| author | wenzelm | 
| Wed, 30 Dec 2015 20:12:26 +0100 | |
| changeset 61992 | 6d02bb8b5fe1 | 
| parent 60940 | 4c108cce6b35 | 
| child 62916 | 621afc4607ec | 
| permissions | -rw-r--r-- | 
| 23178 | 1 | (* Title: Pure/search.ML | 
| 60940 | 2 | Author: Lawrence C Paulson | 
| 3 | Author: Norbert Voelker, FernUniversitaet Hagen | |
| 1588 | 4 | |
| 16179 | 5 | Search tacticals. | 
| 1588 | 6 | *) | 
| 7 | ||
| 2672 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 8 | infix 1 THEN_MAYBE THEN_MAYBE'; | 
| 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 9 | |
| 1588 | 10 | signature SEARCH = | 
| 32940 | 11 | sig | 
| 12 | val trace_DEPTH_FIRST: bool Unsynchronized.ref | |
| 13 | val DEPTH_FIRST: (thm -> bool) -> tactic -> tactic | |
| 14 | val has_fewer_prems: int -> thm -> bool | |
| 15 | val IF_UNSOLVED: tactic -> tactic | |
| 16 | val SOLVE: tactic -> tactic | |
| 17 | val THEN_MAYBE: tactic * tactic -> tactic | |
| 18 |   val THEN_MAYBE': ('a -> tactic) * ('a -> tactic) -> 'a -> tactic
 | |
| 19 | val DEPTH_SOLVE: tactic -> tactic | |
| 20 | val DEPTH_SOLVE_1: tactic -> tactic | |
| 38802 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 21 | val THEN_ITER_DEEPEN: int -> tactic -> (thm -> bool) -> (int -> tactic) -> tactic | 
| 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 22 | val ITER_DEEPEN: int -> (thm -> bool) -> (int -> tactic) -> tactic | 
| 33335 
1e189f96c393
DEEPEN: less aggressive tracing, subject to trace_DEEPEN;
 wenzelm parents: 
32940diff
changeset | 23 | val trace_DEEPEN: bool Unsynchronized.ref | 
| 32940 | 24 | val DEEPEN: int * int -> (int -> int -> tactic) -> int -> int -> tactic | 
| 25 | val trace_BEST_FIRST: bool Unsynchronized.ref | |
| 26 | val THEN_BEST_FIRST: tactic -> (thm -> bool) * (thm -> int) -> tactic -> tactic | |
| 27 | val BEST_FIRST: (thm -> bool) * (thm -> int) -> tactic -> tactic | |
| 28 | val BREADTH_FIRST: (thm -> bool) -> tactic -> tactic | |
| 29 | val QUIET_BREADTH_FIRST: (thm -> bool) -> tactic -> tactic | |
| 30 | val trace_ASTAR: bool Unsynchronized.ref | |
| 31 | val THEN_ASTAR: tactic -> (thm -> bool) * (int -> thm -> int) -> tactic -> tactic | |
| 32 | val ASTAR: (thm -> bool) * (int -> thm -> int) -> tactic -> tactic | |
| 33 | end; | |
| 1588 | 34 | |
| 32940 | 35 | structure Search: SEARCH = | 
| 1588 | 36 | struct | 
| 37 | ||
| 38 | (**** Depth-first search ****) | |
| 39 | ||
| 32738 | 40 | val trace_DEPTH_FIRST = Unsynchronized.ref false; | 
| 1588 | 41 | |
| 42 | (*Searches until "satp" reports proof tree as satisfied. | |
| 43 | Suppresses duplicate solutions to minimize search space.*) | |
| 23178 | 44 | fun DEPTH_FIRST satp tac = | 
| 60940 | 45 | let | 
| 46 | val tac = tracify trace_DEPTH_FIRST tac | |
| 47 | fun depth used [] = NONE | |
| 48 | | depth used (q :: qs) = | |
| 49 | (case Seq.pull q of | |
| 50 | NONE => depth used qs | |
| 51 | | SOME (st, stq) => | |
| 52 | if satp st andalso not (member Thm.eq_thm used st) then | |
| 53 | SOME (st, Seq.make (fn() => depth (st :: used) (stq :: qs))) | |
| 54 | else depth used (tac st :: stq :: qs)); | |
| 55 | in traced_tac (fn st => depth [] [Seq.single st]) end; | |
| 1588 | 56 | |
| 57 | ||
| 58 | (*Predicate: Does the rule have fewer than n premises?*) | |
| 59582 | 59 | fun has_fewer_prems n rule = Thm.nprems_of rule < n; | 
| 1588 | 60 | |
| 61 | (*Apply a tactic if subgoals remain, else do nothing.*) | |
| 62 | val IF_UNSOLVED = COND (has_fewer_prems 1) all_tac; | |
| 63 | ||
| 5754 | 64 | (*Force a tactic to solve its goal completely, otherwise fail *) | 
| 65 | fun SOLVE tac = tac THEN COND (has_fewer_prems 1) all_tac no_tac; | |
| 66 | ||
| 2672 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 67 | (*Execute tac1, but only execute tac2 if there are at least as many subgoals | 
| 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 68 | as before. This ensures that tac2 is only applied to an outcome of tac1.*) | 
| 23178 | 69 | fun (tac1 THEN_MAYBE tac2) st = | 
| 59582 | 70 | (tac1 THEN COND (has_fewer_prems (Thm.nprems_of st)) all_tac tac2) st; | 
| 2672 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 71 | |
| 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 72 | fun (tac1 THEN_MAYBE' tac2) x = tac1 x THEN_MAYBE tac2 x; | 
| 
85d7e800d754
Replaced "flat" by the Basis Library function List.concat
 paulson parents: 
2143diff
changeset | 73 | |
| 1588 | 74 | (*Tactical to reduce the number of premises by 1. | 
| 75 | If no subgoals then it must fail! *) | |
| 3538 | 76 | fun DEPTH_SOLVE_1 tac st = st |> | 
| 59582 | 77 | (case Thm.nprems_of st of | 
| 78 | 0 => no_tac | |
| 79 | | n => DEPTH_FIRST (has_fewer_prems n) tac); | |
| 1588 | 80 | |
| 81 | (*Uses depth-first search to solve ALL subgoals*) | |
| 82 | val DEPTH_SOLVE = DEPTH_FIRST (has_fewer_prems 1); | |
| 83 | ||
| 84 | ||
| 85 | ||
| 2869 
acee08856cc9
DEEPEN now takes an upper bound for terminating searches
 paulson parents: 
2672diff
changeset | 86 | (**** Iterative deepening with pruning ****) | 
| 1588 | 87 | |
| 88 | fun has_vars (Var _) = true | |
| 60940 | 89 | | has_vars (Abs (_, _, t)) = has_vars t | 
| 90 | | has_vars (f $ t) = has_vars f orelse has_vars t | |
| 1588 | 91 | | has_vars _ = false; | 
| 92 | ||
| 93 | (*Counting of primitive inferences is APPROXIMATE, as the step tactic | |
| 94 | may perform >1 inference*) | |
| 95 | ||
| 96 | (*Pruning of rigid ancestor to prevent backtracking*) | |
| 23178 | 97 | fun prune (new as (k', np':int, rgd', stq), qs) = | 
| 60940 | 98 | let | 
| 99 | fun prune_aux (qs, []) = new :: qs | |
| 100 | | prune_aux (qs, (k, np, rgd, q) :: rqs) = | |
| 101 | if np' + 1 = np andalso rgd then | |
| 102 | (if !trace_DEPTH_FIRST then | |
| 103 |                  tracing ("Pruning " ^
 | |
| 104 | string_of_int (1+length rqs) ^ " levels") | |
| 105 | else (); | |
| 106 | (*Use OLD k: zero-cost solution; see Stickel, p 365*) | |
| 107 | (k, np', rgd', stq) :: qs) | |
| 108 | else prune_aux ((k, np, rgd, q) :: qs, rqs) | |
| 109 | fun take ([], rqs) = ([], rqs) | |
| 110 | | take (arg as ((k, np, rgd, stq) :: qs, rqs)) = | |
| 111 | if np' < np then take (qs, (k, np, rgd, stq) :: rqs) else arg; | |
| 112 | in prune_aux (take (qs, [])) end; | |
| 1588 | 113 | |
| 114 | ||
| 115 | (*Depth-first iterative deepening search for a state that satisfies satp | |
| 116 | tactic tac0 sets up the initial goal queue, while tac1 searches it. | |
| 117 | The solution sequence is redundant: the cutoff heuristic makes it impossible | |
| 118 | to suppress solutions arising from earlier searches, as the accumulated cost | |
| 119 | (k) can be wrong.*) | |
| 38802 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 120 | fun THEN_ITER_DEEPEN lim tac0 satp tac1 = traced_tac (fn st => | 
| 60940 | 121 | let | 
| 122 | val countr = Unsynchronized.ref 0 | |
| 123 | and tf = tracify trace_DEPTH_FIRST (tac1 1) | |
| 124 | and qs0 = tac0 st | |
| 1588 | 125 | (*bnd = depth bound; inc = estimate of increment required next*) | 
| 60940 | 126 | fun depth (bnd, inc) [] = | 
| 38802 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 127 | if bnd > lim then | 
| 60940 | 128 | (if !trace_DEPTH_FIRST then | 
| 129 | tracing (string_of_int (! countr) ^ | |
| 130 | " inferences so far. Giving up at " ^ string_of_int bnd) | |
| 131 | else (); | |
| 132 | NONE) | |
| 22025 
7c5896919eb8
Introduction of a bound on DEPTH_ITER, to prevent the meson method from running forever
 paulson parents: 
20852diff
changeset | 133 | else | 
| 60940 | 134 | (if !trace_DEPTH_FIRST then | 
| 135 | tracing (string_of_int (!countr) ^ | |
| 136 | " inferences so far. Searching to depth " ^ string_of_int bnd) | |
| 137 | else (); | |
| 138 | (*larger increments make it run slower for the hard problems*) | |
| 139 | depth (bnd + inc, 10)) [(0, 1, false, qs0)] | |
| 140 | | depth (bnd, inc) ((k, np, rgd, q) :: qs) = | |
| 141 | if k >= bnd then depth (bnd, inc) qs | |
| 1588 | 142 | else | 
| 60940 | 143 | (case | 
| 144 | (Unsynchronized.inc countr; | |
| 145 | if !trace_DEPTH_FIRST then | |
| 146 | tracing (string_of_int np ^ implode (map (fn _ => "*") qs)) | |
| 147 | else (); | |
| 148 | Seq.pull q) of | |
| 149 | NONE => depth (bnd, inc) qs | |
| 150 | | SOME (st, stq) => | |
| 151 | if satp st then (*solution!*) | |
| 152 | SOME(st, Seq.make (fn() => depth (bnd, inc) ((k, np, rgd, stq) :: qs))) | |
| 153 | else | |
| 154 | let | |
| 155 | val np' = Thm.nprems_of st; | |
| 156 | (*rgd' calculation assumes tactic operates on subgoal 1*) | |
| 157 | val rgd' = not (has_vars (hd (Thm.prems_of st))); | |
| 158 | val k' = k + np' - np + 1; (*difference in # of subgoals, +1*) | |
| 159 | in | |
| 160 | if k' + np' >= bnd then depth (bnd, Int.min (inc, k' + np' + 1 - bnd)) qs | |
| 161 | else if np' < np (*solved a subgoal; prune rigid ancestors*) | |
| 162 | then depth (bnd, inc) (prune ((k', np', rgd', tf st), (k, np, rgd, stq) :: qs)) | |
| 163 | else depth (bnd, inc) ((k', np', rgd', tf st) :: (k, np, rgd, stq) :: qs) | |
| 164 | end) | |
| 165 | in depth (0, 5) [] end); | |
| 1588 | 166 | |
| 38802 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 167 | fun ITER_DEEPEN lim = THEN_ITER_DEEPEN lim all_tac; | 
| 1588 | 168 | |
| 169 | ||
| 2869 
acee08856cc9
DEEPEN now takes an upper bound for terminating searches
 paulson parents: 
2672diff
changeset | 170 | (*Simple iterative deepening tactical. It merely "deepens" any search tactic | 
| 
acee08856cc9
DEEPEN now takes an upper bound for terminating searches
 paulson parents: 
2672diff
changeset | 171 | using increment "inc" up to limit "lim". *) | 
| 33335 
1e189f96c393
DEEPEN: less aggressive tracing, subject to trace_DEEPEN;
 wenzelm parents: 
32940diff
changeset | 172 | val trace_DEEPEN = Unsynchronized.ref false; | 
| 
1e189f96c393
DEEPEN: less aggressive tracing, subject to trace_DEEPEN;
 wenzelm parents: 
32940diff
changeset | 173 | |
| 38802 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 174 | fun DEEPEN (inc, lim) tacf m i = | 
| 33380 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 175 | let | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 176 | fun dpn m st = | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 177 | st |> | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 178 | (if has_fewer_prems i st then no_tac | 
| 38802 
a925c0ee42f7
clarified iter_deepen_limit vs meson (cf. 7c5896919eb8) -- eliminated global ref;
 wenzelm parents: 
35408diff
changeset | 179 | else if m > lim then | 
| 33380 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 180 | (if !trace_DEEPEN then tracing "Search depth limit exceeded: giving up" else (); | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 181 | no_tac) | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 182 | else | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 183 |           (if !trace_DEEPEN then tracing ("Search depth = " ^ string_of_int m) else ();
 | 
| 
cd6023a9a922
DEEPEN: all tracing is subject to trace_DEEPEN (NB: Proof General tends to "popup" tracing output);
 wenzelm parents: 
33335diff
changeset | 184 | tacf m i ORELSE dpn (m+inc))) | 
| 2869 
acee08856cc9
DEEPEN now takes an upper bound for terminating searches
 paulson parents: 
2672diff
changeset | 185 | in dpn m end; | 
| 23178 | 186 | |
| 32940 | 187 | |
| 1588 | 188 | (*** Best-first search ***) | 
| 189 | ||
| 32940 | 190 | (*total ordering on theorems, allowing duplicates to be found*) | 
| 191 | structure Thm_Heap = Heap | |
| 192 | ( | |
| 193 | type elem = int * thm; | |
| 59058 
a78612c67ec0
renamed "pairself" to "apply2", in accordance to @{apply 2};
 wenzelm parents: 
55171diff
changeset | 194 | val ord = prod_ord int_ord (Term_Ord.term_ord o apply2 Thm.prop_of); | 
| 32940 | 195 | ); | 
| 196 | ||
| 32738 | 197 | val trace_BEST_FIRST = Unsynchronized.ref false; | 
| 1588 | 198 | |
| 199 | (*For creating output sequence*) | |
| 60940 | 200 | fun some_of_list [] = NONE | 
| 201 | | some_of_list (x :: l) = SOME (x, Seq.make (fn () => some_of_list l)); | |
| 1588 | 202 | |
| 9094 | 203 | (*Check for and delete duplicate proof states*) | 
| 32940 | 204 | fun delete_all_min prf heap = | 
| 205 | if Thm_Heap.is_empty heap then heap | |
| 206 | else if Thm.eq_thm (prf, #2 (Thm_Heap.min heap)) | |
| 207 | then delete_all_min prf (Thm_Heap.delete_min heap) | |
| 208 | else heap; | |
| 1588 | 209 | |
| 210 | (*Best-first search for a state that satisfies satp (incl initial state) | |
| 211 | Function sizef estimates size of problem remaining (smaller means better). | |
| 212 | tactic tac0 sets up the initial priority queue, while tac1 searches it. *) | |
| 23178 | 213 | fun THEN_BEST_FIRST tac0 (satp, sizef) tac1 = | 
| 60940 | 214 | let | 
| 215 | val tac = tracify trace_BEST_FIRST tac1; | |
| 216 | fun pairsize th = (sizef th, th); | |
| 217 | fun bfs (news, nprf_heap) = | |
| 218 | (case List.partition satp news of | |
| 219 | ([], nonsats) => next (fold_rev Thm_Heap.insert (map pairsize nonsats) nprf_heap) | |
| 220 | | (sats, _) => some_of_list sats) | |
| 221 | and next nprf_heap = | |
| 222 | if Thm_Heap.is_empty nprf_heap then NONE | |
| 223 | else | |
| 224 | let | |
| 225 | val (n, prf) = Thm_Heap.min nprf_heap; | |
| 226 | val _ = | |
| 227 | if !trace_BEST_FIRST | |
| 228 |             then tracing("state size = " ^ string_of_int n)
 | |
| 229 | else (); | |
| 230 | in | |
| 231 | bfs (Seq.list_of (tac prf), delete_all_min prf (Thm_Heap.delete_min nprf_heap)) | |
| 232 | end; | |
| 233 | fun btac st = bfs (Seq.list_of (tac0 st), Thm_Heap.empty) | |
| 1588 | 234 | in traced_tac btac end; | 
| 235 | ||
| 236 | (*Ordinary best-first search, with no initial tactic*) | |
| 237 | val BEST_FIRST = THEN_BEST_FIRST all_tac; | |
| 238 | ||
| 23178 | 239 | (*Breadth-first search to satisfy satpred (including initial state) | 
| 1588 | 240 | SLOW -- SHOULD NOT USE APPEND!*) | 
| 60940 | 241 | fun gen_BREADTH_FIRST message satpred (tac: tactic) = | 
| 242 | let | |
| 243 | val tacf = Seq.list_of o tac; | |
| 244 | fun bfs prfs = | |
| 245 | (case List.partition satpred prfs of | |
| 246 | ([], []) => [] | |
| 247 | | ([], nonsats) => | |
| 248 |           (message ("breadth=" ^ string_of_int (length nonsats));
 | |
| 249 | bfs (maps tacf nonsats)) | |
| 250 | | (sats, _) => sats); | |
| 251 | in fn st => Seq.of_list (bfs [st]) end; | |
| 1588 | 252 | |
| 12262 | 253 | val BREADTH_FIRST = gen_BREADTH_FIRST tracing; | 
| 5693 | 254 | val QUIET_BREADTH_FIRST = gen_BREADTH_FIRST (K ()); | 
| 255 | ||
| 1588 | 256 | |
| 60940 | 257 | (* | 
| 258 | Implementation of A*-like proof procedure by modification | |
| 259 | of the existing code for BEST_FIRST and best_tac so that the | |
| 260 | current level of search is taken into account. | |
| 23178 | 261 | *) | 
| 1588 | 262 | |
| 60940 | 263 | (*Insertion into priority queue of states, marked with level*) | 
| 264 | fun insert_with_level (lnth: int * int * thm) [] = [lnth] | |
| 265 | | insert_with_level (l, m, th) ((l', n, th') :: nths) = | |
| 266 | if n < m then (l', n, th') :: insert_with_level (l, m, th) nths | |
| 267 | else if n = m andalso Thm.eq_thm (th, th') then (l', n, th') :: nths | |
| 268 | else (l, m, th) :: (l', n, th') :: nths; | |
| 1588 | 269 | |
| 270 | (*For creating output sequence*) | |
| 60940 | 271 | fun some_of_list [] = NONE | 
| 272 | | some_of_list (x :: xs) = SOME (x, Seq.make (fn () => some_of_list xs)); | |
| 1588 | 273 | |
| 32738 | 274 | val trace_ASTAR = Unsynchronized.ref false; | 
| 1588 | 275 | |
| 23178 | 276 | fun THEN_ASTAR tac0 (satp, costf) tac1 = | 
| 60940 | 277 | let | 
| 278 | val tf = tracify trace_ASTAR tac1; | |
| 279 | fun bfs (news, nprfs, level) = | |
| 280 | let fun cost thm = (level, costf level thm, thm) in | |
| 281 | (case List.partition satp news of | |
| 282 | ([], nonsats) => next (fold_rev (insert_with_level o cost) nonsats nprfs) | |
| 283 | | (sats, _) => some_of_list sats) | |
| 284 | end | |
| 285 | and next [] = NONE | |
| 286 | | next ((level, n, prf) :: nprfs) = | |
| 287 | (if !trace_ASTAR then | |
| 288 |           tracing ("level = " ^ string_of_int level ^
 | |
| 289 | " cost = " ^ string_of_int n ^ | |
| 290 | " queue length =" ^ string_of_int (length nprfs)) | |
| 291 | else (); | |
| 292 | bfs (Seq.list_of (tf prf), nprfs, level + 1)) | |
| 293 | fun tf st = bfs (Seq.list_of (tac0 st), [], 0); | |
| 1588 | 294 | in traced_tac tf end; | 
| 295 | ||
| 296 | (*Ordinary ASTAR, with no initial tactic*) | |
| 297 | val ASTAR = THEN_ASTAR all_tac; | |
| 298 | ||
| 299 | end; | |
| 300 | ||
| 301 | open Search; |