1 (* Title: HOL/Tools/ATP_Manager/atp_wrapper.ML
2 Author: Fabian Immler, TU Muenchen
4 Wrapper functions for external ATPs.
7 signature ATP_WRAPPER =
9 type prover = ATP_Manager.prover
11 (* hooks for problem files *)
12 val destdir : string Config.T
13 val problem_prefix : string Config.T
14 val measure_runtime : bool Config.T
16 val refresh_systems_on_tptp : unit -> unit
17 val setup : theory -> theory
20 structure ATP_Wrapper : ATP_WRAPPER =
23 open Sledgehammer_HOL_Clause
24 open Sledgehammer_Fact_Filter
25 open Sledgehammer_Proof_Reconstruct
28 (** generic ATP wrapper **)
30 (* external problem files *)
32 val (destdir, destdir_setup) = Attrib.config_string "atp_destdir" "";
33 (*Empty string means create files in Isabelle's temporary files directory.*)
35 val (problem_prefix, problem_prefix_setup) =
36 Attrib.config_string "atp_problem_prefix" "prob";
38 val (measure_runtime, measure_runtime_setup) =
39 Attrib.config_bool "atp_measure_runtime" false;
42 (* prover configuration *)
46 arguments: int -> string,
47 failure_strs: string list,
49 insert_theory_const: bool,
50 emit_structured_proof: bool};
55 fun with_path cleanup after f path =
57 |> tap (fn _ => cleanup path)
61 fun find_failure strs proof =
62 case filter (fn s => String.isSubstring s proof) strs of
63 [] => if is_proof_well_formed proof then NONE
64 else SOME "Ill-formed ATP output"
65 | (failure :: _) => SOME failure
67 fun external_prover relevance_filter prepare write cmd args failure_strs
68 produce_answer name ({with_full_types, subgoal, goal, axiom_clauses,
69 filtered_clauses}: problem) =
71 (* get clauses and prepare them for writing *)
72 val (ctxt, (chain_ths, th)) = goal;
73 val thy = ProofContext.theory_of ctxt;
74 val chain_ths = map (Thm.put_name_hint chained_hint) chain_ths;
75 val goal_cls = #1 (Sledgehammer_Fact_Preprocessor.neg_conjecture_clauses ctxt th subgoal);
76 val the_filtered_clauses =
77 (case filtered_clauses of
78 NONE => relevance_filter goal goal_cls
80 val the_axiom_clauses =
81 (case axiom_clauses of
82 NONE => the_filtered_clauses
83 | SOME axcls => axcls);
84 val (thm_names, clauses) =
85 prepare goal_cls chain_ths the_axiom_clauses the_filtered_clauses thy;
87 (* path to unique problem file *)
88 val destdir' = Config.get ctxt destdir;
89 val problem_prefix' = Config.get ctxt problem_prefix;
90 fun prob_pathname nr =
92 Path.basic (problem_prefix' ^ serial_string () ^ "_" ^ string_of_int nr)
94 if destdir' = "" then File.tmp_path probfile
95 else if File.exists (Path.explode destdir')
96 then Path.append (Path.explode destdir') probfile
97 else error ("No such directory: " ^ destdir')
100 (* write out problem file and call prover *)
101 fun cmd_line probfile =
102 if Config.get ctxt measure_runtime
103 then (* Warning: suppresses error messages of ATPs *)
104 "TIMEFORMAT='%3U'; { time " ^ space_implode " " [File.shell_path cmd,
105 args, File.shell_path probfile] ^ " 2> /dev/null" ^ " ; } 2>&1"
107 space_implode " " ["exec", File.shell_path cmd, args,
108 File.shell_path probfile];
111 val split = String.tokens (fn c => str c = "\n");
112 val (proof, t) = s |> split |> split_last |> apfst cat_lines;
113 fun as_num f = f >> (fst o read_int);
114 val num = as_num (Scan.many1 Symbol.is_ascii_digit);
115 val digit = Scan.one Symbol.is_ascii_digit;
116 val num3 = as_num (digit ::: digit ::: (digit >> single));
117 val time = num --| Scan.$$ "." -- num3 >> (fn (a, b) => a * 1000 + b);
118 val as_time = the_default 0 o Scan.read Symbol.stopper time o explode;
119 in (proof, as_time t) end;
121 if Config.get ctxt measure_runtime then split_time s else (s, 0)
122 fun run_on probfile =
123 if File.exists cmd then
124 write with_full_types probfile clauses
125 |> pair (apfst split_time' (bash_output (cmd_line probfile)))
126 else error ("Bad executable: " ^ Path.implode cmd);
128 (* if problemfile has not been exported, delete problemfile; otherwise export proof, too *)
129 fun cleanup probfile = if destdir' = "" then try File.rm probfile else NONE;
130 fun export probfile (((proof, _), _), _) =
131 if destdir' = "" then ()
132 else File.write (Path.explode (Path.implode probfile ^ "_proof")) proof;
134 val (((proof, time), rc), conj_pos) =
135 with_path cleanup export run_on (prob_pathname subgoal);
137 (* check for success and print out some information on failure *)
138 val failure = find_failure failure_strs proof;
139 val success = rc = 0 andalso is_none failure;
140 val (message, real_thm_names) =
141 if is_some failure then ("External prover failed.", [])
142 else if rc <> 0 then ("External prover failed: " ^ proof, [])
143 else apfst (fn s => "Try this command: " ^ s)
144 (produce_answer name (proof, thm_names, conj_pos, ctxt, th, subgoal));
146 {success = success, message = message,
147 theorem_names = real_thm_names, runtime = time, proof = proof,
148 internal_thm_names = thm_names, filtered_clauses = the_filtered_clauses}
152 (* generic TPTP-based provers *)
154 fun generic_tptp_prover
155 (name, {command, arguments, failure_strs, max_new_clauses,
156 insert_theory_const, emit_structured_proof}) timeout =
157 external_prover (get_relevant_facts max_new_clauses insert_theory_const)
158 (prepare_clauses false) write_tptp_file command (arguments timeout)
160 (if emit_structured_proof then structured_isar_proof
161 else metis_lemma_list false) name;
163 fun tptp_prover (name, p) = (name, generic_tptp_prover (name, p));
166 (** common provers **)
170 (*NB: Vampire does not work without explicit timelimit*)
172 val vampire_failure_strs =
173 ["Satisfiability detected", "Refutation not found", "CANNOT PROVE"];
174 val vampire_max_new_clauses = 60;
175 val vampire_insert_theory_const = false;
177 fun vampire_prover_config full : prover_config =
178 {command = Path.explode "$VAMPIRE_HOME/vampire",
179 arguments = (fn timeout => "--output_syntax tptp --mode casc" ^
180 " -t " ^ string_of_int timeout),
181 failure_strs = vampire_failure_strs,
182 max_new_clauses = vampire_max_new_clauses,
183 insert_theory_const = vampire_insert_theory_const,
184 emit_structured_proof = full};
186 val vampire = tptp_prover ("vampire", vampire_prover_config false);
187 val vampire_full = tptp_prover ("vampire_full", vampire_prover_config true);
192 val eprover_failure_strs =
193 ["SZS status: Satisfiable", "SZS status Satisfiable",
194 "SZS status: ResourceOut", "SZS status ResourceOut",
195 "# Cannot determine problem status"];
196 val eprover_max_new_clauses = 100;
197 val eprover_insert_theory_const = false;
199 fun eprover_config full : prover_config =
200 {command = Path.explode "$E_HOME/eproof",
201 arguments = (fn timeout => "--tstp-in --tstp-out -l5 -xAutoDev -tAutoDev" ^
202 " --silent --cpu-limit=" ^ string_of_int timeout),
203 failure_strs = eprover_failure_strs,
204 max_new_clauses = eprover_max_new_clauses,
205 insert_theory_const = eprover_insert_theory_const,
206 emit_structured_proof = full};
208 val eprover = tptp_prover ("e", eprover_config false);
209 val eprover_full = tptp_prover ("e_full", eprover_config true);
214 val spass_failure_strs =
215 ["SPASS beiseite: Completion found.", "SPASS beiseite: Ran out of time.",
216 "SPASS beiseite: Maximal number of loops exceeded."];
217 val spass_max_new_clauses = 40;
218 val spass_insert_theory_const = true;
220 fun spass_config insert_theory_const: prover_config =
221 {command = Path.explode "$SPASS_HOME/SPASS",
222 arguments = (fn timeout => "-Auto -SOS=1 -PGiven=0 -PProblem=0 -Splits=0" ^
223 " -FullRed=0 -DocProof -TimeLimit=" ^ string_of_int timeout),
224 failure_strs = spass_failure_strs,
225 max_new_clauses = spass_max_new_clauses,
226 insert_theory_const = insert_theory_const,
227 emit_structured_proof = false};
229 fun generic_dfg_prover
230 (name, ({command, arguments, failure_strs, max_new_clauses,
231 insert_theory_const, ...} : prover_config)) timeout =
233 (get_relevant_facts max_new_clauses insert_theory_const)
234 (prepare_clauses true)
239 (metis_lemma_list true)
242 fun dfg_prover (name, p) = (name, generic_dfg_prover (name, p));
244 val spass = dfg_prover ("spass", spass_config spass_insert_theory_const);
245 val spass_no_tc = dfg_prover ("spass_no_tc", spass_config false);
248 (* remote prover invocation via SystemOnTPTP *)
250 val systems = Synchronized.var "atp_wrapper_systems" ([]: string list);
254 val (answer, rc) = bash_output ("\"$ISABELLE_ATP_MANAGER/SystemOnTPTP\" -w")
256 if rc <> 0 then error ("Failed to get available systems at SystemOnTPTP:\n" ^ answer)
257 else split_lines answer
260 fun refresh_systems_on_tptp () =
261 Synchronized.change systems (fn _ => get_systems ());
263 fun get_system prefix = Synchronized.change_result systems (fn systems =>
264 (if null systems then get_systems () else systems)
265 |> `(find_first (String.isPrefix prefix)));
267 fun the_system prefix =
268 (case get_system prefix of
269 NONE => error ("System " ^ quote prefix ^ " not available at SystemOnTPTP")
272 val remote_failure_strs = ["Remote-script could not extract proof"];
274 fun remote_prover_config prover_prefix args max_new insert_tc: prover_config =
275 {command = Path.explode "$ISABELLE_ATP_MANAGER/SystemOnTPTP",
276 arguments = (fn timeout =>
277 args ^ " -t " ^ string_of_int timeout ^ " -s " ^ the_system prover_prefix),
278 failure_strs = remote_failure_strs,
279 max_new_clauses = max_new,
280 insert_theory_const = insert_tc,
281 emit_structured_proof = false};
283 val remote_vampire = tptp_prover ("remote_vampire", remote_prover_config
284 "Vampire---9" "" vampire_max_new_clauses vampire_insert_theory_const);
286 val remote_eprover = tptp_prover ("remote_e", remote_prover_config
287 "EP---" "" eprover_max_new_clauses eprover_insert_theory_const);
289 val remote_spass = tptp_prover ("remote_spass", remote_prover_config
290 "SPASS---" "-x" spass_max_new_clauses spass_insert_theory_const);
293 [spass, vampire, eprover, vampire_full, eprover_full, spass_no_tc,
294 remote_vampire, remote_spass, remote_eprover]
295 val prover_setup = fold add_prover provers
299 #> problem_prefix_setup
300 #> measure_runtime_setup