author | blanchet |
Wed, 01 Sep 2010 18:47:07 +0200 | |
changeset 39000 | d73a054e018c |
parent 38998 | f11a861e0061 |
child 39003 | c2aebd79981f |
permissions | -rw-r--r-- |
38021
e024504943d1
rename "ATP_Manager" ML module to "Sledgehammer";
blanchet
parents:
38020
diff
changeset
|
1 |
(* Title: HOL/Tools/Sledgehammer/sledgehammer.ML |
28477
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
2 |
Author: Fabian Immler, TU Muenchen |
32996
d2e48879e65a
removed disjunctive group cancellation -- provers run independently;
wenzelm
parents:
32995
diff
changeset
|
3 |
Author: Makarius |
35969 | 4 |
Author: Jasmin Blanchette, TU Muenchen |
28477
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
5 |
|
38021
e024504943d1
rename "ATP_Manager" ML module to "Sledgehammer";
blanchet
parents:
38020
diff
changeset
|
6 |
Sledgehammer's heart. |
28477
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
7 |
*) |
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
8 |
|
38021
e024504943d1
rename "ATP_Manager" ML module to "Sledgehammer";
blanchet
parents:
38020
diff
changeset
|
9 |
signature SLEDGEHAMMER = |
28477
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
10 |
sig |
38023 | 11 |
type failure = ATP_Systems.failure |
38988 | 12 |
type locality = Sledgehammer_Filter.locality |
13 |
type relevance_override = Sledgehammer_Filter.relevance_override |
|
14 |
type minimize_command = Sledgehammer_Reconstruct.minimize_command |
|
35969 | 15 |
type params = |
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
16 |
{blocking: bool, |
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
17 |
debug: bool, |
35969 | 18 |
verbose: bool, |
36143
6490319b1703
added "overlord" option (to get easy access to output files for debugging) + systematically use "raw_goal" rather than an inconsistent mixture
blanchet
parents:
36064
diff
changeset
|
19 |
overlord: bool, |
35969 | 20 |
atps: string list, |
21 |
full_types: bool, |
|
36235
61159615a0c5
added "explicit_apply" option to Sledgehammer, to control whether an explicit apply function should be used as much or as little as possible (replaces a previous global variable)
blanchet
parents:
36231
diff
changeset
|
22 |
explicit_apply: bool, |
38745
ad577fd62ee4
reorganize options regarding to the relevance threshold and decay
blanchet
parents:
38744
diff
changeset
|
23 |
relevance_thresholds: real * real, |
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
24 |
max_relevant: int option, |
35969 | 25 |
isar_proof: bool, |
36924 | 26 |
isar_shrink_factor: int, |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
27 |
timeout: Time.time, |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
28 |
expect: string} |
35867 | 29 |
type problem = |
38998 | 30 |
{ctxt: Proof.context, |
31 |
goal: thm, |
|
32 |
subgoal: int, |
|
33 |
axioms: ((string * locality) * thm) list} |
|
35867 | 34 |
type prover_result = |
36370
a4f601daa175
centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents:
36369
diff
changeset
|
35 |
{outcome: failure option, |
35969 | 36 |
message: string, |
37926
e6ff246c0cdb
renamings + only need second component of name pool to reconstruct proofs
blanchet
parents:
37627
diff
changeset
|
37 |
pool: string Symtab.table, |
38752
6628adcae4a7
consider "locality" when assigning weights to facts
blanchet
parents:
38748
diff
changeset
|
38 |
used_thm_names: (string * locality) list, |
35969 | 39 |
atp_run_time_in_msecs: int, |
36369
d2cd0d04b8e6
handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents:
36289
diff
changeset
|
40 |
output: string, |
35969 | 41 |
proof: string, |
38818
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
42 |
axiom_names: (string * locality) list vector, |
38083
c4b57f68ddb3
remove the "extra_clauses" business introduced in 19a5f1c8a844;
blanchet
parents:
38061
diff
changeset
|
43 |
conjecture_shape: int list list} |
38100
e458a0dd3dc1
use "explicit_apply" in the minimizer whenever it might make a difference to prevent freak failures;
blanchet
parents:
38098
diff
changeset
|
44 |
type prover = params -> minimize_command -> problem -> prover_result |
35867 | 45 |
|
38023 | 46 |
val dest_dir : string Config.T |
47 |
val problem_prefix : string Config.T |
|
48 |
val measure_runtime : bool Config.T |
|
35969 | 49 |
val kill_atps: unit -> unit |
50 |
val running_atps: unit -> unit |
|
29112
f2b45eea6dac
added 'atp_messages' command, which displays recent messages synchronously;
wenzelm
parents:
28835
diff
changeset
|
51 |
val messages: int option -> unit |
38023 | 52 |
val get_prover_fun : theory -> string -> prover |
38044 | 53 |
val run_sledgehammer : |
54 |
params -> int -> relevance_override -> (string -> minimize_command) |
|
55 |
-> Proof.state -> unit |
|
38023 | 56 |
val setup : theory -> theory |
28477
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
57 |
end; |
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
58 |
|
38021
e024504943d1
rename "ATP_Manager" ML module to "Sledgehammer";
blanchet
parents:
38020
diff
changeset
|
59 |
structure Sledgehammer : SLEDGEHAMMER = |
28477
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
60 |
struct |
9339d4dcec8b
version of sledgehammer using threads instead of processes, misc cleanup;
wenzelm
parents:
diff
changeset
|
61 |
|
38028 | 62 |
open ATP_Problem |
63 |
open ATP_Systems |
|
37578
9367cb36b1c4
renamed "Sledgehammer_FOL_Clauses" to "Metis_Clauses", so that Metis doesn't depend on Sledgehammer
blanchet
parents:
37577
diff
changeset
|
64 |
open Metis_Clauses |
38023 | 65 |
open Sledgehammer_Util |
38988 | 66 |
open Sledgehammer_Filter |
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
67 |
open Sledgehammer_Translate |
38988 | 68 |
open Sledgehammer_Reconstruct |
37583
9ce2451647d5
factored non-ATP specific code from "ATP_Manager" out, so that it can be reused for the LEO-II integration
blanchet
parents:
37581
diff
changeset
|
69 |
|
38023 | 70 |
|
37583
9ce2451647d5
factored non-ATP specific code from "ATP_Manager" out, so that it can be reused for the LEO-II integration
blanchet
parents:
37581
diff
changeset
|
71 |
(** The Sledgehammer **) |
9ce2451647d5
factored non-ATP specific code from "ATP_Manager" out, so that it can be reused for the LEO-II integration
blanchet
parents:
37581
diff
changeset
|
72 |
|
38102
019a49759829
fix bug in the newly introduced "bound concealing" code
blanchet
parents:
38100
diff
changeset
|
73 |
(* Identifier to distinguish Sledgehammer from other tools using |
019a49759829
fix bug in the newly introduced "bound concealing" code
blanchet
parents:
38100
diff
changeset
|
74 |
"Async_Manager". *) |
37585 | 75 |
val das_Tool = "Sledgehammer" |
76 |
||
77 |
fun kill_atps () = Async_Manager.kill_threads das_Tool "ATPs" |
|
78 |
fun running_atps () = Async_Manager.running_threads das_Tool "ATPs" |
|
79 |
val messages = Async_Manager.thread_messages das_Tool "ATP" |
|
35969 | 80 |
|
36281
dbbf4d5d584d
pass relevant options from "sledgehammer" to "sledgehammer minimize";
blanchet
parents:
36235
diff
changeset
|
81 |
(** problems, results, provers, etc. **) |
35969 | 82 |
|
83 |
type params = |
|
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
84 |
{blocking: bool, |
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
85 |
debug: bool, |
35969 | 86 |
verbose: bool, |
36143
6490319b1703
added "overlord" option (to get easy access to output files for debugging) + systematically use "raw_goal" rather than an inconsistent mixture
blanchet
parents:
36064
diff
changeset
|
87 |
overlord: bool, |
35969 | 88 |
atps: string list, |
89 |
full_types: bool, |
|
36235
61159615a0c5
added "explicit_apply" option to Sledgehammer, to control whether an explicit apply function should be used as much or as little as possible (replaces a previous global variable)
blanchet
parents:
36231
diff
changeset
|
90 |
explicit_apply: bool, |
38745
ad577fd62ee4
reorganize options regarding to the relevance threshold and decay
blanchet
parents:
38744
diff
changeset
|
91 |
relevance_thresholds: real * real, |
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
92 |
max_relevant: int option, |
35969 | 93 |
isar_proof: bool, |
36924 | 94 |
isar_shrink_factor: int, |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
95 |
timeout: Time.time, |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
96 |
expect: string} |
35867 | 97 |
|
98 |
type problem = |
|
38998 | 99 |
{ctxt: Proof.context, |
100 |
goal: thm, |
|
101 |
subgoal: int, |
|
102 |
axioms: ((string * locality) * thm) list} |
|
35867 | 103 |
|
104 |
type prover_result = |
|
36370
a4f601daa175
centralized ATP-specific error handling in "atp_wrapper.ML"
blanchet
parents:
36369
diff
changeset
|
105 |
{outcome: failure option, |
35969 | 106 |
message: string, |
37926
e6ff246c0cdb
renamings + only need second component of name pool to reconstruct proofs
blanchet
parents:
37627
diff
changeset
|
107 |
pool: string Symtab.table, |
38752
6628adcae4a7
consider "locality" when assigning weights to facts
blanchet
parents:
38748
diff
changeset
|
108 |
used_thm_names: (string * locality) list, |
35969 | 109 |
atp_run_time_in_msecs: int, |
36369
d2cd0d04b8e6
handle ATP proof delimiters in a cleaner, more extensible fashion
blanchet
parents:
36289
diff
changeset
|
110 |
output: string, |
35969 | 111 |
proof: string, |
38818
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
112 |
axiom_names: (string * locality) list vector, |
38083
c4b57f68ddb3
remove the "extra_clauses" business introduced in 19a5f1c8a844;
blanchet
parents:
38061
diff
changeset
|
113 |
conjecture_shape: int list list} |
35867 | 114 |
|
38100
e458a0dd3dc1
use "explicit_apply" in the minimizer whenever it might make a difference to prevent freak failures;
blanchet
parents:
38098
diff
changeset
|
115 |
type prover = params -> minimize_command -> problem -> prover_result |
35867 | 116 |
|
38023 | 117 |
(* configuration attributes *) |
118 |
||
38991 | 119 |
val (dest_dir, dest_dir_setup) = |
120 |
Attrib.config_string "sledgehammer_dest_dir" (K ""); |
|
121 |
(* Empty string means create files in Isabelle's temporary files directory. *) |
|
38023 | 122 |
|
123 |
val (problem_prefix, problem_prefix_setup) = |
|
38991 | 124 |
Attrib.config_string "sledgehammer_problem_prefix" (K "prob"); |
38023 | 125 |
|
126 |
val (measure_runtime, measure_runtime_setup) = |
|
38991 | 127 |
Attrib.config_bool "sledgehammer_measure_runtime" (K false); |
28484 | 128 |
|
38023 | 129 |
fun with_path cleanup after f path = |
130 |
Exn.capture f path |
|
131 |
|> tap (fn _ => cleanup path) |
|
132 |
|> Exn.release |
|
133 |
|> tap (after path) |
|
134 |
||
135 |
(* Splits by the first possible of a list of delimiters. *) |
|
136 |
fun extract_proof delims output = |
|
137 |
case pairself (find_first (fn s => String.isSubstring s output)) |
|
138 |
(ListPair.unzip delims) of |
|
139 |
(SOME begin_delim, SOME end_delim) => |
|
140 |
(output |> first_field begin_delim |> the |> snd |
|
141 |
|> first_field end_delim |> the |> fst |
|
142 |
|> first_field "\n" |> the |> snd |
|
143 |
handle Option.Option => "") |
|
144 |
| _ => "" |
|
28484 | 145 |
|
38023 | 146 |
fun extract_proof_and_outcome complete res_code proof_delims known_failures |
147 |
output = |
|
38061
685d1f0f75b3
handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents:
38044
diff
changeset
|
148 |
case known_failure_in_output output known_failures of |
685d1f0f75b3
handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents:
38044
diff
changeset
|
149 |
NONE => (case extract_proof proof_delims output of |
685d1f0f75b3
handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents:
38044
diff
changeset
|
150 |
"" => ("", SOME MalformedOutput) |
38023 | 151 |
| proof => if res_code = 0 then (proof, NONE) |
152 |
else ("", SOME UnknownError)) |
|
38061
685d1f0f75b3
handle Perl and "libwww-perl" failures more gracefully, giving the user some clues about what goes on
blanchet
parents:
38044
diff
changeset
|
153 |
| SOME failure => |
38023 | 154 |
("", SOME (if failure = IncompleteUnprovable andalso complete then |
155 |
Unprovable |
|
156 |
else |
|
157 |
failure)) |
|
28582 | 158 |
|
38023 | 159 |
fun extract_clause_sequence output = |
160 |
let |
|
161 |
val tokens_of = String.tokens (not o Char.isAlphaNum) |
|
162 |
fun extract_num ("clause" :: (ss as _ :: _)) = |
|
163 |
Int.fromString (List.last ss) |
|
164 |
| extract_num _ = NONE |
|
165 |
in output |> split_lines |> map_filter (extract_num o tokens_of) end |
|
166 |
||
167 |
val set_ClauseFormulaRelationN = "set_ClauseFormulaRelation" |
|
168 |
||
169 |
val parse_clause_formula_pair = |
|
38515 | 170 |
$$ "(" |-- scan_integer --| $$ "," |
171 |
-- (Symbol.scan_id ::: Scan.repeat ($$ "," |-- Symbol.scan_id)) --| $$ ")" |
|
38023 | 172 |
--| Scan.option ($$ ",") |
173 |
val parse_clause_formula_relation = |
|
174 |
Scan.this_string set_ClauseFormulaRelationN |-- $$ "(" |
|
175 |
|-- Scan.repeat parse_clause_formula_pair |
|
176 |
val extract_clause_formula_relation = |
|
38738
0ce517c1970f
make sure that "undo_ascii_of" is the inverse of "ascii_of", also for non-printable characters -- and avoid those in ``-style facts
blanchet
parents:
38698
diff
changeset
|
177 |
Substring.full #> Substring.position set_ClauseFormulaRelationN |
0ce517c1970f
make sure that "undo_ascii_of" is the inverse of "ascii_of", also for non-printable characters -- and avoid those in ``-style facts
blanchet
parents:
38698
diff
changeset
|
178 |
#> snd #> Substring.string #> strip_spaces_except_between_ident_chars |
0ce517c1970f
make sure that "undo_ascii_of" is the inverse of "ascii_of", also for non-printable characters -- and avoid those in ``-style facts
blanchet
parents:
38698
diff
changeset
|
179 |
#> explode #> parse_clause_formula_relation #> fst |
38023 | 180 |
|
38988 | 181 |
(* TODO: move to "Sledgehammer_Reconstruct" *) |
38023 | 182 |
fun repair_conjecture_shape_and_theorem_names output conjecture_shape |
38698
d19c3a7ce38b
clean handling of whether a fact is chained or not;
blanchet
parents:
38682
diff
changeset
|
183 |
axiom_names = |
38023 | 184 |
if String.isSubstring set_ClauseFormulaRelationN output then |
185 |
(* This is a hack required for keeping track of axioms after they have been |
|
38698
d19c3a7ce38b
clean handling of whether a fact is chained or not;
blanchet
parents:
38682
diff
changeset
|
186 |
clausified by SPASS's Flotter tool. The "ATP/scripts/spass" script is |
d19c3a7ce38b
clean handling of whether a fact is chained or not;
blanchet
parents:
38682
diff
changeset
|
187 |
also part of this hack. *) |
38023 | 188 |
let |
38040
174568533593
fix bug in the SPASS Flotter hack, when a conjecture FOF is translated to several CNF clauses
blanchet
parents:
38039
diff
changeset
|
189 |
val j0 = hd (hd conjecture_shape) |
38023 | 190 |
val seq = extract_clause_sequence output |
191 |
val name_map = extract_clause_formula_relation output |
|
192 |
fun renumber_conjecture j = |
|
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
193 |
conjecture_prefix ^ string_of_int (j - j0) |
38515 | 194 |
|> AList.find (fn (s, ss) => member (op =) ss s) name_map |
38040
174568533593
fix bug in the SPASS Flotter hack, when a conjecture FOF is translated to several CNF clauses
blanchet
parents:
38039
diff
changeset
|
195 |
|> map (fn s => find_index (curry (op =) s) seq + 1) |
38818
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
196 |
fun names_for_number j = |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
197 |
j |> AList.lookup (op =) name_map |> these |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
198 |
|> map_filter (try (unprefix axiom_prefix)) |> map unascii_of |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
199 |
|> map (fn name => |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
200 |
(name, name |> find_first_in_list_vector axiom_names |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
201 |
|> the) |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
202 |
handle Option.Option => |
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
203 |
error ("No such fact: " ^ quote name ^ ".")) |
38023 | 204 |
in |
38040
174568533593
fix bug in the SPASS Flotter hack, when a conjecture FOF is translated to several CNF clauses
blanchet
parents:
38039
diff
changeset
|
205 |
(conjecture_shape |> map (maps renumber_conjecture), |
38818
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
206 |
seq |> map names_for_number |> Vector.fromList) |
38023 | 207 |
end |
208 |
else |
|
38698
d19c3a7ce38b
clean handling of whether a fact is chained or not;
blanchet
parents:
38682
diff
changeset
|
209 |
(conjecture_shape, axiom_names) |
38023 | 210 |
|
211 |
||
212 |
(* generic TPTP-based provers *) |
|
213 |
||
38455 | 214 |
fun prover_fun atp_name |
38645 | 215 |
{exec, required_execs, arguments, has_incomplete_mode, proof_delims, |
38997 | 216 |
known_failures, default_max_relevant, explicit_forall, |
217 |
use_conjecture_for_hypotheses} |
|
38455 | 218 |
({debug, verbose, overlord, full_types, explicit_apply, |
38998 | 219 |
max_relevant, isar_proof, isar_shrink_factor, timeout, ...} : params) |
220 |
minimize_command ({ctxt, goal, subgoal, axioms} : problem) = |
|
38023 | 221 |
let |
38998 | 222 |
val (_, hyp_ts, concl_t) = strip_subgoal goal subgoal |
223 |
val max_relevant = the_default default_max_relevant max_relevant |
|
224 |
val axioms = take max_relevant axioms |
|
38023 | 225 |
(* path to unique problem file *) |
226 |
val the_dest_dir = if overlord then getenv "ISABELLE_HOME_USER" |
|
38998 | 227 |
else Config.get ctxt dest_dir |
228 |
val the_problem_prefix = Config.get ctxt problem_prefix |
|
38023 | 229 |
fun prob_pathname nr = |
230 |
let |
|
231 |
val probfile = |
|
38455 | 232 |
Path.basic ((if overlord then "prob_" ^ atp_name |
38023 | 233 |
else the_problem_prefix ^ serial_string ()) |
234 |
^ "_" ^ string_of_int nr) |
|
235 |
in |
|
236 |
if the_dest_dir = "" then File.tmp_path probfile |
|
237 |
else if File.exists (Path.explode the_dest_dir) |
|
238 |
then Path.append (Path.explode the_dest_dir) probfile |
|
38818
61cf050f8b2e
improve SPASS hack, when a clause comes from several facts
blanchet
parents:
38752
diff
changeset
|
239 |
else error ("No such directory: " ^ quote the_dest_dir ^ ".") |
38998 | 240 |
end |
38023 | 241 |
|
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
242 |
val measure_run_time = verbose orelse Config.get ctxt measure_runtime |
38092
81a003f7de0d
speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents:
38091
diff
changeset
|
243 |
val command = Path.explode (getenv (fst exec) ^ "/" ^ snd exec) |
38023 | 244 |
(* write out problem file and call prover *) |
38645 | 245 |
fun command_line complete timeout probfile = |
38023 | 246 |
let |
247 |
val core = File.shell_path command ^ " " ^ arguments complete timeout ^ |
|
248 |
" " ^ File.shell_path probfile |
|
249 |
in |
|
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
250 |
(if measure_run_time then "TIMEFORMAT='%3U'; { time " ^ core ^ " ; }" |
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
251 |
else "exec " ^ core) ^ " 2>&1" |
38023 | 252 |
end |
253 |
fun split_time s = |
|
254 |
let |
|
255 |
val split = String.tokens (fn c => str c = "\n"); |
|
256 |
val (output, t) = s |> split |> split_last |> apfst cat_lines; |
|
257 |
fun as_num f = f >> (fst o read_int); |
|
258 |
val num = as_num (Scan.many1 Symbol.is_ascii_digit); |
|
259 |
val digit = Scan.one Symbol.is_ascii_digit; |
|
260 |
val num3 = as_num (digit ::: digit ::: (digit >> single)); |
|
261 |
val time = num --| Scan.$$ "." -- num3 >> (fn (a, b) => a * 1000 + b); |
|
262 |
val as_time = the_default 0 o Scan.read Symbol.stopper time o explode; |
|
263 |
in (output, as_time t) end; |
|
264 |
fun run_on probfile = |
|
38092
81a003f7de0d
speed up the minimizer by using the time taken for the first iteration as a timeout for the following iterations, and fix a subtle bug in "string_for_failure"
blanchet
parents:
38091
diff
changeset
|
265 |
case filter (curry (op =) "" o getenv o fst) (exec :: required_execs) of |
38032 | 266 |
(home_var, _) :: _ => |
38023 | 267 |
error ("The environment variable " ^ quote home_var ^ " is not set.") |
38032 | 268 |
| [] => |
269 |
if File.exists command then |
|
270 |
let |
|
38645 | 271 |
fun do_run complete timeout = |
38032 | 272 |
let |
38645 | 273 |
val command = command_line complete timeout probfile |
38032 | 274 |
val ((output, msecs), res_code) = |
275 |
bash_output command |
|
276 |
|>> (if overlord then |
|
277 |
prefix ("% " ^ command ^ "\n% " ^ timestamp () ^ "\n") |
|
278 |
else |
|
279 |
I) |
|
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
280 |
|>> (if measure_run_time then split_time else rpair 0) |
38032 | 281 |
val (proof, outcome) = |
282 |
extract_proof_and_outcome complete res_code proof_delims |
|
283 |
known_failures output |
|
284 |
in (output, msecs, proof, outcome) end |
|
285 |
val readable_names = debug andalso overlord |
|
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
286 |
val (problem, pool, conjecture_offset, axiom_names) = |
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
287 |
prepare_problem ctxt readable_names explicit_forall full_types |
38998 | 288 |
explicit_apply hyp_ts concl_t axioms |
38631
979a0b37f981
prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents:
38600
diff
changeset
|
289 |
val ss = strings_for_tptp_problem use_conjecture_for_hypotheses |
979a0b37f981
prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents:
38600
diff
changeset
|
290 |
problem |
979a0b37f981
prefer TPTP "conjecture" tag to "hypothesis" on ATPs where this is possible;
blanchet
parents:
38600
diff
changeset
|
291 |
val _ = File.write_list probfile ss |
38032 | 292 |
val conjecture_shape = |
293 |
conjecture_offset + 1 upto conjecture_offset + length hyp_ts + 1 |
|
38040
174568533593
fix bug in the SPASS Flotter hack, when a conjecture FOF is translated to several CNF clauses
blanchet
parents:
38039
diff
changeset
|
294 |
|> map single |
38645 | 295 |
val timer = Timer.startRealTimer () |
38032 | 296 |
val result = |
38645 | 297 |
do_run false (if has_incomplete_mode then |
298 |
Time.fromMilliseconds |
|
299 |
(2 * Time.toMilliseconds timeout div 3) |
|
300 |
else |
|
301 |
timeout) |
|
302 |
|> has_incomplete_mode |
|
303 |
? (fn (_, msecs0, _, SOME _) => |
|
304 |
do_run true |
|
305 |
(Time.- (timeout, Timer.checkRealTimer timer)) |
|
306 |
|> (fn (output, msecs, proof, outcome) => |
|
307 |
(output, msecs0 + msecs, proof, outcome)) |
|
308 |
| result => result) |
|
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
309 |
in ((pool, conjecture_shape, axiom_names), result) end |
38032 | 310 |
else |
311 |
error ("Bad executable: " ^ Path.implode command ^ ".") |
|
38023 | 312 |
|
313 |
(* If the problem file has not been exported, remove it; otherwise, export |
|
314 |
the proof file too. *) |
|
315 |
fun cleanup probfile = |
|
316 |
if the_dest_dir = "" then try File.rm probfile else NONE |
|
317 |
fun export probfile (_, (output, _, _, _)) = |
|
318 |
if the_dest_dir = "" then |
|
319 |
() |
|
320 |
else |
|
321 |
File.write (Path.explode (Path.implode probfile ^ "_proof")) output |
|
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
322 |
val ((pool, conjecture_shape, axiom_names), |
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
323 |
(output, msecs, proof, outcome)) = |
38023 | 324 |
with_path cleanup export run_on (prob_pathname subgoal) |
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
325 |
val (conjecture_shape, axiom_names) = |
38023 | 326 |
repair_conjecture_shape_and_theorem_names output conjecture_shape |
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
327 |
axiom_names |
38023 | 328 |
val (message, used_thm_names) = |
329 |
case outcome of |
|
330 |
NONE => |
|
331 |
proof_text isar_proof |
|
332 |
(pool, debug, isar_shrink_factor, ctxt, conjecture_shape) |
|
38998 | 333 |
(full_types, minimize_command, proof, axiom_names, goal, subgoal) |
38744
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
334 |
|>> (fn message => |
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
335 |
message ^ (if verbose then |
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
336 |
"\nATP CPU time: " ^ string_of_int msecs ^ " ms." |
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
337 |
else |
2b6333f78a9e
make relevance filter work in term of a "max_relevant" option + use Vampire SOS;
blanchet
parents:
38741
diff
changeset
|
338 |
"")) |
38597
db482afec7f0
no spurious trailing "\n" at the end of Sledgehammer's output
blanchet
parents:
38590
diff
changeset
|
339 |
| SOME failure => (string_for_failure failure, []) |
38023 | 340 |
in |
341 |
{outcome = outcome, message = message, pool = pool, |
|
342 |
used_thm_names = used_thm_names, atp_run_time_in_msecs = msecs, |
|
38282
319c59682c51
move Sledgehammer's HOL -> FOL translation to separate file (sledgehammer_translate.ML)
blanchet
parents:
38277
diff
changeset
|
343 |
output = output, proof = proof, axiom_names = axiom_names, |
38083
c4b57f68ddb3
remove the "extra_clauses" business introduced in 19a5f1c8a844;
blanchet
parents:
38061
diff
changeset
|
344 |
conjecture_shape = conjecture_shape} |
38023 | 345 |
end |
346 |
||
347 |
fun get_prover_fun thy name = prover_fun name (get_prover thy name) |
|
348 |
||
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
349 |
fun start_prover_thread (params as {blocking, verbose, full_types, timeout, |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
350 |
expect, ...}) |
38998 | 351 |
i n relevance_override minimize_command axioms state |
352 |
(prover, atp_name) = |
|
36379
20ef039bccff
make "ATP_Manager.get_prover" a total function, since we always want to show the same error text
blanchet
parents:
36373
diff
changeset
|
353 |
let |
37584 | 354 |
val birth_time = Time.now () |
355 |
val death_time = Time.+ (birth_time, timeout) |
|
38998 | 356 |
val {context = ctxt, facts, goal} = Proof.goal state |
36379
20ef039bccff
make "ATP_Manager.get_prover" a total function, since we always want to show the same error text
blanchet
parents:
36373
diff
changeset
|
357 |
val desc = |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
358 |
"ATP " ^ quote atp_name ^ " for subgoal " ^ string_of_int i ^ ":" ^ |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
359 |
(if blocking then |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
360 |
"" |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
361 |
else |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
362 |
"\n" ^ Syntax.string_of_term ctxt (Thm.term_of (Thm.cprem_of goal i))) |
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
363 |
fun run () = |
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
364 |
let |
38998 | 365 |
val problem = {ctxt = ctxt, goal = goal, subgoal = i, axioms = axioms} |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
366 |
val (outcome_code, message) = |
38998 | 367 |
prover_fun atp_name prover params (minimize_command atp_name) problem |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
368 |
|> (fn {outcome, message, ...} => |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
369 |
(if is_some outcome then "none" else "some", message)) |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
370 |
handle ERROR message => ("unknown", "Error: " ^ message ^ "\n") |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
371 |
| exn => ("unknown", "Internal error:\n" ^ |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
372 |
ML_Compiler.exn_message exn ^ "\n") |
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
373 |
in |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
374 |
if expect = "" orelse outcome_code = expect then |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
375 |
() |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
376 |
else if blocking then |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
377 |
error ("Unexpected outcome: " ^ quote outcome_code ^ ".") |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
378 |
else |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
379 |
warning ("Unexpected outcome: " ^ quote outcome_code ^ "."); |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
380 |
message |
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
381 |
end |
37584 | 382 |
in |
38985
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
383 |
if blocking then priority (desc ^ "\n" ^ TimeLimit.timeLimit timeout run ()) |
162bbbea4e4d
added "expect" feature of Nitpick to Sledgehammer, for regression testing
blanchet
parents:
38982
diff
changeset
|
384 |
else Async_Manager.launch das_Tool verbose birth_time death_time desc run |
37584 | 385 |
end |
28582 | 386 |
|
38044 | 387 |
fun run_sledgehammer {atps = [], ...} _ _ _ _ = error "No ATP is set." |
38998 | 388 |
| run_sledgehammer (params as {blocking, verbose, atps, full_types, |
389 |
relevance_thresholds, max_relevant, ...}) |
|
390 |
i relevance_override minimize_command state = |
|
38044 | 391 |
case subgoal_count state of |
392 |
0 => priority "No subgoal!" |
|
393 |
| n => |
|
394 |
let |
|
38998 | 395 |
val {context = ctxt, facts = chained_ths, goal} = Proof.goal state |
396 |
val thy = Proof.theory_of state |
|
39000 | 397 |
val _ = () |> not blocking ? kill_atps |
38044 | 398 |
val _ = priority "Sledgehammering..." |
38998 | 399 |
val (_, hyp_ts, concl_t) = strip_subgoal goal i |
400 |
val provers = map (`(get_prover thy)) atps |
|
401 |
val max_max_relevant = |
|
402 |
case max_relevant of |
|
403 |
SOME n => n |
|
404 |
| NONE => fold (Integer.max o #default_max_relevant o fst) provers 0 |
|
405 |
val axioms = |
|
406 |
relevant_facts ctxt full_types relevance_thresholds max_max_relevant |
|
407 |
relevance_override chained_ths hyp_ts concl_t |
|
408 |
val num_axioms = length axioms |
|
409 |
val _ = if verbose then |
|
410 |
priority ("Selected " ^ string_of_int num_axioms ^ " fact" ^ |
|
411 |
plural_s num_axioms ^ ".") |
|
412 |
else |
|
413 |
() |
|
38982
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
414 |
val _ = |
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
415 |
(if blocking then Par_List.map else map) |
820b8221ed48
added "blocking" option to Sledgehammer to run in synchronous mode;
blanchet
parents:
38893
diff
changeset
|
416 |
(start_prover_thread params i n relevance_override |
38998 | 417 |
minimize_command axioms state) provers |
38044 | 418 |
in () end |
419 |
||
38023 | 420 |
val setup = |
421 |
dest_dir_setup |
|
422 |
#> problem_prefix_setup |
|
423 |
#> measure_runtime_setup |
|
424 |
||
28582 | 425 |
end; |