi_blacklist below) to be banned.
   315 
   316   These theorems typically produce clauses that are prolific (match too many equality or
   317   membership literals) and relate to seldom-used facts. Some duplicate other rules.
   318   FIXME: this blacklist needs to be maintained using theory data and added to using
   319   an attribute.*)
   320 val blacklist = ref
   321   ["Datatype.prod.size",
   322    "Divides.dvd_0_left_iff",
   323    "Finite_Set.card_0_eq",
   324    "Finite_Set.card_infinite",
   325    "Finite_Set.Max_ge",
   326    "Finite_Set.Max_in",
   327    "Finite_Set.Max_le_iff",
   328    "Finite_Set.Max_less_iff",
   329    "Finite_Set.max.f_below_strict_below.below_f_conv", (*duplicates in Orderings.*)
   330    "Finite_Set.max.f_below_strict_below.strict_below_f_conv", (*duplicates in Orderings.*)
   331    "Finite_Set.Min_ge_iff",
   332    "Finite_Set.Min_gr_iff",
   333    "Finite_Set.Min_in",
   334    "Finite_Set.Min_le",
   335    "Finite_Set.min.f_below_strict_below.below_f_conv",        (*duplicates in Orderings.*)
   336    "Finite_Set.min.f_below_strict_below.strict_below_f_conv", (*duplicates in Orderings.*)
   337    "Fun.vimage_image_eq",   (*involves an existential quantifier*)
   338    "HOL.split_if_asm",     (*splitting theorem*)
   339    "HOL.split_if",         (*splitting theorem*)
   340    "HOL.All_def",          (*far worse than useless!!*)
   341    "IntDef.abs_split",
   342    "IntDef.Integ.Abs_Integ_inject",
   343    "IntDef.Integ.Abs_Integ_inverse",
   344    "IntDiv.zdvd_0_left",
   345    "List.append_eq_append_conv",
   346    "List.hd_Cons_tl",   (*Says everything is [] or Cons. Probably prolific.*)
   347    "List.in_listsD",
   348    "List.in_listsI",
   349    "List.lists.Cons",
   350    "List.listsE",
   351    "Nat.less_one", (*not directional? obscure*)
   352    "Nat.not_gr0",
   353    "Nat.one_eq_mult_iff", (*duplicate by symmetry*)
   354    "Nat.of_nat_0_eq_iff",
   355    "Nat.of_nat_eq_0_iff",
   356    "Nat.of_nat_le_0_iff",
   357    "NatSimprocs.divide_le_0_iff_number_of",  (*too many clauses*)
   358    "NatSimprocs.divide_less_0_iff_number_of",
   359    "NatSimprocs.equation_minus_iff_1",  (*not directional*)
   360    "NatSimprocs.equation_minus_iff_number_of", (*not directional*)
   361    "NatSimprocs.le_minus_iff_1", (*not directional*)
   362    "NatSimprocs.le_minus_iff_number_of",  (*not directional*)
   363    "NatSimprocs.less_minus_iff_1", (*not directional*)
   364    "NatSimprocs.less_minus_iff_number_of", (*not directional*)
   365    "NatSimprocs.minus_equation_iff_number_of", (*not directional*)
   366    "NatSimprocs.minus_le_iff_1", (*not directional*)
   367    "NatSimprocs.minus_le_iff_number_of", (*not directional*)
   368    "NatSimprocs.minus_less_iff_1", (*not directional*)
   369    "NatSimprocs.mult_le_cancel_left_number_of", (*excessive case analysis*)
   370    "NatSimprocs.mult_le_cancel_right_number_of", (*excessive case analysis*)
   371    "NatSimprocs.mult_less_cancel_left_number_of", (*excessive case analysis*)
   372    "NatSimprocs.mult_less_cancel_right_number_of", (*excessive case analysis*)
   373    "NatSimprocs.zero_le_divide_iff_number_of", (*excessive case analysis*)
   374    "NatSimprocs.zero_less_divide_iff_number_of",
   375    "OrderedGroup.abs_0_eq", (*duplicate by symmetry*)
   376    "OrderedGroup.diff_eq_0_iff_eq", (*prolific?*)
   377    "OrderedGroup.join_0_eq_0",
   378    "OrderedGroup.meet_0_eq_0",
   379    "OrderedGroup.pprt_eq_0",   (*obscure*)
   380    "OrderedGroup.pprt_eq_id",   (*obscure*)
   381    "OrderedGroup.pprt_mono",   (*obscure*)
   382    "Orderings.split_max",      (*splitting theorem*)
   383    "Orderings.split_min",      (*splitting theorem*)
   384    "Power.zero_less_power_abs_iff",
   385    "Product_Type.split_eta_SetCompr",   (*involves an existential quantifier*)
   386    "Product_Type.split_paired_Ball_Sigma",     (*splitting theorem*)
   387    "Product_Type.split_paired_Bex_Sigma",      (*splitting theorem*)
   388    "Product_Type.split_split_asm",             (*splitting theorem*)
   389    "Product_Type.split_split",                 (*splitting theorem*)
   390    "Product_Type.unit_abs_eta_conv",
   391    "Product_Type.unit_induct",
   392    "Relation.diagI",
   393    "Relation.Domain_def",   (*involves an existential quantifier*)
   394    "Relation.Image_def",   (*involves an existential quantifier*)
   395    "Relation.ImageI",
   396    "Ring_and_Field.divide_cancel_left", (*fields are seldom used & often prolific*)
   397    "Ring_and_Field.divide_cancel_right",
   398    "Ring_and_Field.divide_divide_eq_left",
   399    "Ring_and_Field.divide_divide_eq_right",
   400    "Ring_and_Field.divide_eq_0_iff",
   401    "Ring_and_Field.divide_eq_1_iff",
   402    "Ring_and_Field.divide_eq_eq_1",
   403    "Ring_and_Field.divide_le_0_1_iff",
   404    "Ring_and_Field.divide_le_eq_1_neg",  (*obscure and prolific*)
   405    "Ring_and_Field.divide_le_eq_1_pos",  (*obscure and prolific*)
   406    "Ring_and_Field.divide_less_0_1_iff",
   407    "Ring_and_Field.divide_less_eq_1_neg",  (*obscure and prolific*)
   408    "Ring_and_Field.divide_less_eq_1_pos",  (*obscure and prolific*)
   409    "Ring_and_Field.eq_divide_eq_1", (*duplicate by symmetry*)
   410    "Ring_and_Field.field_mult_cancel_left",
   411    "Ring_and_Field.field_mult_cancel_right",
   412    "Ring_and_Field.inverse_le_iff_le_neg",
   413    "Ring_and_Field.inverse_le_iff_le",
   414    "Ring_and_Field.inverse_less_iff_less_neg",
   415    "Ring_and_Field.inverse_less_iff_less",
   416    "Ring_and_Field.le_divide_eq_1_neg", (*obscure and prolific*)
   417    "Ring_and_Field.le_divide_eq_1_pos", (*obscure and prolific*)
   418    "Ring_and_Field.less_divide_eq_1_neg", (*obscure and prolific*)
   419    "Ring_and_Field.less_divide_eq_1_pos", (*obscure and prolific*)
   420    "Ring_and_Field.one_eq_divide_iff",  (*duplicate by symmetry*)
   421    "Set.Collect_bex_eq",   (*involves an existential quantifier*)
   422    "Set.Collect_ex_eq",   (*involves an existential quantifier*)
   423    "Set.Diff_eq_empty_iff", (*redundant with paramodulation*)
   424    "Set.Diff_insert0",
   425    "Set.empty_Union_conv",   (*redundant with paramodulation*)
   426    "Set.full_SetCompr_eq",   (*involves an existential quantifier*)
   427    "Set.image_Collect",      (*involves Collect and a boolean variable...*)
   428    "Set.image_def",          (*involves an existential quantifier*)
   429    "Set.Int_UNIV",  (*redundant with paramodulation*)
   430    "Set.Inter_iff", (*We already have InterI, InterE*)
   431    "Set.psubsetE",    (*too prolific and obscure*)
   432    "Set.psubsetI",
   433    "Set.singleton_insert_inj_eq'",
   434    "Set.singleton_insert_inj_eq",
   435    "Set.singletonD",  (*these two duplicate some "insert" lemmas*)
   436    "Set.singletonI",
   437    "Set.Un_empty", (*redundant with paramodulation*)
   438    "Set.UNION_def",   (*involves an existential quantifier*)
   439    "Set.Union_empty_conv", (*redundant with paramodulation*)
   440    "Set.Union_iff",              (*We already have UnionI, UnionE*)
   441    "SetInterval.atLeastAtMost_iff", (*obscure and prolific*)
   442    "SetInterval.atLeastLessThan_iff", (*obscure and prolific*)
   443    "SetInterval.greaterThanAtMost_iff", (*obscure and prolific*)
   444    "SetInterval.greaterThanLessThan_iff", (*obscure and prolific*)
   445    "SetInterval.ivl_subset"];  (*excessive case analysis*)
   446 
   447 
   448 (*These might be prolific but are probably OK, and min and max are basic.
   449    "Orderings.max_less_iff_conj",
   450    "Orderings.min_less_iff_conj",
   451    "Orderings.min_max.below_inf.below_inf_conv",
   452    "Orderings.min_max.below_sup.above_sup_conv",
   453 Very prolific and somewhat obscure:
   454    "Set.InterD",
   455    "Set.UnionI",
   456 *)
   457 
   458 (*** retrieve lemmas from clasimpset and atpset, may filter them ***)
   459 
   460 (*Hashing to detect duplicate and variant clauses, e.g. from the [iff] attribute*)
   461 
   462 exception HASH_CLAUSE and HASH_STRING;
   463 
   464 (*Reject theorems with names like "List.filter.filter_list_def" or
   465   "Accessible_Part.acc.defs", as these are definitions arising from packages.*)
   466 fun is_package_def a =
   467   let val names = NameSpace.explode a
   468   in
   469      length names > 2 andalso
   470      not (hd names = "local") andalso
   471      String.isSuffix "_def" a  orelse  String.isSuffix "_defs" a
   472   end;
   473 
   474 fun make_banned_test xs =
   475   let val ht = Polyhash.mkTable (Polyhash.hash_string, op =) (6000, HASH_STRING)
   476       fun banned s = isSome (Polyhash.peek ht s) orelse is_package_def s
   477   in  app (fn x => Polyhash.insert ht (x,())) (!blacklist);
   478       banned
   479   end;
   480 
   481 (** a hash function from Term.term to int, and also a hash table **)
   482 val xor_words = List.foldl Word.xorb 0w0;
   483 
   484 fun hashw_term ((Const(c,_)), w) = Polyhash.hashw_string (c,w)
   485   | hashw_term ((Free(a,_)), w) = Polyhash.hashw_string (a,w)
   486   | hashw_term ((Var(_,_)), w) = w
   487   | hashw_term ((Bound i), w) = Polyhash.hashw_int(i,w)
   488   | hashw_term ((Abs(_,_,t)), w) = hashw_term (t, w)
   489   | hashw_term ((P$Q), w) = hashw_term (Q, (hashw_term (P, w)));
   490 
   491 fun hash_literal (Const("Not",_)$P) = Word.notb(hashw_term(P,0w0))
   492   | hash_literal P = hashw_term(P,0w0);
   493 
   494 fun hash_term t = Word.toIntX (xor_words (map hash_literal (dest_disj t)));
   495 
   496 fun equal_thm (thm1,thm2) = Term.aconv(prop_of thm1, prop_of thm2);
   497 
   498 (*Create a hash table for clauses, of the given size*)
   499 fun mk_clause_table n =
   500       Polyhash.mkTable (hash_term o prop_of, equal_thm)
   501                        (n, HASH_CLAUSE);
   502 
   503 (*Use a hash table to eliminate duplicates from xs. Argument is a list of
   504   (thm * (string * int)) tuples. The theorems are hashed into the table. *)
   505 fun make_unique xs =
   506   let val ht = mk_clause_table 7000
   507   in
   508       Output.debug (fn () => ("make_unique gets " ^ Int.toString (length xs) ^ " clauses"));
   509       app (ignore o Polyhash.peekInsert ht) xs;
   510       Polyhash.listItems ht
   511   end;
   512 
   513 (*Remove existing axiom clauses from the conjecture clauses, as this can dramatically
   514   boost an ATP's performance (for some reason)*)
   515 fun subtract_cls c_clauses ax_clauses =
   516   let val ht = mk_clause_table 2200
   517       fun known x = isSome (Polyhash.peek ht x)
   518   in
   519       app (ignore o Polyhash.peekInsert ht) ax_clauses;
   520       filter (not o known) c_clauses
   521   end;
   522 
   523 (*Filter axiom clauses, but keep supplied clauses and clauses in whitelist.
   524   Duplicates are removed later.*)
   525 fun get_relevant_clauses thy cls_thms white_cls goals =
   526   white_cls @ (ReduceAxiomsN.relevance_filter thy cls_thms goals);
   527 
   528 fun all_valid_thms ctxt =
   529   PureThy.thms_containing (ProofContext.theory_of ctxt) ([], []) @
   530   filter (ProofContext.valid_thms ctxt)
   531     (FactIndex.find (ProofContext.fact_index_of ctxt) ([], []));
   532 
   533 fun multi_name a (th, (n,pairs)) =
   534   (n+1, (a ^ "(" ^ Int.toString n ^ ")", th) :: pairs)
   535 
   536 fun add_multi_names_aux ((a, []), pairs) = pairs
   537   | add_multi_names_aux ((a, [th]), pairs) = (a,th)::pairs
   538   | add_multi_names_aux ((a, ths), pairs) = #2 (foldl (multi_name a) (1,pairs) ths);
   539 
   540 val multi_blacklist =
   541   ["Set.ball_simps", "Set.bex_simps",  (*quantifier rewriting: useless*)
   542    "Set.disjoint_insert", "Set.insert_disjoint", "Set.Inter_UNIV_conv"];
   543 
   544 val multi_base_blacklist =
   545   ["defs","select_defs","update_defs","induct","inducts","split","splits","split_asm"];
   546 
   547 (*Ignore blacklisted theorem lists*)
   548 fun add_multi_names ((a, ths), pairs) =
   549   if a mem_string multi_blacklist orelse (Sign.base_name a) mem_string multi_base_blacklist
   550   then pairs
   551   else add_multi_names_aux ((a, ths), pairs);
   552 
   553 fun is_multi (a, ths) = length ths > 1 orelse String.isSuffix ".axioms" a;
   554 
   555 (*The single theorems go BEFORE the multiple ones*)
   556 fun name_thm_pairs ctxt =
   557   let val (mults,singles) = List.partition is_multi (all_valid_thms ctxt)
   558   in  foldl add_multi_names (foldl add_multi_names [] mults) singles  end;
   559 
   560 fun check_named ("",th) = (warning ("No name for theorem " ^ string_of_thm th); false)
   561   | check_named (_,th) = true;
   562 
   563 fun display_thm (name,th) = Output.debug (fn () => name ^ ": " ^ string_of_thm th);
   564 
   565 (* get lemmas from claset, simpset, atpset and extra supplied rules *)
   566 fun get_clasimp_atp_lemmas ctxt user_thms =
   567   let val included_thms =
   568         if !include_all
   569         then (tap (fn ths => Output.debug
   570                      (fn () => ("Including all " ^ Int.toString (length ths) ^ " theorems")))
   571                   (name_thm_pairs ctxt))
   572         else
   573         let val claset_thms =
   574                 if !include_claset then ResAxioms.claset_rules_of ctxt
   575                 else []
   576             val simpset_thms =
   577                 if !include_simpset then ResAxioms.simpset_rules_of ctxt
   578                 else []
   579             val atpset_thms =
   580                 if !include_atpset then ResAxioms.atpset_rules_of ctxt
   581                 else []
   582             val _ = (Output.debug (fn () => "ATP theorems: ");  app display_thm atpset_thms)
   583         in  claset_thms @ simpset_thms @ atpset_thms  end
   584       val user_rules = filter check_named
   585                          (map (ResAxioms.pairname)
   586                            (if null user_thms then !whitelist else user_thms))
   587   in
   588       (filter check_named included_thms, user_rules)
   589   end;
   590 
   591 (*Remove lemmas that are banned from the backlist. Also remove duplicates. *)
   592 fun blacklist_filter ths =
   593   if !run_blacklist_filter then
   594       let val banned = make_banned_test (map #1 ths)
   595           fun ok (a,_) = not (banned a)
   596           val (good,bad) = List.partition ok ths
   597       in
   598         Output.debug (fn () => "blacklist filter gets " ^ Int.toString (length ths) ^ " theorems");
   599         Output.debug (fn () => "filtered: " ^ space_implode ", " (map #1 bad));
   600         Output.debug (fn () => "...and returns " ^ Int.toString (length good));
   601         good 
   602       end
   603   else ths;
   604 
   605 (***************************************************************)
   606 (* Type Classes Present in the Axiom or Conjecture Clauses     *)
   607 (***************************************************************)
   608 
   609 fun setinsert (x,s) = Symtab.update (x,()) s;
   610 
   611 fun add_classes (sorts, cset) = foldl setinsert cset (List.concat sorts);
   612 
   613 (*Remove this trivial type class*)
   614 fun delete_type cset = Symtab.delete_safe "HOL.type" cset;
   615 
   616 fun tvar_classes_of_terms ts =
   617   let val sorts_list = map (map #2 o term_tvars) ts
   618   in  Symtab.keys (delete_type (foldl add_classes Symtab.empty sorts_list))  end;
   619 
   620 fun tfree_classes_of_terms ts =
   621   let val sorts_list = map (map #2 o term_tfrees) ts
   622   in  Symtab.keys (delete_type (foldl add_classes Symtab.empty sorts_list))  end;
   623 
   624 (*fold type constructors*)
   625 fun fold_type_consts f (Type (a, Ts)) x = fold (fold_type_consts f) Ts (f (a,x))
   626   | fold_type_consts f T x = x;
   627 
   628 val add_type_consts_in_type = fold_type_consts setinsert;
   629 
   630 (*Type constructors used to instantiate overloaded constants are the only ones needed.*)
   631 fun add_type_consts_in_term thy =
   632   let val const_typargs = Sign.const_typargs thy
   633       fun add_tcs (Const cT) x = fold add_type_consts_in_type (const_typargs cT) x
   634         | add_tcs (Abs (_, T, u)) x = add_tcs u x
   635         | add_tcs (t $ u) x = add_tcs t (add_tcs u x)
   636         | add_tcs _ x = x
   637   in  add_tcs  end
   638 
   639 fun type_consts_of_terms thy ts =
   640   Symtab.keys (fold (add_type_consts_in_term thy) ts Symtab.empty);
   641 
   642 
   643 (***************************************************************)
   644 (* ATP invocation methods setup                                *)
   645 (***************************************************************)
   646 
   647 fun cnf_hyps_thms ctxt =
   648     let val ths = Assumption.prems_of ctxt
   649     in fold (fold (insert Thm.eq_thm) o ResAxioms.skolem_thm) ths [] end;
   650 
   651 (*Translation mode can be auto-detected, or forced to be first-order or higher-order*)
   652 datatype mode = Auto | Fol | Hol;
   653 
   654 val linkup_logic_mode = ref Auto;
   655 
   656 (*Ensures that no higher-order theorems "leak out"*)
   657 fun restrict_to_logic logic cls =
   658   if is_fol_logic logic then filter (Meson.is_fol_term o prop_of o fst) cls
   659                         else cls;
   660 
   661 (**** Predicates to detect unwanted clauses (prolific or likely to cause unsoundness) ****)
   662 
   663 (** Too general means, positive equality literal with a variable X as one operand,
   664   when X does not occur properly in the other operand. This rules out clearly
   665   inconsistent clauses such as V=a|V=b, though it by no means guarantees soundness. **)
   666 
   667 fun occurs ix =
   668     let fun occ(Var (jx,_)) = (ix=jx)
   669           | occ(t1$t2)      = occ t1 orelse occ t2
   670           | occ(Abs(_,_,t)) = occ t
   671           | occ _           = false
   672     in occ end;
   673 
   674 fun is_recordtype T = not (null (RecordPackage.dest_recTs T));
   675 
   676 (*Unwanted equalities include
   677   (1) those between a variable that does not properly occur in the second operand,
   678   (2) those between a variable and a record, since these seem to be prolific "cases" thms
   679 *)
   680 fun too_general_eqterms (Var (ix,T), t) = not (occurs ix t) orelse is_recordtype T
   681   | too_general_eqterms _ = false;
   682 
   683 fun too_general_equality (Const ("op =", _) $ x $ y) =
   684       too_general_eqterms (x,y) orelse too_general_eqterms(y,x)
   685   | too_general_equality _ = false;
   686 
   687 (* tautologous? *)
   688 fun is_taut (Const ("Trueprop", _) $ Const ("True", _)) = true
   689   | is_taut _ = false;
   690 
   691 (*True if the term contains a variable whose (atomic) type is in the given list.*)
   692 fun has_typed_var tycons =
   693   let fun var_tycon (Var (_, Type(a,_))) = a mem_string tycons
   694         | var_tycon _ = false
   695   in  exists var_tycon o term_vars  end;
   696 
   697 (*Clauses are forbidden to contain variables of these types. The typical reason is that
   698   they lead to unsoundness. Note that "unit" satisfies numerous equations like ?X=().
   699   The resulting clause will have no type constraint, yielding false proofs. Even "bool"
   700   leads to many unsound proofs, though (obviously) only for higher-order problems.*)
   701 val unwanted_types = ref ["Product_Type.unit","bool"];
   702 
   703 fun unwanted t =
   704     is_taut t orelse has_typed_var (!unwanted_types) t orelse
   705     forall too_general_equality (dest_disj t);
   706 
   707 (*Clauses containing variables of type "unit" or "bool" are unlikely to be useful and
   708   likely to lead to unsound proofs.*)
   709 fun remove_unwanted_clauses cls = filter (not o unwanted o prop_of o fst) cls;
   710 
   711 fun tptp_writer logic goals filename (axioms,classrels,arities) user_lemmas =
   712     if is_fol_logic logic
   713     then ResClause.tptp_write_file goals filename (axioms, classrels, arities)
   714     else ResHolClause.tptp_write_file goals filename (axioms, classrels, arities) user_lemmas;
   715 
   716 fun dfg_writer logic goals filename (axioms,classrels,arities) user_lemmas =
   717     if is_fol_logic logic
   718     then ResClause.dfg_write_file goals filename (axioms, classrels, arities)
   719     else ResHolClause.dfg_write_file goals filename (axioms, classrels, arities) user_lemmas;
   720 
   721 (*Called by the oracle-based methods declared in res_atp_methods.ML*)
   722 fun write_subgoal_file dfg mode ctxt conjectures user_thms n =
   723     let val conj_cls = make_clauses conjectures
   724                          |> ResAxioms.assume_abstract_list |> Meson.finish_cnf
   725         val hyp_cls = cnf_hyps_thms ctxt
   726         val goal_cls = conj_cls@hyp_cls
   727         val goal_tms = map prop_of goal_cls
   728         val logic = case mode of
   729                             Auto => problem_logic_goals [goal_tms]
   730                           | Fol => FOL
   731                           | Hol => HOL
   732         val (included_thms,user_rules) = get_clasimp_atp_lemmas ctxt user_thms
   733         val cla_simp_atp_clauses = included_thms |> blacklist_filter
   734                                      |> ResAxioms.cnf_rules_pairs |> make_unique
   735                                      |> restrict_to_logic logic
   736                                      |> remove_unwanted_clauses
   737         val user_cls = ResAxioms.cnf_rules_pairs user_rules
   738         val thy = ProofContext.theory_of ctxt
   739         val axclauses = make_unique (get_relevant_clauses thy cla_simp_atp_clauses user_cls goal_tms)
   740         val subs = tfree_classes_of_terms goal_tms
   741         and axtms = map (prop_of o #1) axclauses
   742         val supers = tvar_classes_of_terms axtms
   743         and tycons = type_consts_of_terms thy (goal_tms@axtms)
   744         (*TFrees in conjecture clauses; TVars in axiom clauses*)
   745         val classrel_clauses = ResClause.make_classrel_clauses thy subs supers
   746         val arity_clauses = ResClause.arity_clause_thy thy tycons supers
   747         val writer = if dfg then dfg_writer else tptp_writer
   748         and file = atp_input_file()
   749         and user_lemmas_names = map #1 user_rules
   750     in
   751         writer logic goal_cls file (axclauses,classrel_clauses,arity_clauses) user_lemmas_names;
   752         Output.debug (fn () => "Writing to " ^ file);
   753         file
   754     end;
   755 
   756 
   757 (**** remove tmp files ****)
   758 fun cond_rm_tmp file =
   759     if !Output.debugging orelse !destdir <> ""
   760     then Output.debug (fn () => "ATP input kept...")
   761     else OS.FileSys.remove file;
   762 
   763 
   764 (****** setup ATPs as Isabelle methods ******)
   765 
   766 fun atp_meth tac ths ctxt =
   767     let val thy = ProofContext.theory_of ctxt
   768         val _ = ResClause.init thy
   769         val _ = ResHolClause.init thy
   770     in Method.SIMPLE_METHOD' (tac ctxt ths) end;
   771 
   772 fun atp_method tac = Method.thms_ctxt_args (atp_meth tac);
   773 
   774 (***************************************************************)
   775 (* automatic ATP invocation                                    *)
   776 (***************************************************************)
   777 
   778 (* call prover with settings and problem file for the current subgoal *)
   779 fun watcher_call_provers sign sg_terms (childin, childout, pid) =
   780   let
   781     fun make_atp_list [] n = []
   782       | make_atp_list (sg_term::xs) n =
   783           let
   784             val probfile = prob_pathname n
   785             val time = Int.toString (!time_limit)
   786           in
   787             Output.debug (fn () => "problem file in watcher_call_provers is " ^ probfile);
   788             (*options are separated by Watcher.setting_sep, currently #"%"*)
   789             if !prover = "spass"
   790             then
   791               let val spass = helper_path "SPASS_HOME" "SPASS"
   792                   val sopts =
   793    "-Auto%-SOS=1%-PGiven=0%-PProblem=0%-Splits=0%-FullRed=0%-DocProof%-TimeLimit=" ^ time
   794               in
   795                   ("spass", spass, sopts, probfile) :: make_atp_list xs (n+1)
   796               end
   797             else if !prover = "vampire"
   798             then
   799               let val vampire = helper_path "VAMPIRE_HOME" "vampire"
   800                   val vopts = "--mode casc%-t " ^ time  (*what about -m 100000?*)
   801               in
   802                   ("vampire", vampire, vopts, probfile) :: make_atp_list xs (n+1)
   803               end
   804              else if !prover = "E"
   805              then
   806                let val Eprover = helper_path "E_HOME" "eproof"
   807                in
   808                   ("E", Eprover,
   809                      "--tstp-in%--tstp-out%-l5%-xAutoDev%-tAutoDev%--silent%--cpu-limit=" ^ time, probfile) ::
   810                    make_atp_list xs (n+1)
   811                end
   812              else error ("Invalid prover name: " ^ !prover)
   813           end
   814 
   815     val atp_list = make_atp_list sg_terms 1
   816   in
   817     Watcher.callResProvers(childout,atp_list);
   818     Output.debug (fn () => "Sent commands to watcher!")
   819   end
   820 
   821 (*For debugging the generated set of theorem names*)
   822 fun trace_vector fname =
   823   let val path = File.explode_platform_path (fname ^ "_thm_names")
   824   in  Vector.app (File.append path o (fn s => s ^ "\n"))  end;
   825 
   826 (*We write out problem files for each subgoal. Argument probfile generates filenames,
   827   and allows the suppression of the suffix "_1" in problem-generation mode.
   828   FIXME: does not cope with &&, and it isn't easy because one could have multiple
   829   subgoals, each involving &&.*)
   830 fun write_problem_files probfile (ctxt,th)  =
   831   let val goals = Thm.prems_of th
   832       val _ = Output.debug (fn () => "number of subgoals = " ^ Int.toString (length goals))
   833       val thy = ProofContext.theory_of ctxt
   834       fun get_neg_subgoals [] _ = []
   835         | get_neg_subgoals (gl::gls) n = #1 (ResAxioms.neg_conjecture_clauses th n) :: 
   836                                          get_neg_subgoals gls (n+1)
   837       val goal_cls = get_neg_subgoals goals 1
   838       val logic = case !linkup_logic_mode of
   839                 Auto => problem_logic_goals (map ((map prop_of)) goal_cls)
   840               | Fol => FOL
   841               | Hol => HOL
   842       val (included_thms,white_thms) = get_clasimp_atp_lemmas ctxt []
   843       val included_cls = included_thms |> blacklist_filter
   844                                        |> ResAxioms.cnf_rules_pairs |> make_unique
   845                                        |> restrict_to_logic logic
   846                                        |> remove_unwanted_clauses
   847       val _ = Output.debug (fn () => "included clauses = " ^ Int.toString(length included_cls))
   848       val white_cls = ResAxioms.cnf_rules_pairs white_thms
   849       (*clauses relevant to goal gl*)
   850       val axcls_list = map (fn ngcls => get_relevant_clauses thy included_cls white_cls (map prop_of ngcls)) goal_cls
   851       val _ = app (fn axcls => Output.debug (fn () => "filtered clauses = " ^ Int.toString(length axcls)))
   852                   axcls_list
   853       val writer = if !prover = "spass" then dfg_writer else tptp_writer
   854       fun write_all [] [] _ = []
   855         | write_all (ccls::ccls_list) (axcls::axcls_list) k =
   856             let val fname = probfile k
   857                 val _ = Output.debug (fn () => "About to write file " ^ fname)
   858                 val axcls = make_unique axcls
   859                 val _ = Output.debug (fn () => "Conjecture Clauses (before duplicate removal)")
   860                 val _ = app (fn th => Output.debug (fn _ => string_of_thm th)) ccls
   861                 val ccls = subtract_cls ccls axcls
   862                 val _ = Output.debug (fn () => "Conjecture Clauses (AFTER duplicate removal)")
   863                 val _ = app (fn th => Output.debug (fn _ => string_of_thm th)) ccls
   864                 val ccltms = map prop_of ccls
   865                 and axtms = map (prop_of o #1) axcls
   866                 val subs = tfree_classes_of_terms ccltms
   867                 and supers = tvar_classes_of_terms axtms
   868                 and tycons = type_consts_of_terms thy (ccltms@axtms)
   869                 (*TFrees in conjecture clauses; TVars in axiom clauses*)
   870                 val classrel_clauses = ResClause.make_classrel_clauses thy subs supers
   871                 val _ = Output.debug (fn () => "classrel clauses = " ^ Int.toString (length classrel_clauses))
   872                 val arity_clauses = ResClause.arity_clause_thy thy tycons supers
   873                 val _ = Output.debug (fn () => "arity clauses = " ^ Int.toString (length arity_clauses))
   874                 val clnames = writer logic ccls fname (axcls,classrel_clauses,arity_clauses) []
   875                 val thm_names = Vector.fromList clnames
   876                 val _ = if !Output.debugging then trace_vector fname thm_names else ()
   877             in  (thm_names,fname) :: write_all ccls_list axcls_list (k+1)  end
   878       val (thm_names_list, filenames) = ListPair.unzip (write_all goal_cls axcls_list 1)
   879   in
   880       (filenames, thm_names_list)
   881   end;
   882 
   883 val last_watcher_pid = ref (NONE : (TextIO.instream * TextIO.outstream *
   884                                     Posix.Process.pid * string list) option);
   885 
   886 fun kill_last_watcher () =
   887     (case !last_watcher_pid of
   888          NONE => ()
   889        | SOME (_, _, pid, files) =>
   890           (Output.debug (fn () => "Killing old watcher, pid = " ^ string_of_pid pid);
   891            Watcher.killWatcher pid;
   892            ignore (map (try cond_rm_tmp) files)))
   893      handle OS.SysErr _ => Output.debug (fn () => "Attempt to kill watcher failed");
   894 
   895 (*writes out the current problems and calls ATPs*)
   896 fun isar_atp (ctxt, th) =
   897   if Thm.no_prems th then ()
   898   else
   899     let
   900       val _ = kill_last_watcher()
   901       val (files,thm_names_list) = write_problem_files prob_pathname (ctxt,th)
   902       val (childin, childout, pid) = Watcher.createWatcher (ctxt, th, thm_names_list)
   903     in
   904       last_watcher_pid := SOME (childin, childout, pid, files);
   905       Output.debug (fn () => "problem files: " ^ space_implode ", " files);
   906       Output.debug (fn () => "pid: " ^ string_of_pid pid);
   907       watcher_call_provers (sign_of_thm th) (Thm.prems_of th) (childin, childout, pid)
   908     end;
   909 
   910 (*For ML scripts, and primarily, for debugging*)
   911 fun callatp () =
   912   let val th = topthm()
   913       val ctxt = ProofContext.init (theory_of_thm th)
   914   in  isar_atp (ctxt, th)  end;
   915 
   916 val isar_atp_writeonly = setmp print_mode []
   917       (fn (ctxt,th) =>
   918        if Thm.no_prems th then ()
   919        else
   920          let val probfile = if Thm.nprems_of th = 1 then probfile_nosuffix
   921                             else prob_pathname
   922          in ignore (write_problem_files probfile (ctxt,th)) end);
   923 
   924 
   925 (** the Isar toplevel hook **)
   926 
   927 fun invoke_atp_ml (ctxt, goal) =
   928   let val thy = ProofContext.theory_of ctxt;
   929   in
   930     Output.debug (fn () => "subgoals in isar_atp:\n" ^
   931                   Pretty.string_of (ProofContext.pretty_term ctxt
   932                     (Logic.mk_conjunction_list (Thm.prems_of goal))));
   933     Output.debug (fn () => "current theory: " ^ Context.theory_name thy);
   934     inc hook_count;
   935     Output.debug (fn () => "in hook for time: " ^ Int.toString (!hook_count));
   936     ResClause.init thy;
   937     ResHolClause.init thy;
   938     if !time_limit > 0 then isar_atp (ctxt, goal)
   939     else (warning ("Writing problem file only: " ^ !problem_name); 
   940           isar_atp_writeonly (ctxt, goal))
   941   end;
   942 
   943 val invoke_atp = Toplevel.no_timing o Toplevel.unknown_proof o Toplevel.keep
   944  (fn state =>
   945   let val (ctxt, (_, goal)) = Proof.get_goal (Toplevel.proof_of state)
   946   in  invoke_atp_ml (ctxt, goal)  end);
   947 
   948 val call_atpP =
   949   OuterSyntax.command
   950     "ProofGeneral.call_atp"
   951     "call automatic theorem provers"
   952     OuterKeyword.diag
   953     (Scan.succeed invoke_atp);
   954 
   955 val _ = OuterSyntax.add_parsers [call_atpP];
   956 
   957 end;